vol1 no3

(IJCNS) International Journal of Computer and Network Security, Vol. 1, No. 3, December 2009

1

Suggestion of New Core Point of Attacks on IEEE 802.16e Networks: A Survey

Adnan Khan Akhunzada1, Saeed Murtaza2, Ahmad Raza Cheema3 and Arif Wahla4

School of Telecommunication National University of Science & Technology (NUST), PK

[email protected], [email protected], [email protected], 4 [email protected]

Abstract: This paper presents a survey of unaddressed security vulnerabilities found in IEEE 802.16e networks. Especially the vulnerabilities leading to denial of service (DoS) attack on IEEE 802.16e based network are discussed in detail. These vulnerabilities include unprotected network entry, unencrypted management communication, unprotected management frames, weak key sharing mechanism in Multi- and Broadcast operation. Moreover, the paper suggest a new core point of attacks on 802.16e networks i.e. the list of twenty unauthenticated management frames which are sent in clear. These unauthenticated management frames will be the cause of different kinds of serious threats in the coming near feature. A new practical scenario based attack regarding Reset command (RES_CMD) message leading to DoS attack has also been identified in this paper.

Keywords: WiMAX, IEEE 802.16e security, DoS Attacks, multi- and broadcast service, shared key vulnerability, hash chaining solution . 1. General Introduction. When bandwidth requirement is combined with ease and portability, the answer is Broadband wireless access (BWA).BWA has been developed to meet fast growing bandwidth requirements for WLANS and has some inherent security and design flaws that made it unsuitable for city wide deployment. In 1999 the working group of IEEE 802 was setup to develop a new standard of BWA for MAN namely IEEE 802.16 [1]. IEEE 802.16 was approved by the IEEE in 2001. It was revised several times and ended in the final standard IEEE 802.16-2004 which corresponds to revision D and is often called Fixed WiMAX [2]. It defines Wireless Metropolitan Broadband access for stationary and nomadic use. This means end devices can not move between base stations

(BS) but they can enter the network at different locations. An extended version IEEE 802.16e was developed to support mobility and is often called Mobile WiMAX [3].Mobile WiMAX introduces new features like different handover types, power saving methods and multi- and broadcast support and eliminates most of the security vulnerabilities exposed in its predecessors [3]. It uses EAP- based mutual authentication, a variety of strong encryption algorithms, nonce’s and packet numbers to defend against replay attacks and reduced key lifetimes. Initially some important parts of the functionality of Mobile WiMAX are introduced. Afterwards different security vulnerabilities are discussed & at the end the list of twenty unauthenticated management frames are shown which are susceptible to different kinds of threats.

1.1 Key management in 802.16e

The MS sets up a security association (SA) for each data communication it wants to establish in a 3-way TEK Exchange processed at initial network. Security association manages the keys for data encryption (the TEKs), their lifetimes and other security associated parameters of this connection. It also includes a TEK state machine which is used to periodically refresh keying material before the life span of a TEK expires. To request new keying material the state machine sends a key request to the BS which responds with a key response including a new TEK. This transferred TEK is encrypted by a key encryption key (KEK) which is derived from AK and is globally used to decrypt received keys of all SAs. To avoid communication interruption each SA simultaneously holds two TEKs. When one TEK expires the second one is used for traffic encryption and a new one is requested.

Figure 1: The standard request/reply mechanism compared with the MBRA

mailto:[email protected]





2

1.2 Multi- and Broadcast Service (MBS) IEEE 802.16e has introduced services for multicast and broadcast communication .This allows the BS to distribute data simultaneously to multiple MSs and uses a common group traffic encryption key (GTEK) for traffic en- /decryption to secure the broadcast communication. Every group member must know this key. Algorithms used for sharing the GTEK between MS and BS are shown in Figure 1. The mandatory key request / reply mechanism and the optional Multi- and Broadcast Rekeying Algorithm (MBRA). In the standard request/reply mechanism a MS has to manage the GTEK update by itself i.e. to request new keying martial before the old key expires while in Multi- and Broadcast Rekeying algorithm (MBRA), the keys are managed by the BS. The BS broadcasts one Key Update Command message to all MSs, if a key lifetime is going to expire. This saves a lot of bandwidth as GTEKs are updated very frequently. It is also distributed by a Key Update Command message, but in a unicast way encrypted by the MS-related KEK. If a MS has not received a new key after a specific time, it requests keying material according to the standard request/reply mechanism. This is also done if the authentication value of a Key Update Command message is not valid.

1.3 Existing Analysis for WiMAX Security

Fixed WiMAX security was analyzed in several papers [3]. With the publication of the Mobile WiMAX revision, most of these vulnerabilities were solved. The security of IEEE 802.16e has only been evaluated by a few papers [4]. [5] Examined the 3-way TEK exchange and the authorization process and could not find any security flaw. Also [6] analyzed the key management protocol using protocol analyzing software and did not notice any problem. The multi-and broadcast service was examined by [7] by applying a protocol analyzing tool. He found out that security of the MBS is based on a few parameters which need to be implemented properly for complete protection. It is also pointed out that the Interoperation with other protocols could be a security problem if these protocols have lower security characteristics. 2. Vulnerabilities in IEEE 802.16e. This section highlights vulnerabilities present in Mobile WiMAX by our analysis. These vulnerabilities are:

2.1. Lack of authenticated management frames

Mobile WiMAX includes some unauthenticated management frames. An SS has no way to know that the message forwarded is from genuine BS or from an adversary.

2.2. Absence of encryption in management communications

The complete management communication between mobile station and base station is unencrypted and sent in clear. A potential adversary can listens to the traffic with

freely available tools and can collect lots of information about both instances.

2.3. Shared keys in the multi- and broadcast Service For symmetric traffic encryption, the multi-and broadcast service in Mobile WiMAX shares keying material with all group members. This introduces the vulnerability that group members can forge messages or even distribute own traffic keying material, thus controlling the multi- and broadcast content.

2.4. Unprotected initial network entry The complete network entry procedure between BS and SS/MS is unprotected. An adversary can not listen to the traffic only but can also use the information to forge different kinds of management frames, e.g. an attacker can forge the (RNG-RSP) messages to manipulate different settings of SS/MS. 3. Possible Denial of Service Attacks on IEEE 802.16e Networks. (HMAC) [8] or Cipher Based Message Authentication Code (CMAC) [9] have been proposed to provide integrity to the majority of management massage. However, some messages are not provided by any authentication mechanism. This introduces some vulnerability. Authentication of broadcasted management messages is hard since there is no common key to generate message digests. Moreover, a common key would not entirely protect the integrity of the message as mobile stations sharing the key can forge these messages and generate valid authentication digits. Maximum DoS vulnerabilities stem from unprotected management frames

3.1. Unprotected management frames Following are some of the possible denial of service attacks based on management frames on IEEE 802.16e networks. We will first briefly explain the purpose of message than discuss possible DoS attacks based on the message. icity.

3.1.1. Power Control Change Request (PMC-REQ) Message

The power control mode of a MS can be changed by sending a Power Control Mode Change Request (PMC_REQ) to BS. The BS then respond with the power control mode change response (PMC_RSP) message. This message can also be sent by the BS in unwanted manner to change MSs Power control mode. It also includes the power adjustment value that should be arranged by the MS. The PMC_REQ message can be used by an opponent to request an undesired change of an MSs Power control mode. The message is accepted as if it came from the genuine MS. Vulnerability regarding forgery of the Power Control Mode Change Response (PMC_RSP) message sent from the BS can openly be used an adversary to change the power control mode of the MS and also adjust its transmission power with the intention to interrupt the communication.


3

3.1.2. Mobile Neighbor Advertisement (MOB_NBR-ADV) Message

Neighbor advertisement message (MOB_NBR-ADV) is also not authenticated used by the serving BS in order to announce the characteristics of neighbor BS to MSs seeking for handover possibilities. An opponent is able to keep back individual BSs by omitting information about their existence when he forges this message. This prevents MSs to handover to BSs which might have better characteristics as their serving BS and can also distribute wrong data about neighbor BSs or announce non existing BSs.

3.1.3. Fast Power Control (FPC) Message Unauthenticated broadcasted Fast Power Control (FPC) message is sent by the BS to one or multiple MS to adjust their transmitting power. It is possible to decrease the transmitting power of all reachable MSs to a minimum so that it is too low to be recognized by the BS. Another misuse of the message is to set the transmitting power of all MSs to the maximum with the objective to stress their batteries.

3.1.4. Mobile Traffic Indication (MOB_TRF-IND) Message

Traffic Indication message (MOB_TRF-IND) is one of the broadcasted and unauthenticated management messages. It is used by the BS to awake a sleeping MS that there is traffic destined to it. So the MS is waked up from sleep mode. An active adversary could use this message to frequently wake up MSs and stress their battery.

3.1.5. Multicast Assignment Request (MSC-REQ) Message

Multicast Assignment Request message (MSC-REQ) an unauthenticated unicast message used by BS to remove a MS from a multicast polling group and subsequently sends a response back to the BS. This exchange is done using the primary management connection between BS and MS. A polling group is a group of MS which can get bandwidth from the BS through a polling mechanism. An attacker can easily remove MSs from polling groups due to no authentication. If a MS is detached from a polling group, it has to use the compulsory contention based bandwidth allocation algorithm which results in a greater uplink delay.

3.1.6. Down Link Burst Profile Change Request (DBPC-REQ) Message

The Downlink Burst Profile Change Request message (DBPC-REQ), the BS sends this message to change the MSs burst profile to a more robust or a more effective one. This message does not have any authentication and integrity protection mechanism. Adversary can temporarily break the communication between MS and BS by changing MSs Burst profile (modulation encoding etc) so that it is not possible for the MS to demodulate the data received from the BS.

3.1.7. Mobile Association Reply (MOB_ASC-REP) Message

The association result report (MOB_ASC-REP) is another unauthenticated message with no integrity protection. An active adversary can change arbitrary response data in the message like time or power adjustments. Moreover the message includes the service prediction of the BS which advertises the services the BS can offer to the MS. Here an opponent can forge the message in a way that it looks like no services are being offered for the requesting MS.

3.1.8.Ranging Request (RNG-REQ)Message Ranging Request (RNG-REQ) message is an unauthenticated, unencrypted and stateless. Hence this message has great potential to be used as follows.

1. Attacker can launch a water torture Dos attack by changing power level to a maximum effectively reducing its battery life.

2. Attacker can change the SSs downlink channel to different frequency range and has multiple facets [9].

3. An adversary can shift only uplink channel to interrupt the communication between SS and BS.

4. An attacker can forge this message the power level of SS to minimum, so that it can barely transmit to BS triggering initial ranging procedure repeatedly [9].

3.1.9. Authorization request (AUTH_REQ) message

This message is well protected with nonce and digital signature of SS/MS. This message is not prone to Dos attack but still it is vulnerable to replay attack because BS has no way to know that whether it is fresh or it is replayed. Many researchers suggested using timestamps instead of nonces in order to eliminate this vulnerability [10]. It is also eliminated using sequence number by our suggestion.

3.1.10. Authorization invalid (Auth_invaled) message.

This message is itself not protected with HMAC/CMAC properly. Therefore it has a great potential to be used as a Dos tool to invalidate legitimate subscriber MSs [9]

3.1.11. Reset command (RES-CMD) Message

This message is used to orders an MS to reinitialize its MAC state machine. The aim of sending this message is to allow a BS to reset a non responsive or malfunctioning SS/MS. It is protected with HMAC which allows the recipient to check the authenticity of the message. But still problem may ensue from this message.Consisder the following step by step scenario.

a. Attacker synchronize with network to receive UL- MAP message

b. Bandwidth is allocated as CID so attacker chooses a victim CID and its burst profile from UL-MAP.

c. Next step is to transmit at scheduled time with scheduled modulation scheme.

d. Rather then rejected by noise filter, the signal will


4

be qualified as intelligible. e. Depending up on the signal strength of both the

SS’s, the signal will be either degraded or completely unintelligible.

f. If this continues, the BS will assume that the victim is malfunctioning and will issue RES- CMD.

After resetting, the attacker may start RNG-RSP attack as mentioned in earlier or may become a rouge BS of the victim by spoofing BSID.

3.1.12. SS Basic Capability Request (SBC-REQ)

message This message is used by the SS during basic connection. Since this message and the response of this message i.e. (SBC-RSP) SS Basic Capability Response sent by BS is both unauthenticated and not provided with any integrity protection mechanism. An active adversary can misuse these messages and can forge the information for a potential attack. There are other non-authenticated messages but a forgery of their carried information can be considered as less dangerous for the operability of the protocol.

3.2. Unencrypted management communication

In Mobile WiMAX management messages are still sent in the clear. The resulting risk shall be outlined in this section. When a MS performs initial network entry it negotiates communication parameters and settings with the BS. Here a lot of information is exchanged like security negotiation parameters, configuration settings, mobility parameters, power settings, vendor information, MSs Capabilities etc. Currently the complete management message exchange in the network entry process is unencrypted and the above mentioned information can be accessed just by listening on the channel. The only messages which are encrypted are key transfer messages. But in this case only the transferred key is encrypted, all other information is still sent in the clear. An adversary collecting management information can create detailed profiles about MSs including capabilities of devices, security settings, associations with base stations and all other information described above. Using the data offered in power reports, registration, ranging and handover messages, a listening adversary is able to determine the movement and approximate position of the MS as well. Monitoring the MAC address sent in ranging or registration messages reveals the mapping of CID and MAC address, making it possible to clearly relate the collected information to user equipment.

3.3. Shared keys in Multi- and Broadcast Service Mobile Wi-max present the service of PMP (point to multipoint) i.e. to distribute data to multiple MS with one single message through multi and broadcast services. IEEE 802.16e Broadcasted messages are encrypted symmetrically with a shared key. Every member in the group has the key and thus can decrypt the traffic because

message authentication is based on the same shared key. Every group member, besides decrypting and validating broadcast messages, can also encrypt and authenticate messages as if they originate from the genuine BS. The distribution of the traffic encryption keys (GTEKs) when the optional Multi-and Broadcast Rekeying Algorithm (MBRA) are used is more challenging feature. Due to broadcasting, the GKEK must also be a shared key and every group member knows it. Thus an active adversary group member can use it to generate valid encrypted and authenticated GTEK key update command messages and distribute its own GTEK. Every group member would set up the adversary’s key as a valid next GTEK. Consequently all traffic sent by the genuine BS can no longer be decrypted by the MS. From an MSs Viewpoint only traffic from the adversary is valid. 4. List of 20 Unauthenticated Management Frames: We will conclude our discussion by showing a list of 20 unauthenticated management frames which presents Type of the message, Message name, Message description, sent by, connection and Authentication. Sent by field shows that management frame is either sent by BS or MS. Authentication filed shows that either it is cryptographically authenticated or not. Since all these messages are totally unauthenticated so these management frames are susceptible to different kinds of serious threats. All though IEEE 802.16e networks have a very promising architecture but still there are some design flaws and inherent weaknesses. After deep study of 802.16e networks I found a list of some management frames which are totally insecure in term of their authentication, confidentiality, and integrity. An active adversary can generate serious kinds of malicious attacks using these management frames.

5. Conclusion Although IEEE 802.16e has a very robust and promising Security Architecture, there are still some flaws which need to be ironed out. In this paper, Vulnerabilities exposing to IEEE 802.16e networks to DoS attack are explored comprehensively. . At the end, a list of some of unauthenticated management frames are shown and must be properly secured by some cryptographic mechanism which provides confidentiality, authentication and integrity known as “CIA” trade in the field of security. Since a group of IEEE is working on the next draft of this protocol (802.16 e). There is a window of opportunity to improve the security measures of the IEEE 802.16 standard before WiMAX certified equipment has been built and sold by the millions. Changes need to be made before there are many “legacy” WiMAX branded systems in customer hands and while there is still time to ensure interoperability with the earliest equipment. Thus all these deficiencies must be kept in mind while designing the next draft of IEEE 802.16e protocol.


5

Type Message

Name

Message Description

Sent By

Connection Authentication

0 UCD Uplink Channel Descriptor BS Broadcast None

1

DCD Downlink Channel Descriptor

BS

Broadcast

None

2 DL-MAP Downlink Access Definition BS Broadcast None

3 UL-MAP Uplink Access Definition BS Broadcast None

4

RNG-REQ

Ranging Request

SS Initial Ranging or Basic

None

5

RNG-RSP

Ranging Response

BS Initial Ranging or Basic

None

23 DBPC- REQ

Downlink Burst Profile Change Request

SS

Basic

None

24

DBPC-RSP Downlink Burst Profile Change Response

BS

Basic

None

26

SBC-REQ

SS Basic Capability Request

SS

Basic

None

27

SBC-RSP SS Basic Capability Response

BS

Basic

None

32

TFTP-RSP Config File TFTP Complete Response

BS Primary Management

None

33 ARQ- Feedback

Standalone ARQ Feedback

BS or SS

Basic

None

34 ARQ- Discard

ARQ Discard message

BS or SS

Basic

None

35

ARQ-Reset

ARQ Reset message

BS or SS

Basic

None

36

REP-REQ Channel measurement Report Request

BS

Basic

None

37

REP-RSP Channel measurement Report Response

SS

Basic

None

39

MSH- NCFG

Mesh Network Configuration

BS or SS

Broadcast

Varies (Reject message is not authenticated)

41 MSH- DSCH

Mesh Distributed Schedule

SS

Broadcast

None

42 MSH- CSCH

Mesh Centralized Schedule

BS

Broadcast

None

43 MSH- CSCF

Mesh Centralized Schedule Configuration

BS

Broadcast

None

44

AAS- FBCK- REQ

AAS Feedback Response

BS

Basic

None (uses Request serial

numbers)

Figure 2: List of Unauthenticated Management Frames.


6

References [1]IEEE std. 802.16-2001,”Local and Metropolitan area

networks, part 16: Air interface for fixed broadband Wireless access system,”2001.

[2] IEEE Std. 802.16-2004, IEEE Standard for Local and Metropolitan Area Networks, part 16, Air Interface for Fixed Broadband Wireless Access Systems, IEEE Press, 2004.

[3] IEEE Std. 802.16e-2005, IEEE Standard for Local and Metropolitan Area Networks, part 16, Air Interface for Fixed and Mobile Broadband Wireless Access Systems, IEEE Press, 2006.

[4] Johnston D., Walker J.: Overview of IEEE 802.16 Security, IEEE Computer Society, 2004.

[5] Data A., He C., Mitchell J.C., Roy A., Sundararajan M.: 802.16e Notes, Electrical Engineering and Computer Science Departments, Stanford University, CA, USA, 2005, available at http://www.iab.org/ liaisons/ieee/EAP/802.16e Notes. PDF [6] Yuksel E.: Analysis of the PKMv2 Protocol in IEEE 802.16e- 2005 Using Static Analysis Informatics and Mathematical Modeling, Technical University Denmark, DTU, 2007, available at http://www.2imm. dtu.dk/pubdb/ views/publication_details.php?id=5159

[7] Ju-Yi Kuo: Analysis of 802.16e Multicast/Broadcast group privacy rekeying protocol, Stanford University, CA, USA, 2006, available at http://www.stanford. edu/class /cs259/projects/project01/01-Writeup.pdf

[8] Andreas Deininger, Shinsaku Kiyomoto, Jun Kurihara, Toshiaki Tanaka :Security Vulnerabilities and Solutions in Mobile WiMAX ,IJCSNS International Journal of Computer Science and Network Security, VOL.7 No.11, November 2007

[9] Dworkin M.: Recommendation for Block Cipher Modes of Operation: The CMAC mode for authentication, NIST special publication 800-38B, National Institute of Standards and Technology (NIST), MD, USA, 2005.

[10] Krawczyk H., Ballare M., Canetti R.: HMAC: Key- Hashing for Message Authentication, RFC 2104, http://www.ietf.org/rfc/rfc2104.txt, IETF, 1997.

[11] Huijie Li, Guangbin Fan, Jigang Qui and Xiaokang Lin:”GKDA: A Group Based Key Distribution Algorithm for WIMAX MBS Security”.

[12]Diffie, W. Hellman, M: New directions in cryptography.IEEE Transactions on information theory 22,644-654 (1976).

[13]Taeshik Shon, Wook Choi: An Analysis of Mobile WiMAX Security: Vulnerabilities and Solutions, First International Conference, NBiS 2007, LNCS, Vol. 4650, pp. 88-97, 2007

http://www.iab.org/

http://www.stanford

http://www.ietf.org/rfc/rfc2104.txt


7

Statistical Analysis of Defect Amplification Index in Domain Variant Commercial Software Application

Development through Fault Injection Patterns

Mr. P Mohammed Shareef1, Dr. M V Srinath2 and Dr.S.Balasubramanian3

1Trimentus Technologies, Vibras Castle, Flat G-2, # 10, 5th Street,

Chowdry Nagar, Valasaravakkam, Chennai - 600087, India [email protected]

2Mahendra Engineering College, Department Of Computer Science Engineering,

Mahendrapuri, Tiruchengode - 637 503, Tamil Nadu, India [email protected]

3Anna University-Coimbatore, # 117, West Ramalinga Road

RS Puram, Coimbatore – 641002, Tamil Nadu, India [email protected]

Abstract: Fault injection involves the deliberate insertion of faults or errors into software in order to determine its response and to study its behaviour. Fault Injection Experiments have proven to be an effective method for measuring and studying response of defects, validating fault-tolerant systems, and observing how systems behave in the presence of faults. This approach can offer both accuracy of fault injection results transparency of the system dynamics in the presence of faults. The objectives of this study are to measure and study defect leakage, analyse amplification of errors and study “Domino” effect of defects leaked. The approach for fault injection patterns presented in this research is validated by two approaches taken to arrive at the Amplification Index (AI) that represents the effect caused by defects in subsequent phases of software development in business applications. The approaches endeavour to demonstrate the phase wise impact of leaked defects, through statistical analysis of defects leakage and amplification patterns of systems, built using domain (education, e-governance, retail, systems) variants under same technology (C#.Net) , and also through a causal analysis done on the defects injected.

Keywords: Amplification Index (AI), Defect Leakage, Domino’s Effect, Fault Injection

1. Introduction Formulating reliable and fault tolerant software is difficult and requires discipline both in specifying system functionality and in implementing systems correctly. Approaches for developing highly reliable software include the use of formal methods [1], [2], [3], and rigorous testing methods [4]. Testing cannot guarantee that commercial and business software is correct, and verification requires enormous human effort and is subject to errors. Automated support is necessary to help ensure software correctness and fault tolerance. Fault injection modelling involves the deliberate insertion of faults or errors into a computer system in order to

determine its response. It has proven to be an effective method for measuring and studying response of defects, validating fault-tolerant systems, and observing how systems behave in the presence of faults. In this study, faults are injected in key phases of software development of business application following a typical water fall software life cycle viz., SRS, Design and Source code.

2. Literature Review The literature review consolidates the understanding on fault injection, associated topics and subsequent studies to emphasis the need to fault injections in business software application. It also crystallizes the need for awareness, tools and analyzes defect leakage/amplification.

Even after 20 years of existence the awareness of fault injection and associated modelling with tools are very rarely used and understood in the commercial software industry and used. The usefulness in the defect modelling and building fault tolerant software systems are not properly preached and/or practiced. Added, the availability of appropriate literature and software tools is very few and not used in commercial and business application design and testing.

After a detailed review by the researcher it was concluded that there is an industrious interest software fault injection in the software industry to develop commercially reliable software.

3. Approach In recent years there has been much interest in the field of software reliability and fault tolerance of systems and commercial software. This in turn has resulted in a wealth of literature being published around the topic, such as the Fault Injection in the form of the ‘Marrying Software Fault Injection Technology Results with Software Reliability’ by Jeffrey Voas, Cigital Norman Schneidewind.

Many critical business computer applications require “fault tolerance," the ability to recover from errors or exceptional conditions. Error free software is very difficult





8

to create and creating fault tolerant software is an even greater challenge. Fault tolerant software must successfully recover from a multitude of error conditions to prevent harmful system failures.

Software testing cannot demonstrate that a software system is completely correct. An enormous number of possible executions that must be examined in any real-world sized software system. Fault tolerance expands the number of states (and thus execution histories) that must be tested, because inconsistent or erroneous states must also be tested.

Mailing lists, websites, research and forums have been created in which all aspects of this fresh new niche software engineering area are discussed. People are interested, partly because it is a new area but also because the whole field of commercial software reliability is in itself so interesting; as it holds so many wide ranging disciplines, perspectives and logic at its core. Software reliability engineering is uniting professionals in disciplines that previously had little to do with one another, it is creating more opportunities for employment in the online environment, and it is changing the face and structure of all information that we seek out on the web. In the era of economic recession, customer demands reliable, certified and fault tolerant commercial and business software applications.

In this research, the focus is on software testing techniques that use fault injection. Several potentially powerful existing systems have drawbacks for practical application. We first examine existing fault injection techniques and evaluate their potential for practical application in commercial and business software applications. Available and accessible literature infrastructure including premium subscribed IEEE and ACM resources were studied and summarized for literature review from 1986 (20 years).

4. Fault Injection Modeling Fault Injection Modelling (FIM) involves the deliberate insertion of faults or errors into a computer system in order to determine its response. It has proven to be an effective method for measuring and studying response of defects, validating fault-tolerant systems, and observing how systems behave in the presence of faults. In this study, faults are injected in all phases of Software Development Life Cycle viz., Requirements, Design and Source Code.

4.1 Objectives The key objective of this research is to understand and statistically analyze the behaviour of faults and defects pattern by injecting known defects in business software applications artifacts (Software Requirements Specification, Design, Source Code) to study the “Amplification Index”, Domino Effect and defect leakage in key Software Development Life Cycle (SDLC) phases (Requirements, Design, Coding, Testing) of business application development, with domain variants (education, e-governance, retail, systems) build on same technology(C #.Net).

The goal of this research is to understand the behaviour of faults and defects pattern in commercial and business software application and defect leakage in each phase of application development.

Throughout the literature certain questions reoccur, which one would anticipate when a new field emerges in commercial software fault tolerance. People are interested, and want to understand and define commercial software reliability and fault tolerance, so the following questions which are recurrent throughout the literature are not surprising: Why study Fault Injection Modelling? Why study business software fault tolerance requirements? Why are they called ‘Fault Injection & Error Seeding’? Why review Software Implemented Fault Injection (SWIFI)? What work was performed, current status and work proposed?

These questions will be expanded upon throughout the research, and seek to bring clarity to those who want to find the answers to the above, or to see if there truly are any answers!

4.2 Background Concepts A fault is a hardware or software defect, inconsistency, transient electrical field, or other abnormal circumstance. An error is an invalid internal state, which may or may not be detected by the system.

A failure is an invalid output. Thus a fault or error becomes a failure when it propagates to the output. There is a natural progression from fault to error to failure. Recovery code is the part of a program that is designed to respond to error states. Recovery code executes after the program recognizes that some erroneous or abnormal state has been entered. This code should gracefully restore the system to a valid state before a failure occurs.

Figure 1 shows the progression from faults to errors and finally to failures. The recovery code should serve as a safety net to prevent the progression from error to failure. A fault tolerant system should never fail, even if it has faults.

Figure 1. Fault Tolerance Terms

Testing recovery code requires the modeling of bad states that accurately simulate exceptional situations. As much as 50% of a fault tolerant program can consist of recovery code. Although testing might include invalid data that executes some of the recovery code, often much of this code is never executed during normal testing.

Any recovery code testing technique must be based upon an assumed fault model [7]. We assume that all faults will behave according to some specific rules. Any fault model can only consider a subset of all possible faults.


9

For example, a common debugging practice is to insert a series of \print" statements in key positions. This debugging practice assumes a particular fault model.

Faults will cause the program to execute in the incorrect order and will be demonstrated Figure 2: Taxonomy of Fault Injection Techniques in the printed output. Clearly, not all faults will adhere to this model.

Figure 2. Taxonomy Of Fault Injection Techniques

No one fault model will fit all faults. However, a fault model can be very effective in detecting faults that fit the model.

Fault Injection technique of fault injection dates back to the 1970s when it was first used to induce faults at a hardware level. This type of fault injection is called Hardware Implemented Fault Injection (HWIFI) and attempts to simulate hardware failures within a system. The first experiments in hardware fault injection involved nothing more than shorting connections on circuit boards and observing the effect on the system (bridging faults). It was used primarily as a test of the dependability of the hardware system. Later specialised hardware was developed to extend this technique, such as devices to bombard specific areas of a circuit board with heavy radiation. It was soon found that faults could be induced by software techniques and that aspects of this technique could be useful for assessing software systems. Collectively these techniques are known as Software Implemented Fault Injection (SWIFI) [8].

Martin defines software fault injections as faults which are injected at the software level by corrupting code or data. So faults are applicable at the implementation phase when the code of the system is available, and it can be applied on an application to simulate either internal or external faults.

Internal faults represent design and implementation faults, such as variables/parameters that are wrong or not initialized, incorrect assignments or condition checks. External faults represent all external factors that are not related to faults in the code itself but that alter the system's state.

The injection of failures can discover errors that normal procedures cannot. First, it tests the mechanisms of exception and treatment of failures that in normal circumstances are not sufficiently proven and, helps to evaluate the risk, verifying how much defective can be the

system behaviour in presence of errors. All of the injection failures methods are based on concrete hardware or software characteristics associated to systems which are applied, then, to realize generalizations is a very complicated task.

4.3 Prior Work on Fault Injection Fault injection can be used to modify either a program's source code text or the machine state of an executing program. Figure 2 shows taxonomy of the key methods of fault injection. Fault injection techniques based on static analysis -program source modification - are modelled by the left sub tree.

The most common static fault injection is mutation testing. The right sub tree in Figure 2 models dynamic fault injection techniques where changes are made to an actively running program's state. Much of the recent fault injection research is concerned with dynamic injection.

4. 4 Domino’s effect Domino’s effect is the cascading effect of defects from the initial stages of the project to all the subsequent stages of the software life cycle. Errors undetected in one work product are ‘leaked’ to the child work product and amplifies defects in the child work product. This chain reaction causes an exponential defect leakage. E.g.: undetected errors in requirements leak and cause a significant number of defects in design which, in turn, causes more defects in the source code. The result of this study is to arrive at an “Amplification Index” which will characterize the extent of impact or damage of phase-wise defects in subsequent Software Development Life Cycle (SDLC) phases.

The defect components in a work product and leakage into subsequent phases is illustrated below:

Figure 3. Fault Injection Patten

5. Trimentus Approach For Fault Injection Experiments Defects were deliberately injected into each phase (work product) in the software development life cycle of a typical application development project and the effect of the defects injected was studied subsequently. The injected defects are typical defects that are characteristic of the software systems of a commercial application developed in the following domains;


10

Table 1: Application and Domain

An approach was adopted towards studying the impact of

defect amplification in a software system was causal analysis of the defects occurring in subsequent phases caused due to injected defects.

Fault injection can occur in several ways: Additional code can be linked to the target program and executed synchronously with the program flow.

A separate process can perform the injection asynchronously with the flow of the target process.

Separate hardware can directly access the memory to modify the state, thus not affecting the timing characteristics of the target process.

Overlay faults occur when a program writes into an incorrect location due to a faulty destination operand. Chillarege and Bowen claim that overlay faults account for 34% of the errors in systems programs. The experiment involved the use of failure acceleration, decreasing fault and error latency and increasing the probability that a fault will cause an error. The experiment applied failure acceleration by corrupting a large region of memory in a single injection. To inject an overlay fault, all bits in an entire page of physical memory are set to one. Because the page is in physical memory, the probability that the latency will be short is further increased. About 16% of the faults immediately crashed the system; about 14% caused a partial loss of service, which was usually recovered from soon after.

Half of the faults did not cause failures. These potential hazards are failures waiting to occur. The injection process used was manual and only 70 faults were injected during the entire experiment.

Software faults introduced include: Initialization faults: incorrectly or uninitialized variables. They are modelled by dynamically replacing the initializing assembly instructions with incorrect values or no-ops. Assignment faults: incorrect assignment statements. Variable names on the right hand side are changed by dynamically mutating the assembly code.

Condition check faults: missing condition checks, for example, failure to verify return values. Condition checks are either entirely overwritten with no-ops, or replaced an incorrect condition check.

Function faults: Invalid functions. The assembly code for a function is dynamically replaced with the assembly code from a manually rewritten alternate version.

Initialization faults can be caught statically with a good compiler. The assignment and condition check faults are clearly relevant to the testing of recovery code, since an incorrect assignment or condition can be a condition that should force the execution of recovery code. Function faults

are also relevant, especially if they could be automatically generated. Unfortunately, manual rewriting of sections of code is prohibitive in a large system.

6. WHY STUDY FAULT INJECTION MODELLING? Fault Injection Modelling has gradually crept into prominence over the last decade as one of the new buzz words in software design.

However, as Martin observes: ‘The main characteristic of fault injection software is that it is capable of injecting failures into any functional addressing unit by means of software, such as memory, registers, and peripherals. The goal of the fault injection software is to reproduce, in the logical scope, the errors that are reproduced after failures in the hardware. A good characterization of failure model should be allowed that this one was as versatile as possible, allowing a major number of combinations among the location, trigger conditions, kind of fault and duration, so that the coverage was maximum. Recent days, the Fault Injection technique has been considered as a very useful tool to monitor and evaluate the behaviour of computing systems in the presence of faults. It’s because the tool tries to produce or simulate faults during an execution of the system under test, and then the behaviour of the system is detected [9].

The Carnegie Mellon Software Engineering Institute1 reports that at least 42-50 percent of software defects originate in the requirements phase.

The Defence Acquisition University Program Manager Magazine2 reports that a Department of Defence study that over 50 percent of all software errors originate in the requirements phase.

Figure 4. Relative Cost to Fix Defects Vs Development Phases

1. MSDN (November, 2005) “Leveraging the Role of Testing and Quality Across the Lifecycle to Cut Costs and Drive IT/Business Responsiveness “ 2. Direct Return on Investment of Software Independent Verification and Validation: Methodology and Initial Case Studies, James B. Dabney and Gary Barber, Assurance Technology Symposium, 5 June 2003.


11

7. Description of software systems developed A Library Management System (LMS) help in automating functions of the library. It helps in reducing the time spent in record keeping and management effectively. The management information system application was used to conduct the fault injection experiments. The same application was developed in the following technologies in 3G languages;

Table 2: LMS

A Post Office Management System (POMS) help in

automating functions of the post office and e-Governance. Post Office management System is an application which is created for the use of an automated system all over the country. The information system application was used to conduct the fault injection experiments. The same application was developed in the following technologies in 3G languages;

Table 3: POMS

A Point Of Sales (POS) is retail systems for recording

sales and billing at a cash counter of typical establishment. POS is developed to automate the operations and billing applications of super markets. The information system application was used to conduct the fault injection experiments. The same application was developed in the following technologies in 3G languages;

Table 4: POS

An Audit Tracking System (ATS) is a systems application

used to conduct the fault injection experiments. The application chosen was the Audit Tracking System (ATS) that is used to automate the process for recording, communicating and monitoring all the audit findings that take place within the organization. The same application was developed in the following technologies in 3G languages;

Table 5: ATS

The four applications were simultaneously developed by

different project team and were made mutually exclusive. The application development for the projects followed the same process as described in the quality management system for software development of Trimentus. Applications chosen

to FIM had different varied domain implementation and knowledge for the application was high; it can be independently managed and developed; it covers the entire development life cycle; and the technology used is typical of current commercial applications and technologies in vogue.

SDLC, technology, exclusiveness allows different types of faults to be injected at various phases without bias and enables direct comparison. In this paper, the system contains injected defects common across all projects. The same count of defects (5 numbers) were introduced in each phase of SDLC. The defect review efficiency was assumed to independent of application domain and same technology based defect removal efficiency was considered across all applications.

8. Results of the experiments

8.1 Requirements Review SRS (Software Requirement Specification) document was prepared and used as the basis for development of for all the projects. However, after the review of SRS, defects were injected into the same document. The SRS containing the defects were base lined by project team respectively to be used as basis for the design.

8.2 Design Phase Analysis Design document prepared with (fault injected) SRS as basis. There were several defects observed with “source” as SRS. The Injected defects are major cause for design defects.

8.2.1 Design Review Table 6: SRS And Design Of The Four Applications

8.2.2 Design Defect Amplification: Domain Variant The following chart represents the comparison of amplification index between the applications developed on same technology. The amplification of design defects caused due to the injected requirement defects in the application is evidenced in all domains and least prominent in retail application.

AI Trend - Application Wise

0.5 0.5

0.3

0.5

0.00.10.10.20.20.30.30.40.40.50.5

LMS POMS POS ATS

Design Phase - Application

AI

Figure 5. Amplification Index Trend - Design


12

8.2.3 Calculation of Amplification Index (AI)

The following methodology was used to calculate Requirement Amplification Index (Table 7) (i.e. impact of Requirements defects on Design)

Table 7: Methodology Used To Calculate Requirement Amplification Index

8.2.4 Defects in Design

Various types of known design defects were introduced after design review:

8.2.5 Statistical Analysis and Relationship Based on the AI derived from the above requirement data analysis, a statistical study was carried out understand and analyse the statistical significance and relationship of AI across domains.

A regression analysis using was carried out to under the relationship between a dependent variable, i.e., Design AI to independent variable i.e., Requirement AI.

Minitab tool was used to analyse the data set of Amplification Indexes. Analysis Results:

Table 8: AI Derived From Requirement Data Analysis

Test 1: R Square Test Condition: The larger is the R Square value the better is the model (It would range from 0 (Poor) to 1 (Excellent)

Analysis: The R square value for the prediction equation is 0.19. So the Correlation is not strengthened. So the R Square test Failed. Test 2 : Significance F test Condition: It is Generally Accepted that if the p-value is less than 0.05, the input variable is Significant Analysis: The Significance F Value is 0.55, which is Greater than 0.05.So Significance F test failed

To conclude, by the statistical failure of two tests, it is concluded that there is no correlation between the Amplification Index (AI) of Requirement and Design across domains.

8.3 Coding Phase Analysis Coding was performed with (fault injected) design as basis. There were several defects observed with “source” as Design and Requirements. The Injected defects were the major cause for Code defects detected in Code review.

8.3.1 Code Review of Applications

Table 9: SRS, Design And Code

8.3.2 Code Defect Amplification: Domain Variant The following chart represents the comparison of amplification index between different applications developed on same technology. The amplification of coding defects caused due to the injected design defects in applications is evidenced in all domains and more prominent in education domain.


0.9

0.60.5 0.5

0.00.10.20.30.40.50.60.70.80.91.0

LMS POMS POS ATS

Coding Phase - Applications

AI

Figure 6. Amplification Index Trend – Code


13

8.3.3 Amplification Index for Code The following methodology was used to calculate Design Amplification Index (i.e. impact of Design defects on Code)

Table 10: Methodology Used To Calculate Design Amplification Index

8.3.4 Statistical Analysis and Validation

Similarly, based on the AI derived from the above design data analysis, a statistical study was carried out to understand and analyse the statistical significance and relationship of AI across design phases.

A regression analysis using was carried out to under the relationship between a dependent variable, i.e., Code AI to independent variable i.e., Design AI.

Minitab tool was used to analyse the data set of Amplification Indexes. Analysis Results:

Table 11: AI Derived From Design Data Analysis

Test 1: R Square Test Condition: The larger is the R Square value the better is the model (It would range from 0 (Poor) to 1 (Excellent) Analysis: The R square Value for the Prediction equation is 0.79. A Positive Correlation is identified.

Test 2 : Significance F test Condition: It is Generally Accepted that if the p-value is less than 0.05, the input variable is Significant Analysis: The Significance F Value is 0.10, which is Greater than 0.05.So Significance F test failed.

To conclude, by the statistical analysis by the failure of Significance F test, it is concluded that there is no positive correlation between the Amplification Index (AI) of Design and Code across domains.

9. CONCLUSIONS

9.1 AI Trend Analysis The Amplification Index indicates the extent of damage caused by a defect in various phases of the project. The index increases with every step in the life cycle of the project in all domains. Initial analysis of these defects in the education domain application shows substantial increase in the amplification index across phases compared other application domains.

Within the experimentation limits, it was concluded and validated statistically that there is no statistically significant relationship, and correlation between defects amplification index on requirements vs design and design vs code on applications developed under different domains under a common technology


0.5

0.9

1.9

0.50.6

0.8

0.3

0.5 0.60.5 0.5

1.1

0.00.20.40.60.81.01.21.41.61.82.0

Req Design Coding

Phases

AI

LMSPMOSPOSATS

Figure 7. Amplification Index Trend – Application wise

9.2 Defect Leakage and Distribution Analysis The defect leakage analysis emphasizes the importance of thorough and systematic reviews in the early stages of a software project with an emphasis on defect prevention. The analysis indicates a high increase of cost and effort to remove the defects at later stages. The number of defects increases exponentially as a direct result of defects leaked from previous stages.


14

Defect Leakage - Application Wise

4

7

9

14

34

68

35 5

7

35

6

10

024

68

1012

1416

Req Design Coding Testing

Phases

Def

ects

LMSPOMSPOSATS

Figure 8. Defect Leakage – Application Wise

10. Future Experiments Currently, the study is being extended to analyse the effect of the defects and amplification index in the development phases of the different technologies based projects developed with same product. Guidelines for review time and effort estimation are being computed by analysing and defining the review and test stop criteria. Error seeding during testing can be carried out to define the test stop criteria.

11. Limitations of Experiments The following are the limitations of the experiments:

Causal analysis is relatively subjective to understand the cause of amplified defect. This required detailed review and discussion with project team and technical/technology experts.

Defect removal efficiency percentage used for experiments in same technology is based on a test in a sample requirement, design and code with known defects provided to project members and review efficiency percentage derived from the defects detected.

It is verified that the skill set and domain knowledge of the analysts and programmers working in the projects are same and/or similar. References [1] A. Hall. Seven myths of formal methods. IEEE

Software, 7(5):11-19, September 1990.

[2] C.B. Jones. Systematic Software Development Using VDM. Prentice-Hall International, London, 1986.

[3] S.J. Garland, J.V. Guttag, and J.J. Horning. Debugging larch shared language specifications. IEEE Trans. Software Engineering, 16(9):1044-1057, September 1990.

[4] W. Howden. A functional approach to program testing and analysis. IEEE Trans. Software Engineering, SE-12(10):997-1005, October 1986.

[5] L. J. White. Basic mathematical de_nitions and results in testing. In B. Chandrasekaran and S. Radicchi,

editors, Computer Program Testing, pages 13-24. North-Holland, 1981.

[6] R. DeMillo, R. Lipton, and A. Perlis. Social processes and proofs of theorems and programs. Communications of the ACM, 22(5):803-820, May 1979.

[7] Barry W. Johnson. Design and Analysis of Fault-Tolerant Digital Systems. Addison-Wesley, Massachusetts, 1989.

[8] Daniel Dreilinger, Lijun Lin. Using Fault Injection to Test Software Recovery Code, November 1995

[9] Leme, Nelson G. M.; Martins, Eliane; Rubira, Cecilia M. F. “A Software Fault Injection Pattern System”. Proceedings of the IX Brazilian Symposium on Fault-Tolerant Computing. Florianópolis, SC, Brazil, March 5th-7th, 2001, pages 99-113.

Authors Profile

Mr. Paloli Mohammed Shareef, CISA, CISM, CGEIT, PMP, is the Executive Vice President and Principal Consultant of Trimentus Technologies, is a research scholar at Anna University – Coimbatore with over 12 years of experience in Software Engineering, Quality Management and Information Security. He has special interest in software

reliability and information security management. He has authored 15 technical papers and made presentations in forums and institutions such as CSI, SPIN, Anna Universities, and Professional Engineering Colleges.

Dr. M V Srinath, PhD, is the Professor, Mahendra Engineering College, Namakkal, with over 12 years of experience in Multimedia Instructional Design and Delivery and Principles and Practices of Software Engineering and Web Engineering. He has special interest in Instructional Materials and Media. He has authored over 50 technical/journal papers and the resource person for the

courses organized and conducted by National Institute of Technical Teachers Training and Research (NITTTR) for the teachers of Polytechnics and Engineering colleges.

Dr. S Balasubramanian, PhD, is the Former Director – IPR, Anna University – Coimbatore with over 18 years experience in Industrial Engineering and Software Engineering. He has special interest in research and educational initiatives. He has authored and published over 25 technical and managerial papers. He is the authority in IPR and Trademark

management. He is also the technical reviewer and academic council member for various professional institutions and universities.


15

Design of a 2.2-4.0 GHz Low Phase Noise and Low Power LC VCO

Namrata Prasad1, R. S. Gamad2 and C. B. Kushwah3

1Electronics & Instrumentation Engineering Department, SGSITS, 23, Park Road, Indore, M.P., India – 452003

[email protected]

2 Electronics & Instrumentation Engineering Department, SGSITS, 23, Park Road, Indore, M.P., India – 452003

[email protected]

3 Electronics & Instrumentation Engineering Department, SGSITS, 23, Park Road, Indore, M.P., India – 452003

[email protected]

Abstract: This paper reports a design of an integrated Voltage Controlled Oscillator (LC-VCO) with high oscillation frequency, Low power consumption and Low Phase noise. For obtaining the performance parameters, the design was simulated in 0.18µm CMOS technology. Results of the present deign shows that the oscillation frequency of VCO is 2.2GHz to 4.0 GHz; the power consumption of the VCO at oscillation frequency of 2.2GHz is 16.13 mw and phase noise-143 mdb/HZ. In addition at 3.3 GHz and 4.0 GHz is 15.76 mW and 15.31mW with phase noise -151 mdb/Hz and -207mdb/Hz respectively. We have compared the results of the present design with earlier published work and is presented in table 1.

Keywords: LC-VCO, Low power consumption, low phase noise, cadence.

1. Introduction VCO is an important component of Phase Locked Loop (PLL) which itself is a main block in the Transceiver. Requirements of the VCO are its ability for a high frequency operation, low Power consumption, Low Phase noise and small area. It is strongly recommended to design monolithically integrate all these blocks on a single chip [1]. A CMOS VCO can be built using ring structures, LC resonator circuits and relaxation circuits [2]. A low supply voltage integrated CMOS Voltage-controlled oscillator (VCO) with on chip digital VCO calibration control system [3]. Recently work has been reported for the design of the VCO with low power consumption, low phase noise and high Speed [4-6]. Most application requires that oscillators be tunable and their output frequency is a function of a control input, usually a voltage. An ideal VCO is a circuit whose output frequency is a linear function of its controlled Voltage. Figure 1 shows the block representation of the ideal VCO and Fig.2 shows the ideal curve of VCO for output frequency and controlled voltage [7].

Figure 1. Ideal voltage controlled oscillator

Wout can be calculated as follows: Wout = Wo + Kvco. Vcont (1) Where, Wo is the Intercept, Kvco is the gain\sensitivity and Vcont is the controlled voltage

Figure 2. Ideal curve of VCO for output frequency and

controlled voltage.

2. VCO Design and Analysis

Figure 3. Circuit diagram of LC Tank In designing of an oscillator is to choose a circuit topology or type to reduce losses. Figure 3 is considered those losses which is associated with the inductor. In practical there would also be losses associated with the variable capacitors (varactors) and the Metal Oxide Semiconductor Field Effect





16

Transistors (MOSFETs). It is the active devices. In experimental integrated VCOs the inductors are on-chip spiral inductors with low quality factor that dominates the losses of the VCO tank. It can be shown that the oscillation frequency of the circuit shown in Fig. 3. Assuming ideal varactors and MOSFETs is given:

(2)

It can also be shown, under the same set of assumptions that the gm of each MOSFET is given by: (3)

The main design of this paper is presented in Fig. 4. In this differentially tuned LC-VCO is used to reduce the power consumption of VCO, here PMOS and NMOS are cross coupled to provide better phase noise. The cross-coupled pairs (M1 and M2) provide the negative resistances to compensate the parasitic resistance of an LC-resonator to have better quality factor. The quality factor Q of the inductor is given by: (4)

Where,

wo is the oscillation frequency [rad/s] L is the value of the inductance [H] R is inductor’s equivalent series resistance Q is the Quality Factor

The oscillation frequency of oscillator is given by:

(5)

Where, L is the inductance of LC-tank and C is the capacitance. To achieve low phase noise and low power consumption, we have used complementary cross-coupled structure. Phase noise can be modeled by the modified Lesson's formula [8]. L(FM)=10log (6)

Where, L (FM) phase noise in dBc/Hz, Fm is the frequency offset from the carrier in Hz, f0 is central frequency in Hz, fc is flicker noise corner frequency in Hz, Q is the loaded quality factor of the tuned circuit, F is noise factor, K is Boltzmann's constant in J/K, T is temperature in K, Pav is average power at oscillator output, R is the equivalent noise resistance of the varactor and Kvco is oscillator voltage gain in Hz/V. From equation (6), Kvco dominates the phase noise performance in the modified Lesson's formula, thus

phase noise performance can be improved by reduction Kvco.

Maximum d.c. power dissipation = Vsupply. Ibias (7) Tuning range can be determined as follows:

(8)

where, W o max is the maximum frequency of operation, Wo min is the minimum frequency of operation and Wo is the frequency of operation.

Figure 4. Schematic view of the VCO

3. Simulation Result This work is carried under the environment of cadence software and schematic editor is used for design entry. In this design we have used specter RF simulator for Simulation, by using TSMC 0.18µ technology. The applied voltage is 2v at different center frequencies. Simulation results are obtained with better improvement as compare to earlier work done and is shown in table 1 and output analysis are presented in Fig. 5 and 6. Phase noise simulation of this design is given in Fig.7.


17

Figure 5. Simulation response of the output

Figure 6. Simulation a.c. response of the output voltage

Figure 7. Result of the phase noise

Table 1: Comparison of present work with earlier work

published work:

4. Conclusion In this paper, we have presented a differential tuned LC-VCO with a tuning range of 3.207% at the operating of 3.3GHz and the phase noise of -151mdB. In addition we have also obtained results with higher and lower range center frequency as shown in table 1. Results are presented

Ref. Technology (CMOS)

Center freque-ncy (GHz)

Power consu-mption (mw)

Tuning Range

Phase noise mdB/Hz)

[4]

0.18µm

4.4

4.9

41%

-110

[7]

0.35µm

6

18

17%

-100.2

[8]

0.18µm

2

15

10%

-131.9

0.18µm

2.2

16.13

3.86%

-143

0.18µm

3.3

15.76

3.207

%

-151

This

0.18µm

4.0

15.31

3.25%

-207


18

in table 1 have got improvement over the other recently reported work. We observed from table 1, when center frequency is increasing power consumption, phase noise and tuning ranges are improved. This VCO design is best for the applications where the low power consumption and low phase noise are the main requirements.

Acknowledgment This work has been carried out in SMDP VLSI laboratory of the Electronics and Instrumentation Engineering department of Shri G. S. Institute of Technology and Science, Indore, India. This SMDP VLSI project is funded by Ministry of Information and Communication Technology, Government of India. Authors are thankful to the Ministry for the facilities provided under this project.

References [1] W. Shing, T. Yan, and H. C. Luong, “A 900MHZ

CMOS low phase-noise voltage controlled ring oscillator,” IEEE Transactions on circuits and systems -2: Analog and Digital signal processing, Vol.48, pp. 216-221, Feb. 2001.

[2] Thomas H. Lee, & Ali Hajimiri, “Oscillator Phase Noise: A Tutorial” IEEE Journal of Solid State Circuit” Vol. 35, No.3 March 2000.

[3] Jongsik kim, Jaewook Shin, Seungsoo kim & Hynchol Shin, “A Wide Band LC VCO With Linearized Coarse Tuning Characteristics”, IEEE Transactions on circuit & Systems 2.Vol. 55,No.5 May 2008.

[4] LU Peiming, Huang Shizhen, Song Lianyi,Chen Run “Design of A 2GHZ LOW Phase Noise LC VCO” International Multi Conference Of Engineers and Computer scientists 2009 Vol. 2., IMECS 2009, March 18-20, 2009, Hong Kong.

[5] Paavo Vaananen, Nikomikkola,& Petri Helio, “VCO Design With On-Chip Calibration System” IEEE Transactions On Circuit & System–I: Regular Papers, Vol. 53. No.10, October 2006.

[6] T. Y. Lin, T. Y. Yu, L. W. Ke, G.K. Dehng, “A low Noise VCO with a constant Kvco foGSM/GPRS/EDGE Applications” Media Tek Inc,No.1, Dusing Rd., Hsinchu Science Park, Hsinchu, Taiwan 300,R.o.c. IEEE 2008.

[7] B. Razavi, “A Study of Phase Noise in CMOS Oscillators” IEEE Journal of Solid-State circuit Vol.31 March 1996.

[8] R. M. Weng and J. Y. Lin “A 2.4GHz Low Phase noise Voltage Controlled Oscillator” Department of Electrical Engineering, National Dong Hwa University, Taiwan, R.O.C. PIERS Proceedings, Beijing, China, March 23-27, 2009.

Authors Profile

Namrata Prasad received the B.E. Degree in Electronics and communication Engineering.from S.A.T.I. Vidisha in 2008 and pursuing M.Tech degree in Microelectronics and VLSI Design from S.G.S.I.T.S. Indore. Recently she is working with a project on VCO design and analysis.

R. S. Gamad received the B.E. in Electronics & Communication Engineering from V. University, India in 1995 and M.E. degrees in Digital Techniques & Instrumentation Engineering with honours from Rajiv Ghandhi Technical University Bhopal, India in 2003. He has been working in teaching and research professions since 1996. He is now working as Asst. Prof. in

Department of Electronics & Instru. Engineering of S. G. S. I. T. S. Indore, India. His interested field of research is Dynamic testing of A/D Converter, Design of an A/D converter.

C. B. Kushwah received the B.E. in Electronics & Telecommunication Engineering from Institute of Technology & Management (ITM), Gwalior, India in 2002 and M. Tech. in Microelectronics and VLSI Design from Shri G. S. Institute of Technology & Science, Indore, India in 2009.

During 2002-2007 he was with DCNPL, Indore as BTS Engineer and worked as a Lecturer in MPCT, Gwalior. He is now working as a Lecturer under SMDP-II (VLSI) project in Electronics & Instrumentation Engineering Department of Shri G. S. Institute of Technology & Science. He is interested in analog circuit design, A/D converters and digital front end design.


19

Choase Based Image Encryption Using Block-Based Transformation Algorithm

Kamlesh Gupta1, Sanjay Silakari2

1RJIT, BSF Academy Tekanpur, M.P, India

[email protected]

2UIT,RGPV,Bhopal, M.P. India [email protected]

Abstract: Encryption is used to securely transmit data in open networks. Each type of data has its own features; therefore different techniques should be used to protect confidential image data from unauthorized access. Most of the available encryption algorithms are mainly used for textual data and may not be suitable for multimedia data such as images. In this paper, we introduce a block-based transformation algorithm based on the combination of image transformation and Choase base image encryption algorithm. The original image was divided into blocks, which were rearranged into a transformed image using a transformation algorithm presented here, and then the transformed image was encrypted using the Choase base algorithm. The results showed that the correlation between image elements was significantly decreased by using the proposed technique. The results also show that increasing the number of blocks by using smaller block sizes resulted in a lower correlation and higher entropy.

Keywords—Image correlation, Image encryption, Image

entropy, Permutation, Choase function.

1. Introduction The rapid growth of computer networks allowed large

files, such as digital images, to be easily transmitted over the internet [1]. Data encryption is widely used to ensure security however, most of the available encryption algorithms are used for text data. Due to large data size and real time constrains, algorithms that are good for textual data may not be suitable for multimedia data [2]-[4]. The efficiency and success of e-commerce and business was fueled by the underlying growth of available network bandwidth. Over the past few years, internet-enabled business or e-business has drastically improved revenue and efficiency of large scale organizations. It has enabled organizations to lower operating costs and improves customer’s satisfaction. Such applications require networks which accommodate voice, image, video and protected data. Obviously, privacy must

In most of the natural images, the values of the neighboring pixels are strongly correlated (i.e. the value of any given pixel can be reasonably predicted from the values of its neighbors) [5]-[7]. In order to dissipate the high correlation among pixels and increase the entropy value, we propose a transformation algorithm that divides the image into blocks and then shuffles their positions before it passes

them to the choase based encryption algorithm. By using the correlation and entropy as a measure of security, this process results in a lower correlation and a higher entropy value when compared to using the choase based algorithm alone, and thus improving the security level of the encrypted images. There are two main keys to increase the entropy; the variable secret key of the transformation process and the variable secret key of the choase based algorithm works using an iterative cipher mechanism that is based on the logistic function. The encryption module encrypts the image pixel by-pixel, taking into consideration, in each iteration the values of the previously encrypted pixels. This feedback property, combined with the external secret key of 256-bit, makes our stream cipher robust against cryptanalytic attacks. Furthermore a simple implementation of image encryption achieves high encryption rates on general-purpose computers.

2. Background Encryption is the process of transforming the information

to insure its security with the huge growth of computer networks and the latest advances in digital technologies, a huge amount of digital data is being exchanged over various types of networks. It is often true that a large part of this information is either confidential or private. As a result, different security techniques have been used to provide the required protection [8].

The security of digital images has become more and more

important due to the rapid evolution of the Internet in the digital world today. The security of digital images has attracted more attention recently, and many different image encryption methods have been proposed to enhance the security of these images [9].

Image encryption techniques try to convert an image to

another one that is hard to understand [9]. On the other hand, image decryption retrieves the original image from the encrypted one. There are various image encryption systems to encrypt and decrypt data, and there is no single encryption algorithm satisfies the different image types.

Most of the algorithms specifically designed to encrypt

digital images are proposed in the mid-1990s. There are two




20

major groups of image encryption algorithms: (a) non-chaos selective methods and (b) Chaos-based selective or non-selective methods. Most of these algorithms are Image Encryption Using Block-Based Transformation Algorithm Mohammad Ali Bani Younes and Aman Jantan designed for a specific image format compressed or uncompressed, and some of them are even format compliant. There are methods that offer light encryption (degradation), while others offer strong form of encryption. Some of the algorithms are scalable and have different modes ranging from degradation to strong encryption [10].

Mitra A et al. [11] have proposed a random combinational image encryption approach with bit, pixel and block permutations.Zhi-Hong Guan et al. [12] have presented a new image encryption scheme, in which shuffling the positions and changing the grey values of image pixels are combined to confuse the relationship between the cipher image and the plain image.Sinha A. and Singh K. [13] proposed an image encryption by using Fractional Fourier Transform (FRFT) and JigSaw Transform (JST) in image bit planes. Shujun Li et al. [14] have pointed out that all permutation-only image ciphers were insecure against known/chosen-plaintext attacks. In conclusion, they suggested that secret permutations have to be combined with other encryption techniques to design highly secured images. Maniccam S.S. and Bourbakis N G. [10] proposed image and video encryption using SCAN patterns. The image encryption is performed by SCAN-based permutation of pixels and a substitution rule which together form an iterated product cipher.

Ozturk I. and Sogukpinar I. [15] proposed new schemes

which add compression capability to the mirror-like image encryption MIE and Visual Cryptography VC algorithms to improve these algorithms. Sinha A. and Singh K. [16] proposed a technique to encrypt an image for secure transmission using the digital signature of the image. Digital signatures enable the recipient of a message to authenticate the sender of a message and verify that the message is intact. Droogenbroeck M.V. and Benedett R. [2] have proposed two methods for the encryption of an image; selective encryption and multiple selective encryption. Maniccam S.S., Nikolaos G. and Bourbakis. [17] have presented a new methodology, which performs both lossless compression and encryption of binary and gray-scale images. The compression and encryption schemes are based on SCAN patterns generated by the SCAN methodology.

The proposed algorithm divides the image into random number of blocks with predefined maximum and minimum number of pixels, resulting in a stronger encryption and a decreased correlation. 3. Algorithm CREATE_TRANSFORMATION_ TABLE

Step1: Load Image, Input key, Get ImageWidth and ImageHeight Step2: LowerHorizontalNoBlocks = Int(ImageWidth /10) LowerVerticalNoBlocks = Int(ImageHeight /10)

Step3: Randomize () Step4: HorizontalNoBlocks = RandomNum between (LowerHorizontalNoBlocks and ImageWidth) VerticalNoBlocks = RandomNum between (LowerVerticalNoBlocks and ImageHeight) NoBlocks = HorizontalNoBlocks* VerticalNoBlocks Step5: Seed = | Hash value (Key) | HashVal1 = |Hash val(first half of the Key)| HashVal2 = |Hash val (second half of the Key)| Step6: Randomize using seed Step7: If HashVal1 > HashVal2 Then SEEDALTERNATE = 1 Else SEEDALTERNATE = 2 End If Step8: I = 0 Number-of-seed-changes (N) = 1 While I < NoBlocks R=RandomNum between (zero and NoBlocks -1) If R is not selected Then Assign location R to the block I I +=1 Else If SEEDALTERNATE = 1 Then seed = seed + (HashVal1 Mod I) +1 SEEDALTERNATE = 2 Else seed = seed + (HashVal2 Mod I) + 1 SEEDALTERNATE = 1 Randomize (seed) End If Else Number-of-seed-changes += 1 If Number-of-seed-changes > 500,000 then For K = 0 to NoBlocks -1 If K not selected then Assign location K to Block I I=I+1 End if Next K End if End if End While END CREATE_TRANSFORMATION_TABLE Input: BMP image file, a string Key Output: Transformation table

ALGORITHM PERFORM_TRANSFORMATION For I = 0 to No. Blocks -1 Get the new location of block I from the Transformation table Set block I in its new location END PERFORM_TRANSFORMATION


21

4. Algorithm Choase based feedback image encryption

In this section, we discuss the step by step procedure of

the proposed algorithm for image compression and encryption. The proposed algorithm with block size of 8-bit applies wavelet transform for each block for image compression and 256-bit secret key used for image encryption. The key is used to generate a pad that is then merged with the plaintext a byte at a time.

1. For the image compression, we divide plaintext into blocks of 8-bits.

I1, , I2 , …………………………Ii P0,P1,,P2,,P3,,P4 ..........Pi C0,C1, C2, C3, C4 .........Ci 2. The proposed image encryption process utilizes an

external secret key of 256-bit Long. Further, the secret key is divided into Blocks of 8-bit each, referred as session Keys.

K0,K1,K2,K3,K4 .........K64 (in hexadecimal) here, Ki 's are the alphanumeric characters (0–9 and (A–

F) and each group of two alphanumeric characters represents a session key. Alternatively, the secret key can be represented in ASCII mode as

K0,K1,K2,K3,K4 .........K32 (in ASCII) Here, each Ki represents one 8-bit block of the secret key

i.e. session key. 3. The initial condition ( X 0 ) for the chaotic map and the

initial code C0 are generated from the session keys as:

R = ∑i= 1

32

M1[ Ki ]

X0 =R – [Floor (R)]

C0 = [∑i=1

32

Ki ] mod256

Here Ki , Floor , and M1 are, respectively, the decimal equivalent of the ith session key, the floor function, and mapping from the session, key space, all integers between 0 and 255, into the domain of the logistic map, all real numbers in the interval [0,1].

4. Read a byte from the image file (that represent a block of 8-bits) and load it as plain image pixel Pi..

5. Encryption of each plain image pixel Pi to produce its corresponding cipher image Pixel Ci can be expressed mathematically as:

Ci = Pi+M2∑i= 1

NrXi 1− Xi mod256

Where Xi represents the current input for logistic Map and computed as: Xi = M1[X i-1 + C i-1 + K i] N is the number of iteration of logistic map for its current

input Xi and calculated as: N = K i +1 + C i-1 And M2 maps the domain of the logistic map, back into

the interval [0,255]. 6. Repeat steps 5-6 until the entire image file is

encrypted. 5. An overview of the proposed technique.

Figure 1. Proposed Encryption technique 6. Experiments

The algorithm was applied on a bit mapped (bmp) image

that has the size of 300 pixels x 300 pixels with 256 colors. In order to evaluate the impact of the number of blocks on the correlation and entropy, three different cases were tested. The number of blocks and the block sizes for each case are shown in Table I. Table I Different Cases to Test the Impact of the Number of Blocks on the Correlation and Entropy Case Number of blocks Block size 1 30×30 10 pixels × 10 pixels 2 60×60 5 pixels × 5 pixels3 100×100 3 pixels × 3 pixels.

Table 1: Different Cases to Test the Impact of the

Number of Blocks on the Correlation and Entropy

Case Number

Number of blocks

Block size

1

30×30 10 pixels × 10 pixels

2 60×60 5 pixels × 5 pixels

3 3 100×100 3 pixels × 3 pixels

Case1. .

(a) (b) (c) Figure 2. Results of encryption by using 10 pixels × 10 pixels

blocks. (a) Original image.(b) Transformed image. (c) Encrypted image using transformed followed by the Choase Based algorithm.

Case2.

Pi

+ M1 X

L

Encrypt image

M2 +

+

Ki

Transform Image

Input original image

Block Based Transformation

Ci

Ki - 1


22

(a) (b) (c) Figure 3. Results of encryption by using 60 pixels × 60 pixels

blocks. (a) Original image.. (b) Transformed image. (c) Encrypted image using transformed followed by the Choase Based algorithm

Case3.

(a) (b) (c) Figure 4. Results of encryption by using 100 pixels × 100

pixels blocks. (a) Original image.. (b) Transformed image. (c) Encrypted image using transformed followed by the Choase Based algorithm.

Table2: the comparison of different algorithm based on

correlation and entropy

Commercially available algorithms preceded by the proposed algorithm

System Number of

blocks

Correlation

Entropy

30×30 0.0049 5.5286 60×60 0.0040 5.5439 100×100

0.0034 5.5440

Choase Based

Feedback(256)

300×300

0.0026 5.5437

30×30 0.0063 5.4402 60×60 0.0049 5.5286 100×100

0.0044 5.5407

BLOWFISH(448)

300×300

0.0028 5.5438

30×30 0.0026 5.5437 60×60 0.0040 5.5439 100×100

0.0041 5.5438

TWOFISH (256)

300×300

0.0029 5.5438

30×30 0.0034 5.5440 RijnDael (AES256) 60×60 0.0024 5.5439

100×100

0.0049 5.5438

300×300

0.0016 5.5439

30×30 0.0024 5.5438 60×60 0.0026 5.5437 100×100

0.0034 5.5439

RC4(2048

300×300

0.0034 5.5438

7. Conclusion

In this paper a simple and strong method has been

proposed for image security using a combination of block based image transformation and encryption techniques. Experimental results of the proposed technique showed that an inverse relationship exists between number of blocks and correlation, and a direct relationship between number of blocks and entropy. When compared to many commonly used algorithms, the proposed algorithm resulted in the best performance; the lowest correlation and the highest entropy.

References

[1] W. Lee, T. Chen and C. Chieh Lee, "Improvement

of an encryption scheme for binary images," Pakistan Journal of Information and Technology. Vol. 2, no. 2, 2003, pp. 191-200. http://www.ansinet.org/

[2] M. V. Droogenbroech, R. Benedett, "Techniques for a selective encryption of uncompressed and compressed images," In ACIVS’02, Ghent, Belgium. Proceedings of Advanced Concepts for Intelligent Vision Systems, 2002.

[3] S. Changgui, B. Bharat, "An efficient MPEG video encryption a lgor i thm, " Proc e edings of the symposium on reliable distributed systems, IEEE computer society Press, 1998, pp. 381-386.

[4] S. Fong, P.B. Ray, and S. Singh, "Improving the lightweight video encryption algorithm," proceeding of iasted international conference, single processing, pattern recognition and application, 2002, pp. 25-28.

[5] S.P.Nana'vati., P. K. panigrahi. "Wavelets:applications to image compression- I,". joined of the scientific and engineering computing, vol. 9, no. 3, 2004, pp. 4- 10.

[6] c. Ratael, gonzales, e. Richard, and woods, "Digital image processing," 2nd ed, Prentice hall, 2002.

[7] AL. Vitali, A. Borneo, M. Fumagalli and R. Rinaldo, "Video over IP using standard-compatible multiple description coding, " Journal of Zhejiang

[8] H. El-din. H. Ahmed, H. M. Kalash, and O. S. Farag Allah, "Encryption quality analysis of the RC5 block cipher algorithm for digital images," Menoufia University, Department of Computer Science and Engineering, Faculty of Electronic Engineering. Menouf-32952, Egypt, 2006.

[9] Li. Shujun, X. Zheng "Cryptanalysis of a chaotic image encryption method," Inst. of Image Process. Xi'an Jiaotong Univ., Shaanxi, This paper appears in:

http://www.ansinet.org/


23

Circuits and Systems, ISCAS 2002. IEEE International Symposium on Publication Date: 2002, Vol. 2, 2002, page(s):708,711.

[10] S.S. Maniccam, N.G. Bourbakis, "Image and video encryption using SCAN patterns," Journal of Pattern Recognition Society, vol. 37, no. 4, pp.725– 737, 2004.

[11] A. Mitra, , Y V. Subba Rao, and S. R. M. Prasnna, "A new image encryption approach using combinational permutation techniques," Journal of computer Science, vol. 1, no. 1, p.127, 2006, Available: http://www.enformatika.org

[12] G. Zhi-Hong, H. Fangjun, and G.Wenjie , "Ch a o s - based image encryption algorithm," Department of Electrical and computer Engineering, University of Waterloo, ON N2L 3G1, Canada. Published by: Elsevier, 2005, pp. 153-157.

[13] A. Sinha, K. Singh, "Image enc rypt ion by using Fractional Fourier Transforms and Jigsaw transform in image bit planes," Source: optical engineering, spie-int society optical engineering, vol. 44, no. 5 , 2005, pp.15-18.

[14] Li. Shujun, Li. Chengqing, C. Guanrong, Fellow., IEEE., Dan Zhang., and Nikolaos,G., Bourbakis Fellow., IEEE. "A general cryptanalysis of permutation-only multimedia encryption algorithms," 2004, http://eprint.iacr. Org/2004/374.pdf

[15] I. Ozturk, I.Sogukpinar, "Analysis and comparison of image encryption algorithm," Journal of transactions on engineering, computing and technology December, vol. 3, 2004, p.38. http: //www.enformatika.org/

[16] A. Sinha , K. Singh, "A technique for image encryption using digi tal signature," Source: Opt ics Communications, vol.218, no. 4, 2003, pp.229-234.http://www.elsevier.com/

[17] S . S . Maniccam. , G.Nikolaos , and Bourbakis, "Lossless image compressionand encryption using SCAN," Journal of: Pattern Recognition, vol. 34, no. 6: , 2001, pp.1229– 1245.

Author Profile

Kamlesh Gupta received B.E degree in Computer science and Engineering. FromRajiv Gandhi Proudyogiki Vishwavidyalaya, Bhopal in 2001 and M.Tech degree in Computer science and Engineering. from Rajiv Gandhi Proudyogiki Vishwavidyalaya,Bhopal in 2005 Currently pursuing Ph...D from Rajiv

Gandhi Proudyogiki Vishwavidyalaya, Bhopal Research interests include image processing and image security . He is working as Lecturer in the Department of Information Technology, Rustamji Institute of Technology, BSF Academy Tekanpur, Gwalior Madhya Pradesh, India

http://www.enformatika.org

http://eprint.iacr

http://www.enformatika.org/

http://www.elsevier.com/


24

Use of CASE-technology and provision of SQL-data security for special transportation management

Lyazat Naizabayeva1,

1Kazakh-British Technical University,

59 Tole-bi street, Almaty, 050000, Republic of Kazakhstan [email protected]

Abstract: Logical model of valuable cargo special transportation management was developed, Computer-Aided Software/System engineering (CASE) – “Entity-Relationship” flowcharting technology – applied, physical database – created in the MS SQL Server, SQL-data security – provided by built-in Database management system (DBMS).

Keywords: Сomputer-Aided Software/System engineering,

AllFusion Erwin Data Modeler, create certificate, symmetric and asymmetric key.

1. Introduction Some types of goods make manufacturers’ head bang. How to deliver perishable goods in time? How to transport OOG? How to protect hazardous goods while transportation? How to transport animals? If a transportation company, responsible for delivery of such goods, fails to render such services at a qualitative and professional level, the goods owner will incur material losses. Perhaps, such an owner will be able to cover losses due to poor services, but what should be done with irretrievably damaged goods and wasted time? This is why special goods shall be transported at a highly professional level. Special transportation is special delivery of special goods to satisfy needs of highly defense and state essential requirements. The following shall be referred to special transportation: delivery of hazardous goods, OOG, goods that require special temperature conditions, highly valuable and fragile goods, escorted transportation. Much attention should be focused on delivery of general cargo to a destination point, delivery of out-of-gauge (OOG), liquid, bulk, perishable, LCL goods; goods storage at consolidated warehouses. To transport hazardous goods, such as explosives, poisons, chemicals, it is necessary to agree the route, transportation conditions, and most likely, escort the goods and control its temperature conditions, loading and unloading. It takes additional time to obtain all approvals, so, it is required that all employees, involved in goods transportation strictly observe both delivery schedule and transportation route. This document is dedicated to arrangement of software system automation to control and provide security of special vehicles movement information system, namely securities delivery control.

2. Information system design Today, in conditions of programming automation intensive development, the problem of adequate reflection of a real subject field against information model abstraction is very acute today. If a project is incorrect, the system, designed based thereon, will not satisfy users’ needs and will require costly modifications or complete redesign [1]. Hence, database logical design is required: design of a general information model based on some users’ data models and being independent of currently used DBMS and other physical parameters. Database design shall be a strategy as to determining company’s information requirements during a long period of time. Database design, based on a relation model, has the following advantages to other ones: • Independence of a logical structure of physical

parameters and user’s opinion. • Database structure flexibility – structural solutions do

not restrain future possibilities to meet different requirements.

2.1 Use of Computer-Aided Software/System engineering (CASE) technology

CASE technology is usually considered to be an information system design method plus tools, which allow • Visually simulating the subject field • Analyzing its model at each stage of information system

design and support • Developing users’ applications The main objective of CASE systems and tools is to separate software engineering from its coding and further engineering stages (testing, filing, etc.), as well as to computerize the entire software engineering process. Engineering of advanced information systems requires application of specials methods and tools. It is no wondering that in recent years system analysts and designers are considerably interested in CASE technologies and tools, which allow integrating and automating all stages of software engineering. In this paper, logical database was designed by CASE tool AllFusion Erwin Data Modeler (Erwin) [3], “Entity-Relationship” was developed (Fig. 1). This diagram shows project intuitive interface and may be used by users for idea sharing.



25

ERwin is a key solution for database for design and support of data bases, data marks and data storage, as well as company’s data resource model. ERwin models visualize data structure to simplify data organization and management, complicated data integration and database design and development techniques. Thereby, data based development is simplified and speeded up, and its quality and reliability are improved significantly. ERwin automatically generates tables and thousands of code lines, stored procedures and flip-flops for advanced vendors’ data bases. Complete-Compare technology, used in the system, allows iteratively designing so that the model is always synchronized with the data base. ERwin may be also used for design and servicing of the entire database life cycle.

Figure 1. “Entity-Relationship” diagram for the specialized transportation system database in ERwin

2.2 Conceptual data bank creation for special transportation management in the SQL Server

The next step was the verification of any operational use of organization’s data related to the data processing, and the exclusion of all useless and repetitive data. In the process of database design, to solve tasks of data doubling minimization and facilitation of data processing and updating procedures, the relations were normalized. The tables of designed database are in 3rd normal form (3NF) accordingly to Dr.E.F.Codd [2]. The physical design phase consists of associating of database logical structure and physical environment of storage with purpose to ensure the most efficient data allocation that is the mapping of the database logical structure in the storage

structure. The following issues are in consideration: the stored data allocation in the memory space and the selection of efficient methods of access to different components of the “physical” database. Solutions made within this phase make critical impact on the system performance. SQL Server has advantages over other DBMS, namely: simplified installation, development and use, as well as scalability, creation of data banks and system integration with other server software. Another factor, determining the choice of DBMS MS SQL Server in this work is speed. In relation DBMS, speed is time, required to running query and return of query processing results to the user. SQL Server is more than just simple query running tool, it provides much more opportunities. All leading DBMS providers prefer SQL. Relation database and software they work with may be carried over from one DBMS to another with minimum costs for modifications and staff training. Software tools in DBMS in PC, such as query software, report program and application generators, integrate with relation data bases of different types. Thus, SQL provides independence from certain DBMS and this is why it is in a high demand. The database has been designed for our system using the MS SQL Server tools, and is based on eight tables: Account contains data of the company by the business account, Courier contains information of the route, DT_Declaration – information of the orders, DT_DeclareEvents – event codes in the course of order processing, which may be as follows: order generation, sorting out for the route, transfer to the courier, order rejection, shift of the order by date, adjustments to details, closed-out, open, false call, prohibited order, changed route; DT_DeclareStatus contains information of possible statuses of the order. The order status may be as follows: new, sorted out, en route, shifted by date, adjusted, accepted, open, false call, rejected, prohibited. DT_DeclareEventScheme defines the flow chart related to order processing. There is the order status BeforeStatus, which was prior to an operation. After the operation with the order is accomplished (the operation is determined by the Event Code), the order status is changed to AfterStatus. DT_Associative contains data of the company by the relevant associative words. The associative words are used to automate data entry – the logistician enters an associative word and fills in all fields on a one-time basis, which contain data of the company (company name, business account, telephone, contact person, working time, lunch time, comments). All information is stored in this table. When the associative word is entered next time, the system finds it in the table and automatically displays all data of the company. The database table relation diagram [2] has been developed. The diagram is shown on Figure 3. One of the most important elements of the database design is the development of the database protection. The protection has two aspects: protection against failures and protection against unauthorized access. The file back-up strategy is developed to ensure the failure protection. To ensure the protection against unauthorized access, each user will obtain the access only in compliance with his/her access rights. When developing distributed information systems of transportation management as to the client-server interrelation, the following criteria were focused on:


26

• Personal database carry-over to the server for its further use as a corporate database;

• Query run for the corporate database in the server, on the user’s PC;

• Development of the user’s application for remote access to the corporate database from the user’s PC;

• Server administration from the user’s PC; • And, finally, the most important section for special

information security: tabular data encryption.

Figure 2. The diagram of a database of a specialized -transport in MS SQL Server

2.3 Use of data encryption methods in MS SQL Server It is well known that one of the most important database components is database security. Data security has two aspects: protection against errors and unauthorized access [4]. To protect from errors, a data back-up is developed. To protect against unauthorized access, each user is provided with access in compliance with his/her access rights only. Starting with version 2005, MS SQL Server provides opportunity of data encryption; this project demonstrates three out of four data encryption methods by certificated, asymmetric keys, symmetric keys, and standard encryption by passwords.

2.3.1 Creation of certificates and use of cryptographic functions

Certificate is present in the database in the form of an object, SQL Server Management Studio provides current certificates, symmetric and asymmetric keys (Fig. 3) in container of Databases \data_base_name\ Security\ Certificates:

Figure 3. Review of all certificates.

Developed database DBI provides a table dbo.SecretTable with only one column Secret of nvarchar type; you can input a text, encrypted by a certificate. At first, a certificate is created, using a CREATE CERTIFICATE COMMAND. A standard option of this command is as follows:

USE DBI; CREATE CERTIFICATE SelfSignedCertl

ENCRYPTION BY PASSWORD = 'P@sswOrd ' WITH SUBJECT = Проверка шифрования',

START_DATE = '03/10/2006';

Please note that to create a certificate you do not need any certification center – all required tools are already built in the SQL Server. However, you can download a certificate into the database, which was generated by an external certification center and stored in a file (a private key shall be in a separate file), for instance:

USE DBI; CREATE CERTIFICATE ExternalCertl FROM FILE - 'C:\Certificates\Certl.cer' WITH PRIVATE KEY (FILE – 'C:\Certificates\CertlKey.pvk', DECRYPTION BY PASSWORD = 'P@sswOrd); GO

Moreover, existing certificate may be extracted from setup .NET, signed by this certificate or from signed used file. DECRYPTION BY PASSWORD Parameter will request to provide password, used for this certificate protection. ENCRYPTION BY PASSWORD parameter will identify the password required for data decryption and secured by the certificate (the latter is not required for data encryption). If this parameter is not used, existing certificate will be automatically secured by the database master key. This key cannot be created automatically. To get opportunity to use this key, it should be created in advance:

USE DB1;


27

CREATE MASTER KEY ENCRYPTION BY PASSWORD = 'P@sswOrd';

In addition to the password, database master key is also secured by the service master key. This key is automatically generated in SQL Server when being installed. When using the database master key, a great care must be taken: if you reinstall the server (hence, the service master key will be changed), in this case, encrypted data may be lost. To prevent it, it is necessary to back up the database master key or export the service master key to a file, using a backup service master key command. Obligatory parameter subject of a create certificate command will identify a purpose of the certificate issue (its value will be put into the relevant certificate cell in compliance with X.509vl standard). Upon creation of a certificate, the latter may be used for data encryption. A special function EncryptByCert is used for purpose:

INSERT INTO SecretTable values(EncryptByCert(Cert_ID('SelfSignedCertl'),N 'Секретные данные'));

If any user runs query for table SecretTable, he/she may be astonished by the obtained results (Fig. 4).

Figure 4. Query result for table SecretTable

Please note that function EncryptByCert accepts not only the certificate itself, but its identifier as an initial parameter. Required identifier may be easily obtained, using Cert_ID function. Encrypted data may be decrypted by using a DecryptByCert function. The only problem in using this function is that it returns decrypted information with data of varbinary type, so it is recommended to convert this data type to nvarchar:

SELECT (Convert(Nvarchar(100), DecryptByCert(Cert_ID('SelfSignedCertl'), Secret, N'P@sswOrd'))) FROM SecretTable;

Initial parameter, accepted by a DecryptByCert function is a certificate identifier, returned with the same function cert_ID; second parameter is a string value (or a variable, or a column description, as in our case); the third parameter is a password, securing the certificate, being generated.

2.3.2 Creation of asymmetric keys Let’s review the following encryption method by asymmetric keys. An asymmetric key differs from the certificate by absence of additional fields with information on who, for what purpose, for what period, etc. this key was provided. Current asymmetric keys are in a container Asymmetric Keys, placed in the same place where the container Certificates is. The asymmetric key is used almost in the same way. First of all, it is necessary to create an asymmetric key:

CREATE ASYMMETRIC KEY AsymKeyl WITH ALGORITHM = RSA_512 ENCRYPTION BY PASSWORD = 'P@sswOrd';

Please note that when creating an asymmetric key, it is necessary to provide such key length in addition to the password. You have three options: 512, 1024 and 2048 bits. Afterwards, you may encrypt and decrypt data by this key:

INSERT INTO SecretTable values (EncryptByAsymKey (AsyrnKey_ID ('AsymKeyl') , N 'Секретные данные')) ;

SELECT (Convert(Nvarchar(100), DecryptByAsymKey (AsyraKey_ID ('AsymKeyl'),Secret, N'P@sswOrd') )) FROM SecretTable;

Figure 5. Query result for encrypt data by asymmetric key

2.3.3 Creation of symmetric keys Let’s review the following encryption method by symmetric keys. Faster algorithms are used for symmetric keys creation. Symmetric key themselves are also created as database objects and may be secured by a certificate, other symmetric key, asymmetric key or just a password. You can find them in a Symmetric Keys container.

When using symmetric keys, data encryption process is faster than when asymmetric algorithms are used, so working with large databases, it is recommended to use exactly symmetric keys. There are some differences in using symmetric keys. Firstly, when creating a symmetric key, you may secure it not only by a password, but also by other symmetric key, asymmetric key or certificate. Secondly, when creating a symmetric key, you may identify one of eight encryption algorithms, supported by SQL Server [n]. Symmetric key creation procedure itself may be as follows:

CREATE SYMMETRIC KEY SymKeyl WITH ALGORITHM = AES_128 ENCRYPTION BY PASSWORD = 'P@sswOrd';

Prior to any key use (for data encryption or decryption), it is necessary to open it. If you open it once during the user’s working session, it is more than enough:

OPEN SYMMETRIC KEY SymKeyl DECRYPTION BY PASSWORD = 'P@sswOrd';

Afterwards, we use it as usually. Just function names differ: INSERT INTO SecretTable values(EncryptByKey(Key__GUID('SymKeyl'), N'Секретные данные')); GO

Please note that while data decryption, it is unnecessary to provide symmetric key name and password to the DecryptByKey function. Key data, opened by an OPEN SYMMETRIC KEY command will be provided automatically:

SELECT (Convert(Nvarchar(100), DecryptByKey(Secret))) FROM SecretTable;


28

Figure 6. Encrypted data query run results

SQL Server also allows encrypting data so easily, using a password. For this purpose, an EncryptByPassPhrase function is used. In case of a standard option, this function requires only password and data to be encrypted. INSERT INTO SecretTable values(EncryptByPassphrase(' P@sswOrd', N'Секретные данные') ) ; GO

Data shall be decrypted by a DecryptByPassphrase function: SELECT(Convert(Nvarchar(100), DecryptByPassphrase('P@sswOrd', Secret))) FROM SecretTable;

Figure 7. Data decryption results

It is worth noticing that it is unnecessary to limit oneself by built-in SQL Server tools only. This server also allows using queries to setup .NET in the Transact-SQL code. So, for purposes of data encryption, classes from System.Security.Cryptography namespace in .NET Framework or own setups may be used [6].

3. CONCLUSION As a result of developed information model for special goods transportation safety and efficient management in the course of the optimization process, the following tasks shall be solved: city traffic analysis, identification of an efficient route for securities delivery in a city, considering city traffic, option efficiency assessment; reliable information system protection was provided; considered efficient operation of specialized vehicles.

Information model, adapted for the goods transportation standards, was developed, implemented and is used now for courier services [5]. It allows automating staff work and render more qualitative services to customers.

References [9] A.J. Brast, S. Forte. Development of Microsoft-Based

Applications. Master Class./ Translated from English.-Moscow: Russian Redaktsiya Publishing House, 2007.

[10] T. Connoli, K. Begg. Databases. Designing, realization and support. Theory and practice. 3rd edition, “Williams” publishing house, Moscow, 2003.

[11] S.V. Maklakov Creation of information systems with an AllFusing Modeling Suite. – M.: Dialogue - MIFI, 2003.

[12] R.N. Mikheyev MS SQL Server for administrators.- SPb.: BKhB-Petersburg, 2006.

[13] L.Naizabayeva. “Information System Modeling to Control Transport Operations Process”. In Proceedings of the International MultiConference of Engineers and Computer Scientists (IMECS), Hong Kong, рр. 1813-1816, 2009

[14] A. Troelsen, PRO C# 2005 and The .NET 2.0 Platform, Third Edition, après, Sankt- Petergburg -Kiev, 2007.

Author Profile

Lyazat Naizabayeva majors in math, Kazakh State University after S.M.Kirov (1986); was awarded degree of Ph.D. (candidate) Physical and mathematical sciences (High Academic Attestation Committee of USSR, 1992); Academic title Associate-professor of Informatics, Computer Systems and Management (High Attestation Committee of Republic of Kazakhstan, 2003 ).

Now, she is an Assistant - Professor in Kazakh-British University.


29

Handwritten Tamil Character Recognition Using SVM

Prof. Dr.J.Venkatesh1, C. Sureshkumar2

1 Assistant Professor – Systems and Production, Anna University Coimbatore, Tamilnadu, India.

[email protected] 2Assistant Professor- CSE, J.K.K.M College of Technology, Tamilnadu, India.

[email protected]

Abstract: Hand written Tamil Character recognition refers to the process of conversion of handwritten Tamil character into Unicode Tamil character. The scanned image is segmented into paragraphs using spatial space detection technique, paragraphs into lines using vertical histogram, lines into words using horizontal histogram, and words into character image glyphs using horizontal histogram. Each image glyph is subjected to feature extraction procedure, which extracts the features such as character height, character width, number of horizontal lines(long and short), number of vertical lines(long and short), horizontally oriented curves, the vertically oriented curves, number of circles, number of slope lines, image centroid and special dots. The extracted features considered for recognition are given to Support Vector Machine (SVM) where the characters are classified using supervised learning algorithm. These classes are mapped onto Unicode for recognition. Then the text is reconstructed using Unicode fonts. This character recognition finds applications in document analysis where the handwritten document can be converted to editable printed document. This approach can be extended to recognition and reproduction of hand written documents in South Indian languages.

Keywords: Character recognition, Unicode, Support Vector Machines (SVM). 1. Introduction Tamil is an ancient language with a rich literary tradition and Ancient India was popular in several fields such as medicine, astronomy and business. Ancient people recorded their knowledge in various fields in palm leaves. The handwritten text written in palm leaves decayed over a period of time. It is very difficult to preserve them in the same form. This paper proposes a new approach for converting handwritten Tamil script using unicode. The style of writing and the font were different compared to present day scripts. Lot of software tools is available only to read present day printed Tamil text with better recognition and accuracy. 1.1 Tamil Language Tamil is a South Indian language spoken widely in Tamilnadu in India. Handwritten character recognition is a difficult problem due to the great variations of writing styles, different size and orientation angle of the characters. Among different branches of handwritten character recognition it is easier to recognize English alphabets and numerals than

Tamil characters. Tamil has the longest unbroken literary tradition amongst the Dravidian languages. Tamil is inherited from Brahmi script. The earliest available text is the Tolkaappiyam, a work describing the language of the classical period. There are several other famous works in Tamil like Kambar Ramayana and Silapathigaram but few supports in Tamil which speaks about the greatness of the language. For example, Thirukural is translated into other languages due to its richness in content. It is a collection of two sentence poems efficiently conveying things in a hidden language called Slaydai in Tamil. Tamil has 12 vowels and 18 consonants. These are combined with each other to yield 216 composite characters and 1 special character (aayutha ezhuthu) counting to a total of (12+18+216+1) 247 characters. 1.2 Vowels

Tamil vowels are called uyireluttu (uyir – life, eluttu – letter). The vowels are classified into short (kuril) and long (five of each type) and two diphthongs, /ai/ and /auk/, and three "shortened" (kuril) vowels. The long (nedil) vowels are about twice as long as the short vowels. The diphthongs are usually pronounced about 1.5 times as long as the short vowels, though most grammatical texts place them with the long vowels.

1.3 Consonants

Tamil consonants are known as meyyeluttu (mey - body, eluttu - letters). The consonants are classified into three categories with six in each category: vallinam - hard, mellinam - soft or Nasal, and itayinam - medium. Unlike most Indian languages, Tamil does not distinguish aspirated and unaspirated consonants. In addition, the voicing of plosives is governed by strict rules in centamiḻ. Plosives are unvoiced if they occur word-initially or doubled. Elsewhere they are voiced, with a few becoming fricatives intervocalically. Nasals and approximants are always voiced. As commonplace in languages of India, Tamil is characterised by its use of more than one type of coronal consonants. Retroflex consonants include the retroflex approximant , which among the Dravidian languages is also found in Malayalam (example Kozhikode), disappeared from Kannada in pronunciation at around 1000 AD (the dedicated letter is still found in Unicode), and was never present in Telugu. Dental and alveolar consonants also contrast with




30

each other, a typically Dravidian trait not found in the neighboring Indo-Aryan languages.

1.4 Tamil Unicode The Unicode Standard is the Universal Character encoding scheme for written characters and text. It defines the uniform way of encoding multilingual text that enables the exchange of text data internationally and creates the foundation of global software. The Tamil Unicode range is U+0B80 to U+0BFF [3].The Unicode characters are comprised of 2 bytes in nature. For example, the Unicode for the character is 0B85; the Unicode for the character is 0BAE+0BC0. The Unicode is designed for various other Tamil characters. 2. Tamil character recognition functional block diagram The schematic block diagram of handwritten Tamil Character Recognition system consists of various stages as shown in figure. They are Scanning phase, Preprocessing, Segmentation, Feature Extraction, Classification, Unicode mapping and recognition and output verification.

Scan Document

Preprocessing

Segmentation

Feature Extraction

Classification (SVM)

Unicode Mapping

Recognized Text

Figure 1. Handwritten Character Recognition System 2.1 Character Recognition Functions—Phase I This phase includes the scanning, preprocessing, segmentation and feature extraction.

2.2 Scanning A properly printed document is chosen for scanning. It is placed over the scanner. A scanner software is invoked which scans the document. The document is sent to a program that saves it in preferably TIF, JPG or GIF format, so that the image of the document can be obtained when needed. This is the first step in OCR. The size of the input image is as specified by the user and can be

of any length but is inherently restricted by the scope of the vision and by the scanner software length. 2.3 Preprocessing This is the first step in the processing of scanned image. The scanned image is pre processed for noise removal. The resultant image is checked for skewing. There are possibilities of image getting skewed with either left or right orientation. Here the image is first brightened and binarized. The function for skew detection checks for an angle of orientation between ±15 degrees and if detected then a simple image rotation is carried out till the lines match with the true horizontal axis, which produces a skew corrected image.

Figure 2. Histograms for skewed and skew corrected images

Figure 3. Original Texts

Figure 4. Character Segmentation 2.4 Segmentation

After pre-processing, the noise free image is passed to the segmentation phase, where the image is decomposed into individual characters. Fig.4 shows the image and various steps in segmentation. Algorithm for segmentation: (1) The binarized image is checked for inter line spaces. (2) If inter line spaces are detected then the image is segmented into sets of paragraphs across the interline gap. (3)The lines in the paragraphs are scanned for horizontal space intersection with respect to the background. Histogram of the image is used to detect the width of the horizontal lines. Then the lines are scanned vertically for vertical space intersection. Here histograms are used to detect the width of the words. Then the words are decomposed into characters using character width computation. 2.5 Feature extraction The next phase to segmentation is feature extraction where individual image glyph is considered and extracted for features. Each character glyph is defined by the following attributes: (1) Height of the character. (2) Width of the


31

character. (3) Numbers of horizontal lines present—short and long. (4) Numbers of vertical lines present—short and long. (5) Numbers of circles present. (6) Numbers of horizontally oriented arcs. (7) Numbers of vertically oriented arcs. (8) Centroid of the image. (9) Position of the various features. (10) Pixels in the various regions. 3. Character Recognition Functions Phase II The second phase of the Character Recognition functions consists of classification and Unicode mapping and recognition strategies. 3.1 Classification The various classification methods are as follows: 3.1.1 A typical rule based Classifier The height of the character and the width of the character, various distance metrics are chosen as the candidates for classification when conflict occurs. Similarly, the classification rules are written for other characters. This method is a generic one since it extracts the shape of the characters and need not be trained. When a new glyph is given to this classifier block it extracts the features and compares the features as per the rules and then recognizes the character and labels it. 3.1.2 Support Vector Machine based classifier The architecture chosen for classification is Support Vector machines, which in turn involves training and testing the use of Support Vector Machine (SVM) classifiers has gained immense popularity in recent years. SVMs have achieved excellent recognition results in various pattern recognition applications [1]. Also in off-line character recognition they have been shown to be comparable or even superior to the standard techniques like Bayesian classifiers or multilayer perceptrons. SVMs are discriminative classifiers based on vapnik’s structural risk minimization principle. They can implement flexible decision boundaries in high dimensional feature spaces. The implicit regularization of the classifier’s complexity avoids over fitting and mostly this leads to good generalizations. Some more properties are commonly seen as reasons for the success of SVMs in real-world problems. [11] The optimality of the training result is guaranteed. 3.2 Classification using SVM Support Vector Machine (SVM) is a classifier which performs classification tasks by constructing hyper planes in a multidimensional space [12] .It supports both classification and regression tasks. It is classified into two types namely 1. Classification SVM type-1(also known as C -SVM classification). 2. Classification SVM type-2 (also known as nu - SVM classification).

Figure 5. Classification using SVM

Figure 6. The SVM Classification Algorithm 3.2.1 Classification SVM Type-1 For this type of SVM, training involves the minimization of the error function:

--- (1) subject to the constraints:

--- (2) Where C is the capacity constant, w is the vector of Coefficients, b a constant and ξi are parameters for handling nonseparable data (inputs). The index i label the N training cases [13]. Note that y±1 represents the class labels and xi is the independent variables. The kernel φ is used to transform data from the input (independent) to the feature space. It should be noted that the larger the C, the more the error is penalized. Thus, C should be chosen with care to avoid over fitting. 3.2.2 Classification SVM Type-2 In contrast to Classification SVM Type-1, the Classification SVM Type 2 model minimizes the error function:

--- (3) subject to the constraints:

--- (4)

Input Space Input Space

Margin

Mapping Solution


32

3.3 Kernel Function s There are number of kernels that can be used in Support Vector Machine models. These include 1. Linear, 2.Polynomial, 3.Radial basis function (RBF), 4.Sigmoid.

Figure 7. Sample SVM Classification SVM consists of a learning module (svm_learn) and a classification module (svm_classify)

Figure 8. Sample identified character

Figure 9. Sample Character Encoding

4. Unicode Mapping The Unicode standard reflects the basic principle which emphasizes that each character code has a width of 16 bits. Unicode text is simple to parse and process and Unicode characters have well defined semantics [3] [7]. Hence Unicode is chosen as the encoding scheme for the current work. After classification the characters are recognized and a mapping table is created in which the unicodes for the corresponding characters are mapped.

Figure 10. Sample Tamil Letters for Unicode Characters 5. Character recognition The scanned image is passed through various blocks of functions and finally compared with the recognition details from the mapping table [6] from which corresponding unicodes are accessed and printed using standard Unicode fonts so that the Character Recognition is achieved. 6. Conclusion Character Recognition is aimed at recognizing handwritten Tamil document. The input document is read preprocessed, feature extracted and recognized and the recognized text is displayed in a picture box. The Tamil Character Recognition is implemented using a Java Neural Network. A complete tool bar is also provided for training, recognizing and editing options. Tamil is an ancient language. Maintaining and getting the contents from and to the books is very difficult. Character Recognition eliminates the difficulty by making the data available in handwritten format. In a way Character Recognition provides a paperless environment. Character Recognition provides knowledge exchange by easier means. If a knowledge base of rich Tamil contents is created, it can be accessed by people of varying categories with ease and comfort.

X1

X2

f (x)=0


33

References

[1] C. Papageorgiou and T. Poggio, “A trainable system

for object detection,” International Journal of Computer Vision, vol. 38, no. 1, pp. 15–33, 2000.

[2] G. Guodong, S. Li, and C. Kapluk, “character recognition by support vector machines,” in Proc. IEEE International Conference on Automatic Face and Gesture Recognition, 2000, pp. 196–201.

[3] K. Jonsson, J. Matas, J. Kittler, and Y. Li, “Learning Support Vectors for face verification and recognition,” in Proc. IEEE Int Conf on AutomaticFace and Gesture Recognition, 2000.

[4] B. Heisele, P. Ho, and T. Poggio, “Face recognition with support vector machines: global versus component-based approach,” in ICCV, 2001, pp. 688–694.

[5] A. Mohan, C. Papageorgiou, and T. Poggio, “Example-based object detection in images by

Components,” IEEE Trans. Patt. Anal. Mach. Intell., vol. 23, pp. 349–361, 2001.

[6] B. Scholkopf, P. Simard, A. Smola, and V. Vapnik, “Prior knowledge in support vector kernels,” in Advances in Neural Inf. Proc. Systems, vol. 10. MIT Press, 1998, pp. 640–646.

[7] O. Chapelle, P. Haffner, and V. Vapnik, “SVMs for histogram-based image classification,” IEEE Transactions on Neural Networks, special issue on Support Vectors, 1999.

[8] F. Jing, M. Li, H. Zhang, and B. Zhang, “Support Vector Machines for region-based image retrieval,” in Proc. IEEE International Conference on Multimedia and Expo, 2003.

[9] S. Belongie, C. Fowlkes, F. Chung, and J. Malik, “Spectral partitioning with indefinite kernels using the nystr¨om extention,” in ECCV, part III, Copenhagen, Denmark, may 2002,

[10] C. Wallraven, B. Caputo, and A. Graf, “Recognition with local features: the kernel recipe,” in Proceedings of the International Conference on Computer Vision, vol. I, 2003, p. 257ff.

[11] T. Evgeniou, M. Pontil, and T. Poggio, “Regularization networks and support vector machines,” Advances in Computational Mathematics, vol. 13, pp. 1–50, 2000.

[12] V. Vapnik, The nature of statistical learning Theory. John Wiley and sons, New York, 1995.

[13] D. M. J. Tax and R. Duin, “Uniform object generation for optimizing one-class classifiers,” Journal of Machine Learning Research, Special Issue on Kernel methods, no. 2, pp. 155–173, 2002.

Authors Profile

Prof. Dr. J. Venkatesh received a MBA degree in 1997 and a PhD in System Information in 2008, both from the University of Bharathiar, Tamilnadu, India. He is working as an Assistant Professor at Anna University Coimbatore, Tamilnadu, India, specialised in the field of Systems and Production. He published many papers on computer vision applied to automation,

motion analysis, image matching, image classification and view-based object recognition and management oriented empirical and conceptual papers in leading journals and magazines. His present research focuses on statistical learning and its application to computer vision and image understanding and problem recognition. Mr. C. Sureshkumar received a ME degree in Computer Science in 2006, from the University of Anna, Tamilnadu, India. He is a part-time PhD research scholar in the Department of Computer Science and Engineering, Anna University Coimbatore. His main interests in research are Hand written Tamil Character recognition refers to the process of conversion of handwritten Tamil character into printed Tamil character.


34

Enhancement in the Identities-Exchange Process during the Authentication Process

Amr M. Kisk1, Nagy W. Messiha2, Nabil M. A. Ayad 3, Nawal A. Elfeshawy4, and Fathi E. Abdel-Samie5

1Egyptian Atomic Energy Authority EAEA

[email protected]

2Faculty of Electronic Engineering, Menouf, Egypt [email protected]

3Egyptian Atomic Energy Authority EAEA [email protected]



Abstract: The key exchange is the most important target to the hackers. The asymmetrical encryption algorithms are used for the key-exchange in the authentication process. These algorithms help the authentication protocols to authenticate the networks. The enhancement of the key-exchange and identities-exchange is presented in this paper. The enhancement of the key-exchange keeps the shared keys to be secure from the man-in-the-middle attacks.

Keywords: Public key; Key-Updating; Secret Key; Shared Key.

1. Introduction Wireless Local Area Network (WLAN) is one of the fastest-growing technologies. The demand for connecting devices without the use of cables is increasing everywhere. WLAN can be found in the office buildings, and in many other public areas [1]. The security in WLAN is based on cryptography, the science and art of transforming messages to make them secure and immune to attacks. Cryptography can be used to authenticate the sender and receiver of the message to each other within WLAN. The data security in WLAN needs a key for encryption and decryption processes [2]. The key exchange should be safer to avoid the attacks to get that key. The authentication protocols use an asymmetrical encryption algorithm for the key exchange. The Diffie and Hellman [1], RSA [2], and Elliptic-Curve cryptography [2] are examples of the asymmetrical encryption algorithms. These algorithms depend on two types of keys, secret and public keys. The client and server exchange a two authenticated keys used to generate the key used for the encryption and decryption of data. The authentication protocols have been used for authentication and key-exchange processes, such as EAP-TLS [3], EAP-TTLS [4], and PEAP [5]. This paper is organized as follows. Section 2 gives a short review of the

asymmetrical-encryption algorithms. Section 3 presents the enhancement in the Authentication Process. Section 4 presents the results. Finally, conclusion is presented in section 5. 2. Review on the Asymmetrical-Encryption Algorithms In the RSA algorithm, the authentication server announces two keys and keeps two keys to be secret keys. The client uses the two public keys to encrypt the message and the authentication server uses one of the secret key to decrypt the message. The RSA steps are as the following: Step 1: Authentication server chooses two very large numbers p and q. Step 2: It calculates qpn ×= Step 3: It calculate )1()1( −×−= qpφ Step 4: It chooses a random integer e, and then determines d from the relation φmod1=× ed (1) Step 5: The client encrypts the plain text, p, with the two keys, n and e. C=Pe mod n (2) Step 6: The authentication server decrypts C by using d and n to get P. P=Cd mod n (3) The drawback of RSA is that clients in this network can analysis this encrypted message because all the clients have the same public keys to encrypt the message. This gives the







35

man-in-the middle attacks to monitor the key-exchange process in the authentication process. In the Diffie and Hellman, the authentication server announces two keys, p and g, and uses a secret key, x. The client has a secret key, y. The authentication server encrypts its two public keys, p and g, by the secret key, x, see equation (4), then sends the encrypted message Ks to the client. The client will use its secret key, y, to get the shared key, K, see equation (5). The client will send an encryption message, Kc, to the authentication server. Kc can be obtained from equation (6). The authentication server will use its secret key, x, to get the shared key, K, see equation (7). The man-in- the middle attacks can easily exchange the data with the authentication server from the starting point. He can generate a secret key, y, and then obtain the shared key, K, from the previous steps. Ks=gx mod p (4) K=gxy mod p (5) Kc=gy mod p (6) K=gyx mod p (7) The proposed algorithm keeps the server and the keys stored in them to be secure from the outside attacks. That prevent unauthorized user to discover the shared key and the stored keys between them. Any authentication algorithm can use the proposed algorithm for authentication process and it will give the high security to the identification exchange between the authentication server and the client.

3. The Authentication Process Enhancement The authentication server and the clients in Wireless Local Area Network (WLAN) look for the best authentication protocol to keep their environment to be safe from any attacks. The public encryption algorithms are one of the methods used to safe WLAN environments. The explanation of the new key-exchange algorithm requires two processes, the key-distribution process and the authentication process. The key-distribution process is the initial process used at the initial configuration of the network, and it is used to distribute the network keys from the authentication server to all clients. The authentication process is the process used to authenticate the clients and the authentication server to each other using the generated keys in the key-distribution process before the data exchange process. 3.1 The Key-Distribution Process This process used to distribute the keys to all clients. The first steps of the key-distribution process are the same as RSA algorithm. The steps of the key-generation are as the following: Step 1: Authentication server chooses two very large numbers p and q. Step 2: It calculates qpnv ×= Step 3: It calculates )1()1( −×−= qpφ Step 4: It chooses a random integer number, e, and then

determine d from the relation φmod1=× ed (8) Step 5: It chooses a prime number, nk, to be a secret key in the authentication server. Step 6: It chooses a prime numbers, ns, for each client in WLAN. Step 7: It calculates shared key of the client, ks, from the following equations. pcnsnk ×= nknsqc ⊕= (9) nkqcpcks mod)( ×= Step 8: The authentication server will announce nv and e as public keys for all clients in WLAN, and it gives each client its two keys, ns and ks. Where ns is exchanged between the client and the server, and it is used to generate the ks used to encrypt the identities of the client and the server to complete the authentication process. 3.2 The Authentication Process The authentication process will appear as the following steps: Step 1: The Access Point (AP) sends a request packet to the client to start the authentication process. Step 2: The client encrypts ns with the two public keys, nv and e. C=nse mod nv (10.a) Step 3: The C-packet is passes to the authentication server through the AP Step 4: The authentication server decrypts C by using d and nv to get ns. ns=Cd mod nv (10.b) Step 5: The authentication server determines Ks of that client according to equations (9). The shared key, Ks, is kept out of the man-in-the middle attacks, because the flying shared-key, ns, cannot be analyzed except by the secret-key, n, stored in the authentication server. The outside attack can be appeared as:

• A client: in this case, the ks generated by ns in the authentication server will not be matched with ks of the client.

• The authentication server: in this case, the server cannot match the generated ks with the client key, ks. Each client has their shared keys that will save the clients in WLAN from the man-in-the middle attacks.

Step 6: The authentication server and the client exchange their identities, as description in the next sub-section.


36

3.3 Identities Exchange Stage The dual authentication between the client and the server is based on the exchange of the identities. After the key-exchange stage, the client encrypts its identities by using the shared key, Ks, to send them to the authentication server. But, the same key used to encrypt all identities packet can give the possibility to the outside attacks to discover the key based on the fixed shape of the packet such that used in EAP-methods. The solution of that problem is the key-updating with each packet. The key-updating adds many benefits such as: - Increase the impossibilities to the outside attackers to

discover the shared key. - No need to update the shared key with each authentication process. The shared-key is divided into bytes. Each byte of the shared key is applied to the S-Box. The S-box is a matrix of 1616× , and it is used to map the input code to another code at the output. The contents of the S-Box are shown in Table (1). For example, if some byte appears as 39 in a hexadecimal form at the sender, then the output of the S-Box of table (1) will be the code that takes the row number 3 and the column number 9, or the input byte, 39, is mapped into FF at the output. The encryption and/or the decryption of the identities are shown in the figure (1).

The feedback, kv, is the backbone of the key-updating as shown in figure (1). The encryption/decryption procedures are as the following: Step 1: KvKsKn ⊕= Where:

• Ks is the shared-key. • Kv is the feedback value, and its value is

zero at the first time Step 2: )(KnSKv = Where: S(Kn) is the output value of the S-Box when the input is Kn, it is applied on each byte inside Kn. Step 3: KvMM ⊕=* Where:

• M* is the encrypted/ decrypted identity • M is the plain/encrypted identity

respectively. 4. The Results The enhancement in the key-exchange improves the drawbacks of the Diffie Helman and RSA algorithms to exchange the keys during the authentication process. The shared key is exchanged in a safe environment. The attackers must know n which is stored in the authentication server to obtain the shared key. The enhancement in the identities-exchange during the authentication process keeps the WLAN environment to be safer than that used Diffie Helman and RSA:

i. The difficulties to discover the shared key. ii. The unauthorized client cannot get the shared key except by n which is stored in the authorized authentication server. iii. The unauthorized server cannot get the shared key except by n which is stored in the authorized authentication server. iv. The key-updating with each packet of the identities during the authentication process add more difficulties to the attacks to discover the shared key.

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

XOR XOR S-Box Ks Kn Kv

Kv

M

M*

Figure 1. The Encryption/Decryption of the Identities

Table 1: S-Box Contents


37

5. Conclusion

The authentication process is used to authenticate the clients and the authentication server in WLAN. The asymmetrical encryption algorithm is used for key-exchange in the authentication process. The enhancement of the key-exchange process increases the difficulties to discover the shared key. The key-updating with each Identity-packet adds more difficulties to crack the WLAN environment. The authentication process becomes more secure because of the key-exchange enhancement and the key-updating process.

References [1] F. Majstor, “WLAN security, threats and

solutions”,28th IEEE International Conference on Local Computer Networks, Bonn, Germany, Oct 2003.

[2] William Stallings, “Network Security Essentials (Applications and Standards)”, Pearson Education, 2004.

[3] Simon, D., Aboba, B., and R. Hurst, "The EAP-TLS Authentication Protocol", RFC 5216, March 2008.

[4] P. Funk and S. Blake-Wilson, " EAP Tunneled TLS Authentication Protocol Version 1 (EAP-TTLSv1) ", The Internet Society, Mar. 2006.

[5] Palekar, A., Simon, D., Zorn, G., Salowey, J., Zhou, H., and S. Josefsson, "Protected EAP Protocol (PEAP) Version 2", work in progress, October 2004.

[6] Aboba, B., Blunk, L., Vollbrecht, J., Carlson, J., and H. Levkowetz, "Extensible Authentication Protocol (EAP)", RFC 3748, June 2004.

[7] Simpson, W., "The Point-to-Point Protocol (PPP)", STD 51, RFC 1661, July 1994.

[8] Kasera and N. Narang, "3G Mobile Networks - Architecture, Protocols and Procedures", McGraw-Hill, 2004.

Authors Profile

Nagy Wadie Messiha received the B.S. in Electrical Engineering Telecommunication Department, Ein Shams University, Cairo, Egypt, June 1965, and M.S. in "Telecommunication Engineering", Helwan University, Cairo, Egypt, 1973, and the Ph.D in Computer Science from University of Stuttgart, in 1981. From 1981 to

1987, He was an associate professor in the department of communication engineering, Menoufia University, Menouf, Egypt. Currently, he is a Professor in the department of communication engineering, Menoufia University, Menouf, Egypt. He is interested in Communication systems, Electrical circuit Theory, Systems and Networks, Electronic Measurements, Computer Networks, and Information Theory and Coding.

Nabil M. A. Ayad received the B.S. in Electronics and Electrical Communications Department, Cairo University, Cairo, Egypt, June 1974, and M.S. in "A Microprocessor-Based Data Acquisition System for Exchanges", April 1979, and the Ph.D in “Performance Evaluation of Routing Techniques for Packet- Switched

Computer Networks”, Oct., 1984. From 1995 to 2002, He was an associate professor in the department of Reactors, Egyptian Atomic Energy Authority. Currently, he is a Professor in the department of Reactors, Egyptian Atomic Energy Authority. He is interested in Communication systems, Designing and implementing of PC business accounting packages Intelligent Database systems, Evaluation of LAN performance, and Modeling of computer networks.

Nawal El-Fishawy She received the PhD degree in mobile communications the faculty of Electronic Eng., Menoufia University, Menouf ,Egypt ,in collaboration with Southampton University in 1991 .Now she is the head of Computer Science and Engineering Dept., Faculty of Electronic Eng. Her research interest includes computer communication networks

with emphasis on protocol design, traffic modeling and performance evaluation of broadband networks and multiple access control protocols for wireless communications systems and networks. Now she directed her research interests to the developments of security over wireless communications networks (mobile communications, WLAN, Bluetooth), VOIP, and encryption algorithms.


38

An Approach to Apply Fuzzy Set in Classical Database

Sadia Husain1, Afshar Alam2 and Yasir Ahmad3

1Jazan University, Department of Computer Science, P.O box-706, Jizan, Kingdom of Saudi Arabia

[email protected]

2Jamia Hamdard University, Department of Computer Science & Information Technology Hamdard Nager, New Delhi-65 India

[email protected]

3Jazan University, Department of Computer Science, P.O box- 706, Jizan, Kingdom of Saudi Arabia [email protected]

Abstract: In this paper we present an application of fuzzy sets theory in a relational database with an objective of easily retrieval of meaningful information. This approach includes equivalence relation and equivalence class defined for the domain of some attribute. But a better approach is presented using Fuzzy attributes with fuzzy membership function with them. The degrees of membership in different terms can be evaluated by putting different values of attributes in membership function.

Keywords: Fuzzy attributes, Equivalence relation, Equivalence class, Fuzzy membership function, Fuzzy membership degree.

1. Introduction In recent years, in order to satisfy the requirement of easy information retrieval various new concepts are applied to the database. On such technique is to use fuzzy set with crisp database to retrieve meaning full data. Retrieving information from a large database is not easy. At the same time it is critical too as our decisions depend on them. By introducing equivalence classes and relation [1]-[4] we can reduce the complexity of the database but by using fuzzy sets [5]-[8] over the database attribute we can further simplify our approach to retrieve needed information and also the scope of our query is broaden. In this paper we first reduce the complexity of the database [9]-[10] by defining equivalence classes but at one point we will notice that attributes are fixed to one group even if they keep on changing their values. This problem will be solved by using fuzzy set over different attributes in the given database.

2. Preliminry We present some basic preliminaries for the better understanding of our work. Note that in [10] H.J Zimmerman has given very useful description about Equivalence relation and classes based on which some definition are presented in this section.

2.1 Equivalence Relation/Class

Normally, we consider values present in the domain of an attribute are different from another. Let say a domain of attribute have four values {good, excellent, average, poor} each of the values are different but we might consider two different values as indifferent in context of some query. In such cases we use equivalence relation. Further, for reducing complexity domain of the attributes on which Equivalence relation is defined can partition its values into equivalent values which can be called as equivalence classes. 2.2 Membership function

The membership function of a fuzzy set is a generalization of the indicator function in classical sets. In fuzzy logic, it represents the degree of truth as an extension of valuation. Degrees of truth are often confused with probabilities, although they are conceptually distinct, because fuzzy truth represents membership in vaguely defined sets, not likelihood of some event or condition. For any set X, a membership function on X is any function from X to the real unit interval [0, 1]. Figure 1 showing membership functions.

Figure 1. Memberships function

The Membership functions on X represent fuzzy subsets of X. The membership function which represents a fuzzy set is usually denoted by μx. For an element x of X, the value μx (x) is called the membership degree of x in the fuzzy set





39

.The membership degree μx(x) quantifies the grade of membership of the element x to the fuzzy set .The value 0 means that x is not a member of the fuzzy set; the value 1 means that x is fully a member of the fuzzy set. The values between 0 and 1 characterize fuzzy members, which belong to the fuzzy set only partially.

3. Application Here is an application that shows the use of fuzzy queries to fetch the required information from a large database. Table 1 shows the sample database.

Table 1 : Teachers Teacher Name Subjects Sectio

n Performance Irregularity

Paul Operating system

MCA-Sec A Average 10

Sameer Operating system

MCA-Sec B Good 7

Sameer Multimedia

MCA-Sec A Excellent 10

Meera Operating system

MCA-Sec C Average 9

Meera Multimedia

MCA-Sec B Average 5

Meera Database MCA-Sec A Poor 11

Raj Operating system

MCA-Sec D Average 10

Raj Multimedia

MCA-Sec C Good 6

Aryan Operating system

MCA-Sec E Excellent 4

Aryan Database MCA-Sec B Excellent 6

Laiba Operating system

MCA-Sec F Poor 9

Laiba Database MCA-Sec C Good 8

John Multimedia

MCA-Sec D Excellent 10

John Database MCA-Sec D Good 4

Different attribute of the given database are Teacher name, Subject, section assigned to each teacher, performance of each teacher in their subjects and finally irregularity attribute tells about the percentage of irregularity of the teacher with his/her classes. Class (performance) = {{Excellent, Good},{Average, Poor }} Now we can say values of Excellent and Good are equivalent to superior performance. Similarly Average and poor are equivalent to inferior performance. Assumption taken here:

a) One Teacher can teach more than one subject. b) One subject can be assigned to more than one teacher but for different sections i.e. suppose we have more than one section of the same class , example MCA Ist yr section A and section B. Now a subject say Operating system can be

assigned to teacher X for section A and Teacher Y for section B. For better understanding of the given application we first explain the use of equivalence relation/classes. Suppose a user only wish to know whether the performance is either superior or inferior. The domain of the attribute “performance” has four attribute. Dper ={ Excellent, Good, Average, Poor } Now we can obtain this by using equivalence relation, E given in table2.

Table 2: Equivalence Relation E Excellent Good Average Poor

Excellent 1 1 Good 1 1

Average 1 1 Poor 1 1

Here we have divided the domain of performance in two equivalence values or equivalence classes. So by introducing the concept of equivalence classes we are reducing the complexity of the data. Our next goal is to divide the teacher database in groups so, it will be easy to find teachers whose classes are not doing well and necessary measures should be taken to make them fit for the organization. Let’s define the domain of Performance and Irregularity: Dper ={Excellent, Good, Average, Poor } Diir={ 1, 12 } We can define the query with the following perspective- Ø Excellent and Good as superior performance. Ø Average and Poor as inferior performance.

Gper={{ Excellent ,Good },{ Average , Poor }} Similarly, Ø Irregularity {1, 7} within this range is acceptable. Ø Irregularity {8, 14} within this range is

unacceptable. Girr= {{1, 7}, {8, 14}} Based on above perspective we can define our database in four groups shown in table 3.

Table 3: Group

Teachers Performance Irregularity {Aryan , Sameer ,Raj , John}

{ Excellent , Good}

{4,6,7}

{ Sameer , John , Laiba}

Excellent ,Good

{8,10}

{Meera} Average {5}

{Paul ,Raj ,Meera, Laiba}

Average , Poor

{9,10,11}

We can deduce the group in the matrix given in figure 2.

G1

G2

G3

G4


40

Figure 2. Matrix1 So from the deduced matrix, it’s easy to find what action should be taken against each teacher depending on the group they fall. Problem with this matrix is that even if the teacher improves the performance and reduces the irregularities up to some extent, but still he/she remains in the same class. To solve this problem we can use Fuzzy sets with the attributes performance and irregularities. Now we will define these attributes with the Fuzzy membership function. So in our case we can define fuzzy membership function with attribute Irregularities as acceptable and unacceptable.

μ Acc (u)= {

μ Unacc (u) = {

Now let’s define fuzzy membership function with attribute Performance as Superior performance and Inferior performance.

μ sup= { (Excellent, 1) , (Good , .65), (Average, .35) }

μ inferior= {(Good, .35),( Average, .65),(Poor, 1) }

The following example will show the use of above mention membership function and we further deduce the group into matrix2 given in figure 3

14

13

12

11

10

9

UNACCEPTABLE

8

G2

Ask to reduce the irregularities

G4

Word of warning to teacher.

7

6 5 4 3 2

ACCEPTABLE

1

G1

Give Appraisal.

G3

Ask to improve Performance.

Excellent

Good Average Poor

0.00 ------------------------------------------------ 0.33 ---------- 0.67 ---------- 1.00 ------------------- ------------------ Figure 3. Matrix 2 From figure 3 we can easily evaluate performance of a teacher. For example teacher Mr Paul in the database, would give performance which is .33 degree superior and inferior to the degree .67. His Irregularities are accepted to the degree 0 and unaccepted to the degree 1.This data indicate he really needs to improve. Similarly we can evaluate performance and irregularities for Teacher Mr John who is teaching Database to section D. Mr John would give performance which is .65 degree superior and inferior to the degree .35. His Irregularities are accepted to the degree 1 and unaccepted to the degree 0. So, the degree of membership for different terms can be easily determined by substituting the values of the attributes in membership function.

4. Conclusion This paper deals with the application of fuzzy set basically fuzzy queries in crisp database with an objective to reduce the complexities of large database. By defining suitable membership function on the different attributes, we can find

14

13 12 11 10 9

UNACCEPTABLE

8

G2

Ask to reduce the irregularities

G4

Word of warning to teacher.

7 6

5 4 3 2

ACCEPTABLE

1

G1

Give Appraisal.

G3

Ask to improve Performance.

Excellent Good Average Poor Superior performance

Inferior performance

0 for 9 ≥ u

(9-u) / (9-5) for 5 ≤ u<9

1 for 1 ≤ u<5

(u-5) / (9-5) for 5 ≤ u<9

1 for 9 ≤ u

0 for u<5

μ Superior Inferior


41

the appropriate degree of acceptance / unacceptable for a given value of an attribute. It has been shown that how the use of equivalence the complexity of the large database when the query is expressed in some context. A problem was encountered that attributes are confined in the same class even if they improve themselves. To overcome this problem we use fuzzy sets over the attributes and define proper membership function for each of them. And finally, we check membership function with different attributes value. So we concluded that degrees of the membership in the different terms can easily be determined by substituting the values of the attributes in the membership functions.

References [1] Sanchez, E., “Resolution of composite fuzzy relation

Equations”, Inform. And control., 30(1996) 38-48.

[2] K. C. Gupta, R. K. Gupta “Fuzzy equivalence relation redefined “, Fuzzy Sets and Systems, Volume 79 Issue 2 (1996) 227-233.

[3] Miroslav Ćirić, Jelena Ignjatović, Stojan Bogdanović, “Fuzzy equivalence relations and their equivalence classes”, Fuzzy Sets Volume 158 , Issue 12 (June 2007), 1295-1313.

[4] J. Hale, Sujeet Shenoi, "Catalyzing database inference

with fuzzy relations," isuma, pp.408, 3rd International Symposium on Uncertainty Modelling and Analysis, 1995-pp 408.

[5] ZADEH, L. “Fuzzy sets as a basis for a theory of possibility”, Fuzzy Sets and Systems 1 (1978), 3-28.

[6] ZADEH, L. “Fuzzy sets”, Information and Control 8 (1965) 338-356.

[7] D. Dubois, H Prade, Fuzzy Sets and Systems:Theory and Application, Academic Press, New York,1980.

[8] Jose Galindo/ Mario Piattni/ A.Urrutia :Fuzzy database Modeling Design & Implementation: Idea group publishing, USA,2005.

[9] Ullman, J.D, principal Of database Systems, Galgotia publication India, 1998.

[10] H.J Zimmerman, Fuzzy sets Theory And its Applications, Fourth edition, Kluwer Academic Publishers, Boston, USA, 1991.


42

MOSAODV: Solution to Secure AODV against Blackhole Attack

N. H. Mistry1, D. C. Jinwala2 and M. A. Zaveri3

1Computer/IT Engineering Department, Shri S’ad Vidya Mandal Institute of Technology, Bharuch 392-001, Gujarat, India

[email protected] 2,3 Computer Engineering Department, Sardar Vallabh National Institute of Technology ,

Surat 395-007, Gujarat, India [email protected], [email protected]

Abstract: Mobile ad hoc network (popularly known as MANET) showing promising applications have now gained significant importance in research as well as in practice, due to their autonomous and self-maintaining nature. Unlike other types of networks, MANETs are usually deployed without a centralized control unit. Hence, mutual cooperation amongst the participating entities forms the basis for determining the routes to the destination. This aspect along with the fact that MANET nodes are often constrained in power, storage and computational resources, make MANETs vulnerable to various communications security related attacks. Therefore, the direct application of the conventional routing algorithms is infeasible here. Numerous attempts can be found in the literature that concentrates on improving the security of the routing protocols for MANETS. However, according to our analysis, none of them is complete by itself. In this paper, therefore, we focus on improving the Secure Adhoc On demand Distance Vector (AODV) routing protocol to safeguard it against a Denial of Service attack viz. the Blackhole attack. The proposed modifications to AODV are implemented and tested using Network Simulator (NS-2.33). The performance analysis carried out shows improvement in Packet Delivery Ratio of AODV in presence of Blackhole attack, with marginal rise in average end-to-end delay and normalized routing overhead.

Keywords: MANET, Blackhole attack, Security, Routing Protocols, AODV.

1. Introduction The desire to be connected anytime, anywhere, anyhow has led to the development of wireless networks, with a focus on pervasive and ubiquitous computing. MANETs are no exception [1]. Therefore, traditional wired routing techniques are infeasible here [2]. Due to the open medium, dynamically changing network topology, cooperative algorithms, lack of centralized monitoring and management, and lack of a clear line of defense; MANETS are more vulnerable to attacks than wired networks [3]. Amongst various attacks that the MANETs are susceptible to, a few are eavesdropping by the adversary, spoofing on the control and data packets transacted, malicious modification or alteration of the packet content and several Denial-of-service (DoS) attacks.

In addition, in the absence of any centralized mechanism to support the network operations, the participating nodes in a MANET rely largely on cooperative algorithms establishing the network routes. Hence, the routing protocol obviously becomes susceptible to the nodes with malicious intent. We focus on analyzing the security of the Adhoc On-

Demand Distance Vector (AODV) that is one of the many available reactive routing protocols for MANETs AODV is a reactive routing protocol for adhoc and mobile networks. As all other routing protocols of MANETs, AODV uses two phases, viz., Route Discovery and Route Maintenance. Various control messages used by AODV are Route Request (RREQ), Route Reply (RREP) and Route Error (RERR). The header information of this control messages can be seen in detail in [6]. Every node in an Adhoc network maintains a routing table, which contains information about the route to a particular destination. Route Discovery Phase is initiated by broadcasting RREQ message. After broadcasting RREQ the source node waits for the RREP message. If a route is not received within NET_TRAVERSAL_TIME milliseconds, the node may try again to discover a route by broadcasting another RREQ, up to a maximum of RREQ_RETRIES times at the maximum TTL (Time to Live) value [6]. ReveiveRREP(Packet p) is one of the crucial function of AODV. The pseudocode can be seen in figure 1.

Figure 1. RecvReply Pseudocode

The nodes participating in the communication can be classified as either source node, intermediate node or destination node. Working of a node varies as it plays one of these roles. Source node, once send a RREQ waits for first RREP to come, the figure hence explains only the part at source node after it receives a RREP message. When destination node or intermediate node that has fresh enough route to the destination receives the RREQ message it generates an RREP message and updates its routing table with accumulated hop count and the sequence number of the destination node. Freshness of a route is decided by the magnitude of sequence number. Sequence number is a 32-bit integer. The larger the sequence number, the fresher is the route considered [4]. In Route Maintenance, if a node

At Source Node: AODV 1 ReceiveRREP (Packet P){ 2 if(p has an entry in Route Table) { 3 select Dest_Seq_No from routing table 4 if(P.Dest_Seq_No > Dest_Seq_No){ 5 update entry of P in routing table 6 unicast data packets to the route specified in RREP 7 } 8 else { 9 discard RREP 10 } 11 } 13 else { 14 if(P.Dest_Seq_No >= Src_Seq_No) { 15 Make entry of P in routing table 16 } 17 else { 18 discard this RREP 19 }





43

finds a link break or failure then it sends RERR message to all the nodes that uses the route.

In this paper, therefore, we propose an algorithm to counter Blackhole attack against Secure AODV routing protocol [5] in MANETs. As our results and analysis described in section 5, the proposed modification to Secure AODV is indeed effective in preventing the Blackhole attacks with marginal performance penalty.

The rest of the paper is organized as follows: In Section 2, we describe working of AODV routing protocol, the Blackhole attack and then survey of the related work in the area. In section 3, we discuss our solution MOSAODV (MOdified Secure AODV) algorithm. In section 4, we discuss the methodology of evaluating the MOSAODV and will discuss the metrics used to compare the algorithm relative to the existing traditional AODV. In Section 5, we describe the simulation results and analyze the same. Finally we conclude in Section 6.

2. Theoretical Background and Related Work Blackhole attack is one of the active DoS attacks possible in MANETs. In blackhole attack a malicious node sends a forged RREP packet to a source node that initiates the route discovery in order to pretend to be a destination node itself or a node immediate neighbor of the destination. So, source node will forward all of its data packets to the malicious node; which were intended for the destination. The malicious node will never forward these data packets to the destination and therefore, source and destination nodes became unable to communicate with each other [7].

Figure 2 illustrates blackhole attack in MANETs. In the figure node S wants to communicate with node D. It can be seen in the figure that S starts broadcasting RREQ, which is received by nodes N1, N2, N3 and M (malicious node). Node N1 and N3 again forwarded the RREQ to node D. Assuming RREQ from node N1 reaches first to D, so it generates RREP and replied back to N1. Then it receives same RREQ from node N3 and hence is dropped (ignored) by D. Node M being malicious also generates RREP and sends it to node S. So, now node S will ignore the genuine RREP from N1 (Destination Sequence Number in RREP from M is higher). Node S will now starts sending data packets to node M. Node M being malicious absorb all data packets.

Figure 2. Blackhole Attack in MANET

First, we shall explore how a malicious node succeeds in injecting blackhole attack to a MANET using AODV as its routing protocol. Security of AODV is compromised as it accepts the received RREP having fresher route. The malicious node always sends RREP as soon as it receives RREQ without performing standard AODV operations keeping Destination Sequence number very high. As AODV

considers RREP having higher value of destination sequence number to be fresher, the RREP sent by malicious node is treated fresh. Thus, malicious node succeeds in injecting blackhole attack. According to the solution in [8] the requesting node without sending the DATA packets to the reply node at once, it has to wait till other replies with next hop details from the other neighboring nodes for a predetermined time value. After the timeout value, it first checks in CRRT table whether there is any repeated next hop node. If any repeated next hop node is present in the reply paths it assumes the paths are correct or the chance of malicious paths is limited. The solution adds a delay and the process of finding repeated next hop is an extra addition to overhead.

In [9], DPRAODV check to find whether the RREP_seq_no is higher than the threshold value. The threshold value is dynamically updated at every time interval. As the value of RREP_seq_no is found to be higher than the threshold value, the node is suspected to be malicious and it adds the node to the black list. As the node detected an anomaly, it sends a new control packet, ALARM to its neighbors. The ALARM packet has the black list node as a parameter so that, the neighboring nodes know that RREP packet from the node is to be discarded. Further, if any node receives the RREP packet, it looks over the list, if the reply is from the blacklisted node; no processing is done for the same. In DPRAODV, one can imagine overhead of updating threshold value at every time interval. Along with this, generation of ALARM packet will considerably increase Routing Overhead.

In [10], the protocol requires the intermediate nodes to send RREP message with next hop information. When the source node get this information it will send a RREQ to the next hop to verify that the node has a route to the intermediate node that sends back the RREP packet, and that it has a route to the destination. When the next hop receives Further Request, it sends Further Reply which includes check result to source node. Based on information in Further Reply, the source node judges the validity of the route.

In [11], source node verifies the authenticity of node that initiates RREP by finding more than one route to the destination. When source node receives RREPs, if routes to destination shared hops, source node can recognize the safe route to destination.All solutions discussed [9] [10] [11], puts some overhead on either/both intermediate and destination nodes in one or other way. Keeping in mind, the limitations of mobile nodes in MANETs (battery life, processing power, storage) we need to device an algorithm or protocol that satisfies the following criterions: • The algorithm should put minimum routing overhead and

end-to-end delay. • It should put minimum efforts on either intermediate or

destination node. Otherwise, sometimes intermediate nodes tend to act selfishly.

• The selection procedure (of a fresh route) must be computationally simple.


44

3. Modified Secure AODV (MOSAODV) Though there are many solutions available to deal with blackhole attack. All of them are incomplete in one or other way. As discussed in section 2 d, there is still scope for a solution that takes care of all the points discussed there. Our solution is designed such that it does not modify working of either intermediate nodes or of destination. In addition to normal AODV MOSAODV have a new table Cmg_RREP_TAB, a timer MOS_WAIT_TIME and a variable Mali_node. The structure of the table can be seen in table 1.

Table 1: Fields in Cmg_RREP_Tab Table Field Name Size in

Bytes Value

Node Id 2

Node Id from where RREP arrived

Destination Sequence Number (Destination Seq. No.)

4 Value of destination sequence number in the RREP.

Unlike AODV, source node in MOSAODV does not accept every first RREP but calls Pre_ReceiveRREP(Packet p) which stores all the RREPs in the newly created (Cmg_RREP_Tab) table till MOS_WAIT_TIME. Then it analyses all the stored RREPs from Cmg_RREP_Tab table, and discards the RREP having exceptionally high destination sequence number. The node that sent this RREP is suspected to be the malicious node. MOSAODV maintains the identity of the malicious node as Mali_node so that in future it can discard any RREPs from that node. Now since malicious node is identified the routing table for that node is not maintained and also control messages from the malicious node will not be forwarded in the network. Cmg_RREP_Tab is flushed once an RREP is chosen from it. Our solution; after detecting the malicious node acts as normal AODV by accepting the RREP with higher destination sequence number. The pseudocode of MOSAODV is given in figure 1. Line number 14 shows that after selecting one RREP, MOSAODV calls ReceiveRREP(Packet p) method of AODV.

Figure3. Pseudocode of MOSAODV Figure 4 explains working of MOSAODV. It is assumed

that RREP from M is received at time t0. Figure 4(a) show that unlike traditional AODV, MOSAODV rather than dropping the second RREP coming from N1to S is saved in Cmg_RREP_Tab table. Figure 4(b) shows the entries in the Cmg_RREP_Tab table of S. figure 4(c) shows a scenario at T1= T0 + MOS_WAIT_TIME, MOSAODV node S picks RREP from node N1 to be used. And now, the flow of data

packets is from S to D via node N1. MOSAODV is very simple and the algorithm will fail only if there exists a single route to the destination and that route is compromised. Our solution adds a table of size 6 byte ( Table 1), a variable Mali_Node of size 2 bytes and a timer variable of size 10 bytes. The overall memory consumption is 20 Bytes more than that of AODV. This is worthy for the rise in Packet deliver Ratio (PDR). The time overhead in MOSAODV is MOS_WAIT_TIME which is a constant value in terms of milliseconds (1500 ms) and time required to execute Pre_ReceiveRREP() is also in terms of milliseconds. So again that is acceptable.

Figure 4 .

(a) MOSAODV at T1 > T0 : Saving RREPs in cmg_RREP_Tab (b) Entries in Cmg_RREP_Tab at time T0+MOS_WAIT_TIME (c) MOSAODV at T2 > = T0 + MOS_WAIT_TIME

4. Simulation Results For the simulations, we use NS-2 (v-2.33) network simulator. NS-2 provides faithful implementations of the different network protocols. At the physical and data link layer, we used the IEEE 802.11 algorithm. The channel used is Wireless Channel with Two Ray Ground radio propagation model. At the network layer, we used the routing algorithms AODV and MYSAODV. Finally UDP is used at the transport layer. All the data packets are CBR (continuous bit rate) packets. The details of CBR packets can be seen in table 3.

The connection pattern is generated using cbrgen and the mobility model is generated by setdest. Setdest generates the random positions of the nodes in the network and mobility in the network. The terrain area is 800m X 800m with number of nodes varying from minimum 10 to maximum 80 with chosen maximum speed up to from 10 m/s to 70 m/s and pause time varying from 1s to 5s. The simulation parameters are summarized in table 2.

Each data point represents an average of ten runs. The same connection pattern and mobility model is used in simulations to maintain the uniformity across the protocols. Table2: Simulation Parameters Table 3: Details of CBR

At Source Node: MOSAODV 1 Pre_ReceiveRREP (Packet P){ 2 t0 = get(current time value) 3 settimer(to + MOS_WAIT_TIME) 4 till timer expires Store P.Dest_Seq_No and P.NODE_ID in Cmg_RREP_Tab table 5 after timer expires 6 while (Cmg_RREP_Tab is not empty) { 7 Select Dest_Seq_No from table 8 if (Dest_Seq_No >>>= Src_Seq_No){ 9 Mali_Node=Node_Id 10 discard entry from table 11 } 12 } 13 select Packet q for Node_Id having highest value of Dest_Seq_No 14 ReceiveRREP(Packet q)


45

To analyze the performance of MYSAODV, various contexts are created by varying the number of nodes, nodes mobility and nodes pause time. The metrics used to evaluate the performance of these contexts are given below. Packet Delivery Ratio: The ratio between the number of packets originated by the “application layer” CBR sources and the number of packets received by the CBR sink at the final destination. Average End-to-End Delay: This is the average delay between the sending of the data packet by the CBR source and its receipt at the corresponding CBR receiver. This includes all the delays caused during route acquisition, buffering and processing at intermediate nodes, retransmission delays at the MAC layer, etc. It is measured in milliseconds. Normalized routing overhead: This is the ratio of number of control packets to data transmissions in a simulation. A transmission is one node either sending or forwarding a packet. Either way, the routing load per unit data successfully delivered to the destination [8].

To evaluate the packet delivery ratio, End-to-End Delay and Normalized Routing Overhead; simulation is done with nodes with the source node transmitting maximum 1000 packets to the destination node. Figure 5 shows the graphs when network size (number of nodes) is varying. It can be seen from the figure5 (a), that PDR of AODV drops by 81.812 % in presence of blackhole attack. The same increases by 81.811 % when MOSAODV is used in presence of blackhole attack. At the same time, figure5 (b) and figure5 (c) shows that the rise in End-to-End delay and Normalized Routing overhead is 13.28 % and 15.05% respectively. Figure 6 shows the graphs when mobility of nodes is varying. It can be seen from the figure 6 (a), that PDR of AODV drops by 70.867 % in presence of blackhole attack. The same increases by 70.877 % when MOSAODV is used in presence of the attack. At the same time, figure 6 (b) and figure6 (c) shows that the rise in End-to-End delay and Normalized Routing overhead is 6.28 % and 7.81 % respectively. Figure 7 shows the graphs when pause time of

nodes is varying. It can be seen from the figure 7(a), that PDR of AODV drops by 92.84 % in presence of blackhole attack. The same is gained back when MOSAODV is used in presence of the attack. At the same time, figure 7 (b) shows a drop by 0.55 % in End-to-End Delay. Figure 7 (c) shows that the rise in Normalized Routing overhead is 7.71 %. This is acceptable.

5. Conclusion The algorithm presented in this paper provides protection against blackhole attack in MANET. Inclusion of MOS_WAIT_TIME variable and Cmg_RREP_Tab table, helps us to suspect malicious node. From the experimental results, it shows that the solution achieves a very good rise in PDR with acceptable rise in End-to-End delay and Normalized Routing Overhead. Neither intermediate nodes nor the destination node need to do anything extra. As compared to the various solutions; we had seen in the paper the algorithm is simple to implement. Though the algorithm is implemented and simulated with AODV routing algorithm, we believe that the solution can also be used by other routing algorithms as well.

References [1] Anil Kumar Verma, “Design And Development Of A

Routing Protocol For Mobile Ad Hoc Networks (Manets)”, A Thesis Of Doctor Of Philosophy In Computer Science And Engineering, Thapar Universtiy Patiala .

[2] Ebrahim Mohamad, Louis Dargin;” Routing Protocols Security In Ad Hoc Networks”. A Thesis Oakland University School of Computer Science and Engineering.

[3] Yian Huang, Wenke Lee; “A Cooperative Intrusion Detection system for Ad Hoc Networks”. In Proceedings of the 1st ACM Workshop Security of Ad Hoc and Sensor Networks, Fairfax, Virginia, pp135-147, 2003.

[4] N.H.Mistry, D.C.Jinwals, M.A.Zaveri;” Prevention of Blackhole Attack in MANETs”. In Proceedings of EPWIE-2009, Gujarat, India, pp 89-94, July 2009.

[5] M.G.Zapata; “Secure On Demand Distance Vector (SAODV) Routing”. INTERNET-DRAFT draft-guerrero-manet-saodv-06.txt, Sep. 2006

[6] C. Perkins, “(RFC) request for Comments-3561”, Category:Experimental, Network, Working Group, July 2003.

[7] Satoshi kurosawal, Hidehisa, Nakayama, Nei Kato, Abbas Jamalipour and Yoshiaki Nemoto. “Detecting Blackhole Attack on AODV-based Mobile Ad Hoc Networks by Dynamic Learning Method”, International Journal of Network Security, Vol.5, No.3, pp.338–346, Nov. 2007.

[8] Latha Tamilselvan, V Sankaranarayanan, “Prevention of Blackhole Attack in MANET”. In Proceedings of The 2nd International Conference on Wireless Broadband and Ultra Wideband Communications (AusWireless 2007), pp. 21-21, Aug. 2007.

[9] Payal N. Raj, Prashant B. Swadas. “DPRAODV: A Dyanamic Learning System Against Blackhole Attack In Aodv Based Manet”, International Journal of Computer Science Issues, Vol. 2,pp 54-59,2009.

[10] H. Deng, W. Li, and D. P. Agrawal, “Routing security in ad hoc networks”, IEEE Communications Magazine, vol. 40, no. 10, pp. 70-75, Oct. 2002.

[11] M. A. Shurman, S. M. Yoo, and S. Park, “Black hole attack in wireless ad hoc networks”, in ACM 42nd Southeast Conference (ACMSE’04), pp 96-97, Apr. 2004.


46

Exploring spatial information in spectral features for Texture image retrieval

1Renuka Methre and 2M. Ravindranath

Research Scholar

University of Hyderabad Email: [email protected], [email protected]

Abstract: Texture analysis plays an vital role in many areas such as remote sensing, medical imaging, content based image retrieval. The development in multi-resolution analysis such as Gabor and wavelet transform have helped for good results in texture image retrieval. This paper investigates the texture retrieval using combination of local features of Haralick derived from one level discrete wavelet transform coefficients and global statistical features computed form three level wavelet transformed images. Further retrieval performance is considerably increased by updating query images features through human relevance feedback using Query Vector Movement method.

Keywords: Texture image retrieval.

1. Introduction

Texture is an important feature to characterize the region of an image. Texture in an image can be perceived at different scales or levels of resolution. Texture is defined in [14], as “A region in an image has a constant texture if a set of local statistics or other local properties of the picture are constant, slowly varying, or approximately periodic”. Image texture, defined as a function of the spatial variation in pixel intensities (gray values), is useful in a variety of applications and has been a subject of intense study by many researchers. One immediate application of image texture is the recognition of image regions using texture properties.

Texture analysis approaches is broadly classified under statistical, spectral, structural and stochastic approaches. With a study on human vision system, researchers begin to develop the multi-resolution texture analysis models, such as the wavelet transform and the Gabor transform. Extensive research has demonstrated that these approaches based on the multi-resolution analysis could achieve reasonably good performance, so that they are widely applied to texture analysis, classification, and retrieval. Many other statistical methods have been proposed to extract texture features. They are Markov Random Fields [5][4], Gibbs Random Fields [7], Entropy based applications [3]. Haralick features capture the spatial correlations in gray level between adjacent pixels by calculating statistics based on the grey level co-occurrence matrix. Ohanian and Dubes [13] studied the performance of four types of features: Markov Random Fields parameters, Gabor multi- channel features, fractal-based features and co-occurrence features. Arivazhagan et'al[1] used the statistics (mean and variance) extracted

from the wavelet subbands as the texture representation. To explore the middle-band characteristics, tree-structured wavelet transform was used. Landeweerd and Gelsema [11] extracted various first-order statistics (such as mean gray level in a region) as well as second-order statistics (such as gray level co-occurrence matrices) to differentiate different types of white blood cells. Insana et al., [9] used textural features in ultrasound images to estimate tissue scattering parameters. Chen et al., [15] used fractal texture features to classify ultrasound images of livers, and used the fractal texture features to do edge enhancement in chest X-rays. In [13] the comparison of different features obtained by fourier spectrum, co-occurrence statistics, run length statistics, second order grey level statistics.

The most common multi-resolution analysis approach is to transform a texture image into a local spatial/frequency representation by convolving the image with a bank of filters with some tuned parameters. This clearly motivates researchers to study how to extract more discriminable texture feature based on the multi-resolution techniques. Compared to the wavelet transform, the Gabor transform needs to select the filter parameters according to different texture. In order to strengthen these multi resolution based features, we explored spatial relationship in coefficients of multi resolution transformed images.

The main aim of this paper is to investigate fusion of global and local texture features extracted from discrete wavelet transform to improve retrieval performance. Further human feedback is also considered to improve query features using Query Vector Movement (QVM). This paper is organized as follows. In Section II theory of DWT is discussed. In section III features used in this paper are presented. Section IV gives experimental results. Finally, Section V presents concluding remarks of our work.

2. Discrete Wavelet Transform The image is decomposed into sub-bands using Discrete

Wavelet Transform (DWT) and subbands are labelled as LL, LH, HL and HH where LL corresponds to coarse approximation and remaining represent the finest scale wavelet coefficients of original image. On Sub-band LL alone being further decomposed and sampled results in two-level wavelet decomposition and thus detail and approximation features are obtained as shown in Fig.1. The size of the subband images is halved on each decomposition. The values in sub-band images or their combinations or the




47

derived features from these bands uniquely characterize a texture.

Figure1. A three level wavelet decomposition schema of

an image

3. Features extraction

3.1 Global features The texture images in database are decomposed using

DWT. Mean and Standard Deviation of approximation and detail sub-bands of three level decomposed images (ie, LLk, HLk and HHk; for k=1, 2, 3) are calculated as features and stored in features library.

∑N

=ji,

j)x(i,N

=(m)1

2

1mean

where x is the DWT transformed image for any sub-band

of size N×N.

3.2 Local features Gray Level Co-occurrence Matrices are computed for 1-

level DWT decomposed detail subband images. The co-occurrence features such as contrast, energy, entropy, local homogeneity, cluster shade, cluster prominence and maximum probability are calculated from the co-occurrence matrix C(i,j) for different angles

( θ= 00 ,450 ,900 ,1350) using the formulas given

below.

∑N

=ji,x j)iC(i,=M

1

∑N

=ji,y j)jC(i,=M

1

4. Texture Retrieval Query image feature vector is compared against all the

feature vectors in features library using the Euclidean distance

[ ] 21

esNooffeatur

1

2

−∑

=jqx (j)f(j)f=q)D(x, where f q ,

f x represents the feature vector of query and database

image respectively. The top few images are retrieved based

on the minimum Euclidean distance.

Relevance feedback (RF) is a feature of information retrieval systems. In relevance feedback the results that are initially returned for a given query are used to update the query vector for improved retrieval results. Some of the techniques for RF are QVM, Feature relevance estimation (FRE), Bayesian Inference (BI).We have used the classical Rocchio Method of query vector Movement . The feature vector q (k+1) is updated from the kth query using the following formula

where DR are the set of relevant images, NR is the number of relevant(similar) images. Here similar images mean the images from same class.

5. Experimental results We have implemented our work using Matlab7. The

image database used is Brodatz database of 1856 images of 116 classes. Each class has 16 similar images. The sample Brodatz image library is shown in Fig.2.. Our feature database is created by computing the feature vector each of length 136 for each image 128x128 original texture images by extracting GLCM features such as contrast, energy, entropy, local homogeneity, cluster shade, cluster prominence and maximum

Figure 2. Sample Brodatz database having 116 classes of 16 similar images for each class

∑ −N

j=i,

m]j)[x(i,N

=)(1

22

1sddeviation standard

∑∑−i j

j)C(i,j)C(i,= logEntropy

∑∑ −i j

j)C(i,j)(i= 2Contrast

Cluster prominence = ∑i,j=1

N

i− Mx+j− My 4C i,j

∑ −−N

j=i,yx j)C(i,)Mj+M(i=

1

3shadeCluster ∑∑i j

j)(i,C= 2Energyj)][C(i,= Maxyprobabilit Maximum

∑Rjεε

jR

k)+(k DN

+q=q 11

∑ −ij)C(i,

j)(i+= 21

1y homogeneit Local


48

probability derived from co-occurrence matrices computed for different angles (i.e. θ= 00 ,450 ,900 ,1350

). Feature vector is formed by mean and standard deviation of 3 level DWT subband images i.e ( 24 features = 4 means x 3 levels + 4 variance x 3 levels) . 7 Haralick features calculated from occurrence matrix along four directions from original image i.e. (24 features = 7 features x 4 directions) and the same features calculated from l level decomposed detailed DWT image i.e. (84 features = 7 features x 4 directions x 3 detail images) . This results in vector length of 136 for each image. Each feature vector is normalised in range [0, 1] by following formula.

−

−(x)(x)

(x)x=xminmax

min

Experiment I: The fig. 3 shows the sample query and the top ranked 20 retrieved images using our proposed method. The first image in the list is the query image, followed by the retrieved images. The average retrieval performance for 116 different classes is shown as plot in fig. 5

Experiment II: The fig.4 shows the retrieval performance comparison using 3 level DWT features and Haralick features separately. It is observed that Haralick features perform well in retrieving Brodatz images. The use of spatial information in spectral coefficients is found to clearly outperform in texture characterization. Using both features average retrival performance is 14.56 out of 16 i.e. (90.56%) for 116 classes using the best retrieval performance in each classes as shown in Table 1.

Table 1: the average retrieval performance comparison using different features.

Features DWT GLCM DWT & GLCM

After QVM

performance(%) 81.5 90.4 90.56 91.38

Experiment III: RF is captured using QVM method to further improve the retrieval performance. We have done one itertation of RF. Fig. 5 shows comparative retrieval performance plots for 116 different classes. The Fig. 6 shows average precision versus recall curve.

The precision and recall are defined as

Figure 3. Retrieval of top 20 images for query from T75

class which is at top left

Figure 4. Average retrieval performance of 116 classes using DWT, GLCM and both features

Figure 5. Average retrieval performance of 116 classes

using DWT, GLCM features using QVM and without QVM

retrieved images of no totalretrieved imagesrelevant of noPrecision =

database from retrieved images of no totalretrieved imagesrelevant of noRecall =


49

Figure 6. Precision Vs. Recall curve

6. Conclusions We have explored the statistical approach by finding

Gray level co-occurrence matrix features to capture spatial information from multi resolution approach discrete wavelet transform coefficients. Our experimental results show that considerable improvement in retrieval efficiency by using the proposed methodology. The average retrieval performance achieved is 14.6 out of top 20 images being retrieved.

References [1] Arivazhagan S, Ganesan L. “Texture Classification

using wavelet transforms.” Pattern Recognition Letters 24, Pages (PP) 1513-1521, 2003

[2] Brodatz., P., “Textures: A photo graphic Album for Artists and Designers”, New York : Dover , New York 1996.

[3] Coifman, R.R Wickerhauser , M.A., Entropy based algorithm for best basis selection”, IEEE Trans., Information Theory, 38,1992,204-222.

[4] Cohen F.S., Fan, Z. Patel, M.A., Classification of rotated and scaled textured images using Gaussian Markov random Fields models”, IEEE TRANS. Pattern Anal. Machine Intell., 13(2), 1991,192-202.

[5] Chellapa, R., Chatterjee, S. “Classification of texture using Gaussian Markov random Fields”, IEEE Trans. Acoustics Speech Signal Process. ASSP-33(4),959-963.

[6] Conners, Harlow R.W, “A theoretical comparison of texture algorithms”, IEEE Transactions on Pattern Analyses and Machine Intelligence. Vol. PAMI-2, Pages (PP) 204-222. May 1980.

[7] Derin, H., Elliot, H., “Modeling and segmentation of noisy and textured images using Gibbs random fields”, IEEE Trans. Pattern Anal. Machine Intell., PAMI – 9(1), 1987, 39-55.

[8] Haralick R.M, Shanmugam.K and Dinstein I, “Textural features for image classification”, IEEE Transactions on System, Man, Cybernetics, Pages (PP):610-621, 1973.

[9] Insana, M. F., R. F. Wagner, B. S. Garra, D. G. Brown, and T. H. Shawker, “Analysis of Ultrasound Image Texture via Generalized Rician Statistics,” Optical Engineering, 25,Pages(PP)743-748, 1986.

[10] Jain, A.K. and Farrokhnia, F. “Unsupervised Texture Segmentation Using Gabor Filters”, Pattern Recognition, Vol. 24, No. 12, Pages (PP) 1167-1186, 1991.

[11] Landeweerd, G. H. and E. S. Gelsema, “The Use of Nuclear Texture Parameters in the Automatic Analysis of Leukocytes,” Pattern Recognition, Vol 10, Pages (PP) 57-61, 1978.

[12] Manjunath B.S and W.Y. Ma. “Texture features for browsing and retrieval of image data.”, IEEE Transactions on Pattern Analysis and Machine Vol. 18, No. 8, Pages(PP) 837-42, Aug. 1996.

[13] P.Oanian und R.Dubes. Performance evaluation for four classes of textural features” Pattern Recognition ,25: 819-8133,1992.

[14] Sutton, R.N. Hall, E.L. “Texture Measures for Automatic Classification of Pulmonary Disease”, Computers, IEEE Transactions on, Vol C-21, Issue: 7, July 1972.

[15] Skalanskv.J “Image segmentatio and feature extraction”, IEEE Trans. System Man Cybernat., 8(4), 1978, 237-247.

[16] Yung-Chang Chen, Chung-Ming Wu, “Texture Features for Classification of Ultrasonic Liver Images”, IEEE Transactions on Medical, Vol.11, No. 2, Pages 141-151. June 1992


50

Design and Experimental Analysis of High Performance Microstrip Antenna

for Wireless Broadband Communication

Raheel M. Hashmi1, Arooj M. Siddiqui 2, M. Jabeen 2, K. Shehzad2, S. Muzahir Abbas2, K.S.Alimgeer2

1Department of Electronics & Information, Politecnico di Milano, Italy

1,2Department of Electrical Engineering, COMSATS Institute of Information Technology

Islamabad, Pakistan

Abstract: Antennas have fundamental importance in the field of wireless communication systems. With advances in miniature design for communicating devices and broadband technologies, needs for low-cost and small sized antennas supporting broadband communications have grown like never before. In this paper, we present a simple and very low-cost design of a wide-band micro patch antenna for operation in the broadband wireless spectrum. The proposed design has been optimized using quarter-wave strip-line approach to provide high performance at a wide functional bandwidth while minimizing reflection effects. The design has been fabricated and the prototype been analyzed using the results from Network Analyzer. The results of the analysis suggest that the characteristics of the fabricated prototype comply with the simulated outcomes. Hence, the proposed design proves to be simple, scalable and very low-cost with provision of optimum features by combining the advantages of broadband communication with micro-strip design.

Keywords:Broadband Communication; Micro-strip lines; Radio Frequency, Radiation Pattern;

1. Introduction Development of broadband communications has achieved

a very high pace in modern era. Antennas serve as the foremost interface between air and transceivers to transform electrical signals into electromagnetic energy in the free space and vice versa [1]. Modern communication devices demand portability, low manufacturing cost, small size and high performance with excessive bandwidth requirements. These needs are being met by new developments in the field of antenna design and analysis.

There are several classes of antennas like narrowband antennas, fractal or frequency independent antennas and wideband antennas. Broadband or wideband antennas are termed as those which can cover an octave or two around the designated centre frequency [2]. Present broadband communications involve IEEE 802.11 based Wireless Local Area Networks (WLANs) [3] and IEEE 802.16 based Worldwide Interoperability for Microwave Access (WiMAX) networks [4]. WiMAX serves as a solution for Wireless Metropolitan Area Networks and hence, can accommodate several WLANs for backhaul purposes [5]. The most widely employed spectral band for WLANs is 2.45 GHz [3]. However, depending upon the allocation

authorities of the realm, 2.45 GHz is also employed for operation of WiMAX networks [6]. Microstrip antennas are low profile antennas which can be easily mounted on surfaces due to their planar geometrical designs [7]. These antennas are manufactured by etching the designed prototype on a dielectric substrate with dielectric constants ranging as 2.2 ≤ εr ≤ 12 [7]. Designs with greater substrate thickness and lesser value of dielectric constants can increase efficiency of the antenna but introduce application constraints as well. Smaller and cheaper designs are required for implementation in practical systems to support wideband communications and offer low reflection losses.

In this paper, the proposed design is aimed to provide very low cost hardware and small size for operation on 2.45 GHz frequency band with optimum performance. Furthermore, the proposed design has been optimized using strip-line matching approach and offers very low reflection losses on the resonant frequencies. The resonant or nominal bandwidth offered by this prototype is more than 100 MHz and is supported by a very close to ideal voltage standing wave ratio (VSWR).

The rest of the paper is organized as: Section 2 describes the proposed antenna geometry. Section 3 highlights the modeling and designing process while Section IV depicts the radiation characteristics. Section 5 presents the results of the hardware prototype and Section 6 concludes the paper.

2. Proposed Geometrical Design The prototype has been designed for operation at 2.45

GHz band and the dimensions have been calculated by the equation set provided in [7]. The design consists of a rectangular patch printed on 1.6mm thick; FR-4 substrate. The substrate bed has dimensions of 50mm x 80mm and has ground plane printed on the lower face. Figure 1 shows the image of the proposed design. The antenna is fed using a 3mm x 17mm transmission line with 50 ohms resistance. The radiating patch and the transmission line (T-Line) have been matched using a strip-line to act as a quarter-wave transformer to manipulate the fact suggested by T-Line theory that open and short circuit effects alternate at distances separated by factor of λ/4. The impedance of the quarter-wave transformer is calculated using (1) as:

Ζ1 = (Ζ0 RL) 1/2 (1)


51

Where RL is the load resistance calculated by using the relation of transmission coefficient and impedance ratios [8] and Z0 is the characteristic impedance equal to 50 ohms. The dimensions of quarter-wave transformer are 0.1mm x 29.2mm.

Figure 1. Radiating Patch Prototype Layout

The design and software simulations have been performed using the software package AWR Microwave Office (MWO-228, VSS-100). This package has an embedded graphical user interface which can help to calculate the dimensions of the quarter-wave transformer. The patch is fed using female SMA connector which has 50 ohms impedance to match the T-line.

3. Modelling and Design The antenna design has been modeled in AWR

Microwave Office Design Environment which is based on the Method of Moments to calculate the current distributions on the patch and determine the radiation characteristics [9]. The modeling structure has been designated to be of copper metal with radiating and ground planes sandwiching the FR-4 substrate between them. The radiating patch elements consist of strip-lines and bends as shown in figure 1. The layered metallic frame is then filled by using copper conductor plane and the final shape is presented in figure 2. Layering properties are adjusted for simulation in the design environment. The adjustments of the layering parameters are made according to data presented in Table 1.

Table 1: layering parameters adjustments

Layer Type Thickness

(mm) Permittivity

(εr)

Loss Tangent

(δ)

1 Air 16 1 0

2 Substrate (FR-4) 1.6 4.7 0.019

Figure 2. The basic building frame of radiating patch

4. Radiation Characteristics

The results obtained from the simulations include the radiation pattern and the return loss. The return loss experienced is measured to be -34.59 dB whereas the resonating bandwidth calculated at -10 dB is approximately 150 MHz. According to IEEE 802.16 specifications, the channel bandwidth for WiMAX is scalable between 1.25 - 20 MHz whereas of 802.11 specifications state a channel bandwidth of 20 MHz (non-overlapping) [3]. This endorses the fact that not only the designed antenna possesses wide-band characteristics, but also it provides ease of operation over the 2.45 GHz band for both types of broadband communication networks.

Figure 3. Radiation Plot in Elevation Hemisphere


52

Figure 4. Radiation Plot in Azimuthal Hemisphere

Figure 3 and 4 show the simulated radiation patterns of the antenna in horizontal and vertical planes. The radiation plots suggest a symmetric behavior with low directional properties. The half power beam-width of the antenna is calculated to be 60o whereas the first null beam-width is approximately 160o.

5. Experimental Results & Analysis The hardware prototype of the simulated design was

fabricated using the process of photolithography and was connected with a SMA 4-hole female connector for practical analysis. The prototype was analyzed using Agilent ENA Series Network Analyzer Model-E5062A. The tests were made to measure return-loss, VSWR, group delay experienced by the transmission and the impedance characteristics. The obtained results are presented in this section.

Figure 5. Return Loss and Resonant Bandwidth

Figure 5 shows the plot of RL which has increased in the practical prototype to -40 dB with a resonant bandwidth of

approximately 100 MHz at RL -10 dB. This low value of RL yields a reflection coefficient (τ) to be 1x10-4.

Figure 6. Voltage Standing Wave Ratio (VSWR)

Figure 6 shows the values of VSWR plotted against a range of frequencies. At 2.45 GHz, the value of VSWR is measured to be 1.0545 which is very close to the ideal response valued unity. This result clearly depicts that the reflections and thus, the creation of standing waves has been extremely suppressed due to the optimization using the strip-line approach.

Figure 7. Antenna Impedance plotted on Smith Chart

Figure 7 shows that the plot of input impedance on a smith chart generated by the network analyzer. The point in figure 7 bounded by a triangle and marked ‘1’ represents the value of impedance at operating frequency of 2.45 GHz. This has been plotted using a frequency sweep of 2.3 – 2.7 GHz.

In figure 8, the plot of group delay experienced by the transmission has been shown which comes out to be


53

2.9152x10-9 seconds, which is very low. The results illustrated in figures 5 – 8 reveal that the performance of the simulated design in comparison to that of hardware prototype is considerably close.

Figure 8. Plot of group delay experienced by the

transmission

Moreover, the performance results obtained from the fabricated prototype can be expressed as even better than the simulated characteristics.

6. Conclusion The proposed design, its simulation results and the

experimental analysis agree together and hence, it can be concluded that this antenna can provide favorable outcomes when employed in WLANs or WiMAX networks, specifically with user-end communicating devices. Moreover, it can also be extracted that extension of this prototype into a scalable array of similar antennas can be helpful for implementation as a base station transmitter as well. The design and analysis has been successful and the hardware antenna performs quite close to the simulated structure.

Appendix I

Antenna Prototype under test on E5062A Network Analyzer at Communications Lab., CIIT Main Campus,

Islamabad, Pakistan

References [1] Constantine A. Balanis, “Antenna Theory: Analysis &

Design”, Chapter 1,3rd Edition, Jhon Wiley & Sons Inc. USA, 2003.R. Caves. Multinational Enterprise and Economic Analysis, Cambridge University Press, Cambridge, 1982. (book style)

[2] Philip Felber, “A literature study: Fractal Antennas”, Illinois Institute of Technology, USA, December 2000.

[3] H. Labiod, H. Afifi, C. De Santis,”Wi-Fi, Bluetooth, Zigbee and WiMAX”, Chapter 2: Introduction to Wi-Fi, Springer, Netherland, 2007. ISBN 978-1-4020-5397-9

[4] R. M. Hashmi et. al., “Improved Secure Network Authentication Protocol for IEEE 802.16”, 3rd IEEE International Conference on Information and Communication Technologies, Pakistan, 2009., in press

[5] Alberto Escudero-Pascual, “WLAN (IEEE 802.11 B) and WMAN (802.16 A) Broadband Wireless Access: when opportunities drive solutions”, Royal Institute of Technology, Stockholm, Sweden, 2002.

[6] WiMAX Forum, “Online MAPS”, retrieved 30th July 2009 from http://www.wimaxmaps.org/

[7] Constantine A. Balanis, “Antenna Theory: Analysis & Design”, Chapter 14, 3rd Edition, Jhon Wiley & Sons Inc. USA,pp 811-820, 2003.

[8] David M. Pozar, “Microwave Engineering”, 3rd Edition, Jhon Wiley & Sons, USA, 2005.

[9] Robert E. Collin, “Antennas and Radiowave Propagation”, Chapter 2, Mc-Graw Hill, Singapore,pp 46-50, 1985.

http://www.wimaxmaps.org/


54

Optimal Channel Allocation in Wireless Sensor Network using OFDMA

Mohd. Sabir1, Rakesh Kumawat2 and Dr. V.S.Chouhan3

1Electronics & Communication Department, Sobhasaria Engineering College,

Sikar (Raj.), India [email protected]

2 Electronics & Communication Department, Sobhasaria Engineering College,

Sikar (Raj.), India [email protected]

3Prof. & Head, Electronics & Communication Department,

I.E.T. Alwar (Raj.), India [email protected]

Abstract: Wireless Sensor Network is a Wireless network of Spatially Distribution of Various Autonomous Device which using Sensors to Co-operative Monitor Physical or real environment conditions like temp., light etc. in wireless sensor network, we require a fast data transfer between one node to another node in critical real-time applications, in present ,we face slow data rate between one sensor node to another sensor node.In this Paper, we want to solve this problem and propose unique approach for optimal channel allocation in wireless sensor network using OFDMA. We assume that sensor node knows channel state information (CSI) partially, and perfect queue state information (QSI). Our Objective is to minimize long-term average packet delay over multiple time slots.

Keywords: Wireless sensor network, channel state information, data rate, OFDMA, Channel allocation.

1. Introduction A wireless sensor network (WSN) is a wireless network consisting of spatially distributed autonomous devices using sensors to cooperatively monitor physical or environmental conditions, such as temperature, sound, vibration, pressure, motion or pollutants, at different locations. The development of wireless sensor networks was originally motivated by military. Applications such as battlefield surveillance. However, wireless sensor networks are now used in many civilian application areas, including environment and habitat monitoring, healthcare applications, home automation, and traffic control. In addition to one or more sensors, each node in a sensor network is typically equipped with a radio transceiver or other wireless communications device, a small microcontroller, and an energy source, usually a battery. The envisaged size of a single sensor node can vary from shoebox-sized nodes down to devices the size of grain of dust, although functioning 'motes' of genuine microscopic dimensions have yet to be created. A sensor network normally constitutes a wireless ad-hoc network, meaning that each sensor supports multi-hop routing algorithm (several nodes may forward data packets to the base station).We propose a heuristic Approaches, which we show through simulation performs better, in comparison with the

other optimal algorithms, which consider perfect CSI, when error in channel estimation is high. We then consider the problem of Designing a centralized scheduler in data transmission that maximizes the average sensor node throughput, and satisfies the average delay constraint of individual node. We assume a data transfer between one node to another node in wireless sensor node is use single channel and multiple queues at the base station. We propose an online throughput optimal delay constraint satisfying (OTODCS) algorithm for this model, and we will show through simulation that it performs better than multi-queue TDM system under the same set up. Further we extend the same problem to a problem with multicasting data transfer with optimal channel allocation. We will show via simulation that we will be get, in both throughput and delay in comparison with the other online algorithm with single server. OFDM is a promising technique for broadband wireless communication in a multi-path environment with frequency selective fading. OFDM achieves high spectral efficiency in multi user environment by dividing the total available bandwidth to narrow-bands in efficient way. This allows the mobile to spread information selectively in order to avoid sub-bands where fading occurs. For the user downloading data, it is required to have as much as throughput as possible. OFDM achieves high spectral efficiency by converting a frequency selective channel into a set of parallel non-interfering frequency flat channels on which efficient power allocation can be done. When we assume that each user has infinite amount of data stored in buffer water filling is throughput optimal. However, when considering stochastic packet arrival with finite rate, water filling is not throughput optimal. However in all previous work they assume perfect CSI at the base station, but in practice only the inaccurate estimate is available and optimal policies may not be robust to this estimation error. It is interesting to study performance of these scheduling algorithms under partial CSI.

2. Problem Definition and assumption

We consider a multi-queue multi-server system with stochastic connectivity’s. There are N queues (users) and K





55

servers (sub-carriers). Fixed-size packets arrive stochastically for each user and are transmitted over a set of allocated servers. Each user has an infinite buffer to store the data packets that cannot be immediately transmitted. The system is time slotted. The users have the same priority and are symmetric, i.e., they have statistically identical arrival and channel connectivity processes. At the beginning of each time slot, the assignment of servers to users is instantaneous and made by a centralized resource manager. The resource manager has perfect knowledge of the current queue backlogs and the connectivity’s which are assumed to be constant during a time slot but varying independently over time slots (e.g. block fading model). We do not allow sharing of any server and assume no error in the transmission. We use the following convention: lower case for scalar, bold faced lower case letters for row vector, upper case letter for matrices and scripted upper case letter for space of matrices. b(n) = (b1, b2, ..., bN) : Backlogs (in units of packets) of each queue in the beginning of time slot n. • a(n) = (a1, a2, ..., aN) : Stochastic number of fixed-length packets arrived to each queue during a time slot n. The new packet arrivals at time n can be served only at time n + 1 or after. • C(n) = [ci,j ]: the K-by-N stochastic connectivity matrix at time slot n where ci,j Є{0, 1, ..., cmax < ∞} denotes the maximum number of packets sub-carrier i can serve from queue j at time n, and we further assume that ci,j takes discrete values form 0 to cmax, where cmax is the maximum number of packets that a user j can transmit in channel i. • ξ(n) = [ξi,j ]: the K- by- N matrix , each element of this matrix is a circularly symmetric complex Gaussian random variable. • W(n) = [wi,j ]: K-by-N allocation matrix at the beginning of time slot n where wi,j {0, 1} and wi,j = 1 denotes that sub-carrier i is assigned to serve queue j during time n. In our model we assume that there is error in channel estimation. We denote the Erroneous connectivity matrix by H(n) = C (n) + ξ (n), If above is the equation for channel visible to base station, we assume that base station Does MMSE estimation on the received channel matrix H(n). And with MMSE estimate. 3. Problem Formulation and Assumption

The channel estimation at base station is given by [14]: Ĥ(n) = SNRest / 1+ SNRest {C(n) + ξ(n)} Where SNRest = E[|C|2] / E[|C|2] + E[|ξ|2] The dynamics of queue length vectors under allocation W (n) are described by equation b(n+1) = [b(n)-1W(n)Θmin C(n),H(n)]+ +a(n),n=1,2… where element wise product (W(n) ּס min(C(n),H(n)) is a matrix [wi,j min(ci,j ,Hi,j)], 1 is a K dimensional row vector of K ones, and, for a vector, [v+] = [v+ 1 , ..., v+ N] with v+ j = max{0, vj}. Definition 1: For a row vector x = (x1, ..., xN) and a matrix Y = [y1, ..., yN] where yj is a column vector, a column by

column matrix permutation Ππ corresponding to a permutation π is defined as, for any j and k Є {1, ...,N}, π (xj) = xk ↔ Ππ (yj) = yk Using the above notation and definition, we make the following symmetry assumptions on the arrival and connectivity processes: A(1) The packet arrival processes [a(n)] to users’ queues during each time slot are i.i.d across time slots. The packet arrival processes are symmetric or exchangeable, i.e., the joint probability mass function is permutation invariant, i.e.,

P[a(n) = π(x)] = P[a(n) = x] for any n, vector x, and permutation π. A(2) The connectivity profiles [C(n)] are i.i.d across time slots and exchangeable across server, i.e. for any n, matrix Y , and column by column permutation matrix Ππ Assumption A(2) is valid when channel and the mobility creates a homogeneous environment for all users. Note that A(1) and A(2) imply independence across time slots but not across users, i.e., at a given time the arrival to various queues need not to be independent. Note that we will some times use the vector w(n) to mean the matrix product 1W(n). 4. Mathematical Formulation Problem (P)

Consider the system described above. We wish to determine a Markov server allocation policy σ that minimizes the cost function at the finite horizon T: = E | ] Where summarizes all information available at time zero and denotes the cost under Markov policy σ over horizon T. = where the cost function ø(b) = g(bj) and g is a convex and strictly increasing function. Note that when g is an identity function, problem reduces to a average total backlog (E [ bj(t)]) minimization problem over horizon T. From Little’s theorem, any optimal policy that achieves minimum average backlog achieves minimum average packet delay as well. Thus, in this special case, the study reduces to the study of the average delay minimization. 5. Proposed Solutions to the Problem In this section we present a heuristic policy as a solution to the above problem. We modify maximum throughput load balancing (MTLB) algorithm and maximum weighted matching (MWM) algorithm proposed by Kittipiyakul and Javidi in [5] to take into account the error in channel estimation. In [13] authors proved that MTLB policy is a special case of MWM when we assume channel is On/Off. Our algorithm is valid for both On/Off channel and general channel with cmax > 1 by the arguments in [13]. We assume full queue state information (QSI) and partial channel state information (CSI). Algorithm: • X = {1, 2, ..., K} • loop (until stop):


56

– If X = Φ, then stop – (i*,j*) = arg maxi∈X,j∈{1,2,...,N} bj( (Ci,j\ Hi,j)) ; – if bj Hi,jP (ci,j |Hi,j)> 0 then w∗ i*,j* = 1 – bj* = bj* − min(Hi*, j*, ci*,j*) and X = X − {i∗ } • Assign W∗ = wi*, j*

6. Simulation parameter Before we provide the details of simulation set up let us first discuss the MTLB and MWM algorithm:

(1) MTLB Algorithm: The subcarrier assignment uses full information about the queue lengths (full QSI) and binary (ON-OFF) information about the channel: a sub-carrier is considered ON if ci, j > cthreshold. Then MTLB policy is used for sub carrier allocation.

=

For state (b,ĉ). Compute an MTLB allocation W∗

(2) MWM Algorithm: The sub carrier assignment uses full queue state information (full QSI) and full channel state information (full CSI) • X = {1, 2, ..., K} • loop (until stop): – If X = Φ, then stop – (i∗ , j∗ ) = arg maxi,j∈{1,2,...,N} bjci,j ; – if bj*ci*,j* > 0 then w∗ i*,j* = 1 else stop; – bj* = bj* − ci*,j* and X = X − i∗ • AssignW∗ = wi*,j*

where ci,j denotes the channel capacity for user j on server i, K denotes the number of sub carrier and N denotes the number of user (queues).We now compare the performance of MTLB algorithm and MWM algorithm in [5] with our heuristic policy. We consider downlink OFDM consisting of 32 statistically independent and identical users, and 128 sub carriers. We assume equal power distribution among all users, and further assume that the loss of throughput due to equal power allocation is tolerable [15]. We can represent the number of packets per time slot the sub carrier i can transmit to user j, in terms of the erroneous channel gain gi,j, as follows:

Hi, j =

where D is the number of QAM symbols per channel in a time slot, β the fixed packet length (in bits) and N0 is the noise power in the sub carrier. The parameters P, D, β and N0 are chosen such that the connectivity Hi,j ∈ {0, 1, 2, 3}. All simulations are conducted over 6,000 time slots. We assume arrivals to each user have the Poisson distribution and are independent from user to user.

7. Results and Comparisons

We compare the performance of our algorithm with the maximum throughput load balancing (MTLB) and maximum weighted matching (MWM) algorithm pro-posed by Kittipiyakul and Javidi in [13] with error in channel estimation error under different channel model. We show

via simulation that our algorithm performs better than MTLB when error in channel estimation is high and channel is On/Off. Figs. 1 to 3 above show the simulated performance in terms of average queue backlogs for Poisson distribution when the channel is On/Off. The simulation results in Fig.2 demonstrate the superior performance of MTLB algorithm with reliability information in comparison with simple MTLB algorithm with channel estimation error and no reliability information. By reliability information we mean probability with which estimation is correct. In MTLB the channel is On/Off. It is clear with simulation results in Fig. 1 that we do well in terms of the average queue backlog when error in channel estimation is high. From Figs. 2 and 3, it is clear that we gain nothing by reliability information, and the performance is almost same as the MTLB even with channel estimation error. While Figs. 4 show the simulated performance under the same set up with the condition that channel is more general, i.e., Hi, j ∈ {0, 1, 2, 3}. It is clear from the simulations that when channel is general rather than just On/Off we gain nothing from reliability information. Our heuristic policy performs almost same as the MWM with error in channel estimation.

Figure 1. Average queue backlog for MTLB policy with

SNR = 0 dB


SNR = 6 dB


57


SNR = 9 dB

Figure4.Average queue backlog for MWM with SNR = 0dB

8. Performance and Comparisons Before we provide the details of simulation parameters let us first discuss a Heuristic algorithm which is also a delay satisfying algorithm. Heuristic Algorithm: We state here the heuristic algorithm proposed by Salodkar et.al, in [12]. Let in a time slot n, i be the estimate of i(n), the average rate at which data can be transmitted to a user i. We calculate i using exponential averaging as follows,

i (n+1) = α i(n) + (1-α) yi(n), Where yi (n) is the rate at which data can be transmitted to a user in time slot n and α ∈ (0, 1). The heuristic scheduler chooses a user i in time slot n such that,

We perform the simulation using MATLAB. We simulate the system for 50 users, then we divide user’s in 10 groups and each group has 5 users. The arrivals to each users queue are i.i.d., and we use the Poisson distribution of arrivals for each user, bounded to 8. The distribution for each user channel is Rayleigh, The mean of |hi(n)|2, i.e., is the average SNR of a user (expressed in dB). We normalize the ratio P/N0 to 1. The rate at which user can receive data from a base station in a time slot and is upper bounded by 25. The throughput achieved in a time slot is expressed in packets/timeslot/Hz. Each user has an

average delay constraint in terms of number of slots. All users in a group have the same system parameters. We do two experiments. In experiment 1, we vary the arrival rate for group 1 and measure the user delay and throughput, while in experiment 2 we vary the average SNR of the users in a single group and measure the same quantities as in experiment 1. For both the experiments, we compare 4 schedulers, i.e., Opportunistic scheduler, heuristic, OTODCS scheduler in [12] and OTODCS scheduler for broadcast channel case. In the OTODCS scheduler for broadcast channel we assume that at the most 2 users can be scheduled at a time, and also assume equal power distribution among them. Figure 5shows the variation of the delay of a user in group 1 against various arrival rates. Figure 6 shows the variation of the system throughput against various arrival rates for group 1. In this experiment, group 1 has poor channel conditions; hence the pure opportunistic scheduler allocates a small share of bandwidth to this group. From figure 5 we observe that pure opportunistic scheduler achieves a very high user delay. The heuristic scheduler, OTODCS scheduler for the broadcast case and its TDM counterpart all satisfy the delay requirements, and we see in Figure 6 that throughput achieved by all the scheduler except OTODCS for broadcast are same. However the throughput for OTODCS broadcast case is higher than the other scheduler. From Figure 5 it is clear that OTODCS scheduler for the broadcast case do much better than the others when we compare in terms of delay.

Figure 5. Single user delay variation with arrival rate

Figure 6. System throughput variation with arrival rate


58

Figure 7. Single sensor node variations with channel

condition

Figure 8. System throughput variations with channel

condition Figure 7 shows the variation of the delay of user in group 10 against various channel gain (average SNR). Figure 8 shows the system throughput against various channel gain (average SNR) for group 10. In Figure 5.3 it is clear that OTODCS scheduler for the broadcast case, OTODCS scheduler with TDM and heuristic policy all are able to satisfy delay requirements, but the delay performance of OTODCS for broadcast case is the best among all. We can observe form Figure 8 that OTODCS for the broadcast case do better than the other in terms of throughput as well. 9. Applications The applications for WSNs are many and varied, but typically involve some kind of monitoring, tracking, and controlling. Specific applications for WSNs include habitat monitoring, object tracking, nuclear reactor control, fire detection, and traffic monitoring. In a typical application, a WSN is scattered in a region where it is meant to collect data through its sensor nodes.

10. Conclusion

The problem of delay optimal server allocation policy studied in literature is with full CSI and QSI. We considered a more practical scenario where there is channel gain estimation error. We proposed a heuristic policy as a solution to this problem. We then compared the policy with

some of the existing ones. We have shown via simulation that our algorithm performs better than MTLB algorithm when error in channel estimation is high, and channel is considered to be On/Off. We then formulated the problem of maximizing average throughput, with individual delay constraints, and we considered a broadcast channel at the downlink. We proposed an online algorithm as a solution to this problem. We compared our algorithm with some of the existing algorithms. We have shown via simulation that our algorithm gives better performance in comparison with its TDM counterpart in terms of throughput and delay. Further we extended our problem to two server case. We have shown that this has the same structure as single server problem. We have shown via simulation that we gain in performance when we increase the number of servers to two.

References [1] R. S. Cheng and S. Verdu, “Gaussian multiaccess

channel with isi: capacity region and multiuser waterfilling,” IEEE Trans. on Inform. Theory, vol. 39, no. 3, pp. 773–785, May 1993.

[2] R. V. Nee and R.Prasad, “OFDM for wireless communications,” Artech House,Boston, 2000.

[3] T. Javidi, “Rate stable resource allocation in OFDM systems: from waterfilling to queue balancing,” Allerton conference on communication, control and computing, September 2004.

[4] G. Li and H. Liu, “Dynamic resource allocation with finite buffer constraint in broad-band OFDMA networks,” IEEE Wireless Communication and Networking, pp. 1037-1042, March 2003.

[5] S. Kittipiyakul and T. Javidi, “Delay-optimal server allocation in multi-queue multi-server systems with time varying connectivities,”IEEETrans. on info. theory.

[6] L. Tassiulas and A. Ephremides, “Dynamic server allocation to parallel queues with randomly varying connectivity,” IEEE Trans. on Inform. Theory, vol. 39, no. 2, pp. 466–478, May 1993.

[7] A. Ganti, E. Modiano, and J. T. Tsitsiklis, “Optimal transmission scheduling in symmetric communication models with intermittent connectivity,” IEEE Trans. On Inform. Theory, vol. 53, no. 3, pp. 998–1008, March 2007.

[8] X. Liu, E. Chong, and N. Shroff, “Transmission scheduling for efficient wireless re-source utilization with minimum performance guarantees,” Proceeding of IEEE VTC,vol. 2, pp. 824–828, October 2001.

[9] S. Kulkarni and C. Rosenberg, “Opportunistic scheduling policies for wireless systema with short term fairness constraint,” IEEE GLOBECOMM, vol. 1, pp. 533–537, December 2003.

[10] R. Berryand and R. Gallager, “Communication over fading channel with delay constraint,” IEEE Trans on Info. Theory, vol. 48, no. 5, pp. 533–537, May 2002.

[11] E. Yeh and A. Cohen, “Dealy optimal rate allocation in multiaccess fading channel,”IEEE Worksop on Multimedia Signal Processing, pp. 404–407, October 2004.


59

[12] N. Salodkar, A. karanidikar, and V. S. Borkar, “Delay constrianed point to multipoint scheduling in wireless fading channel,” IEEE Journal on selected areas in communication.

[13] S. Kittipiyakul and T. Javidi, “Resource allocation in OFDMA with time-varying channel and bursty arrivals,” IEEE Communication Letters, vol. 11, no. 9, pp. 708–710, September 2007.

[14] S. M. Kay, Fundamentals of statistical Signal Processing: Estimation theory. Prentice Hall, 1993.

[15] A. Czylwik, “Adaptive OFDM for wideband radio channels,” IEEE Globecomm, vol. 1, pp. 713–718, 1996.

[16] D. P. Bertsekas, Nonlinear Programming. MA: Athena Scientific, 1999.

[17] V. S. Borkar, “Stochastic approximation with two time scales,” System and Contol Letters, vol. 29, 1996.

[18] J. C. Spall, Introduction to Stochastic Search Optimization. Wiley, 2003.

Mohd. Sabir received the B.E. degree (Electronics & Communication) from MAIET, Jaipur (Raj.) in 2003. And M.E. degree (Digital Communication) from M.B.M. Engineering College, Jodhpur (Raj.) in 2009. He now with Sobhasaria Engineering College, Sikar, as lecturer in department of electronics.

Rakesh Kumawat received the B.E. degree (Electronics & Communication) from SEC, (Raj.) in 2007. And M.Tech. degree (VLSI Design) from C-DAC Mohali in 2009. He now with Sobhasaria Engineering College, Sikar, as lecturer in department of electronics.

Vijay S. Chouhan was born in India in 1960. He received B.E. degree in Electronics & Communication Engineering and M.E. degree in Digital Communication Engineering from J. N. Vyas University, Jodhpur, India. He submitted his Doctoral thesis on Biomedical Signal Processing in 2007. Presently he is working as Professor in the Department of Electronics &

Communication Engineering, Institute of Engineering & Technology, Alwar, India. His research interest includes fields of Biomedical Signal Processing, Soft Computing and Digital Communications.


60

Performance Analysis of Diversity Schemes in Multi hop Ad hoc Networks

Meenu Khurana1 and C. Rama Krishna2

1Department of Computer Science & Engineering, Chitkara Institute of Engineering and Technology,

Chandigarh Patiala Highway, Rajpura, Punjab 140 401, India [email protected]

2 Department of Computer Science & Engineering, National Institute of Technical Teachers’ Training & Research,

Sector 26, Chandigarh 160 026, India [email protected]

Abstract: Transmit diversity in multiple input multiple output (MIMO) systems is one of the best methods of overcoming the detrimental effects of multipath fading in a wireless channel. Increasing the transmit antennas improves network performance but increases complexity in return. To reduce this complexity, transmit antenna selection (TAS) technique is implemented.

In this paper we evaluated the performance of various diversity schemes implementing quasi orthogonal space time block codes (QSTBC) and TAS. In view of the increase in the usage of multi-hop ad hoc networks in various applications, it becomes important to enhance its performance. So we evaluated the performance of multi-hop ad hoc networks by implementing these diversity schemes to get the best possible scheme. The simulation results obtained have been analyzed in terms of packet delivery ratio (PDR), number of packets dropped and energy consumption (EC).

Keywords: diversity, MIMO, multipath fading, QSTBC, TAS,

multi-hop

1. Introduction The nodes with multiple antennas offer significant diversity advantage by exploiting both transmitter and receiver diversity using various space-time coding schemes [1], [2]. One of the simplest and widely used transmit diversity technique that implemented Orthogonal Space Time Block Codes (OSTBCs) using two transmit antennas is Alamouti Scheme [3]. It is proved that OSTBCs, which provides full diversity and full transmission rate, is not possible for more than two transmit antennas [2]. To overcome this limitation QSTBCs, which achieve full rate at the cost of orthogonality, are designed [4], [8]. TAS is a technique to reduce complexity raised by adding more number of antennas. In this method optimum number of antennas for MIMO system is decided on the basis of trade off between increased complexity and improved bit error rate (BER) performance [5], [6].

In this paper we have proposed a diversity scheme for multi-hop ad hoc network that gave better performance in comparison to other schemes. The work is done in two parts, firstly the performance comparison of different diversity schemes has been evaluated with BER and signal to noise ratio (SNR) as metrics. Finally, we evaluated the performance of these schemes i.e. single input single output (SISO), multiple input single output (MISO) and TAS, in ad hoc networks. The performance of these schemes is

compared by carrying out extensive simulations using GloMoSim network simulator.

The rest of the paper is organized as follows. Section 2, explains QSTBC and TAS terminologies. Section 3 represents simulation details and performance comparison of different diversity schemes. Section 4 gives the performance analysis of diversity schemes in ad hoc networks. Finally, conclusions are presented in section 5.

2. QSTBC and TAS An OSTBC is a linear space time block codes (STBC), for which the code matrix S has the following unitary property:

IsSS N

nn

H ∑=

=1

2 (1)

(.)H in equation (1) denotes the Hermitian conjugate, which is a complex conjugate of the symbol.

The orthogonality enables to achieve full transmit diversity and simple, linear and optimal decoding at the receiver side [2]. It allows the receiver by means of simple Maximal Ratio Receive Combining (MRRC) to decouple the signals transmitted from different antennas. Simple ML decoding can be applied for OSTBCs, so that two symbols which are coded together can be detected independently at the receiver [3]. The main disadvantage of OSTBCs is a code rate less than one symbol per time slot for a system with more than two transmit antennas. The code rate may be improved by using QSTBC, which can give full diversity by signal constellation rotations, but require joint ML detection of pairs of symbols. These codes achieve full data rate at the expense of a slightly reduced diversity [4]. In QSTBC, the columns of the transmission matrix are divided into groups, columns within each group are not orthogonal to each other but different groups are orthogonal to each other. The code construction for QSTBC with NT = 2n , (NT ≥ 4) is done in the following recursive way

(2)

( ) ( ) ( )( ) ( )

=

Θ=−Θ+=

+==

=TN

TN

jjxTNGTNTNTNj

jxTNG

TNTNj

jxTNGTN

jjxTNG

TT

NjjxNG

21}{

212

}{2

12

}{2

21}{

2

1}{




61

where T

TN

Njj xxxx ......,}{ 211 == and the diagonal

],[ 22TNTN

matrix TNΘ is given by

)})1({( 11 T

T

Nj

jN diag =

−−=Θ

Using equation (2), QSTBCs for four transmit antennas

can be represented as,

−−−−

−−=

1234

*2

*1

*4

*3

*3

*4

*1

*2

4321

ssssssssssssssss

s (3)

The conventional multiple-antenna system requires the

number of radio-frequency (RF) chains equal to the total number of antennas [7], which presents a hardware challenge in terms of complexity and cost. TAS system (NT, N; NR) implies that subset of N antennas is selected out of total of NT antennas that reduces the complexity. Instead of NT number of RF chains, N chains will suffice. TAS requires the feedback link to find the best subset of antennas. The criterion for selection of active antennas is based on signal to noise ratio (SNR).

3. Performance Comparison of Different Diversity Schemes The scheme with no diversity (1 x 1) is compared with Alamouti scheme for (2 x 1) and further with the extension to (2 x 2) scheme. All these matrix dimensions represent number of transmit and receive antennas in the form (NT x NR). The implementation of diversity schemes is done using MATLAB.

To compare the performance of diversity schemes implementing QSTBC or OSTBC, four transmit antennas are considered. The reason for considering four antennas is the availability of rate one QSTBC for four antennas as specified in (3). The quadrature phase shift keying (QPSK) modulation scheme is used for transmission. The channel considered is Rayleigh fading channel with additive white gaussian noise (AWGN). We have considered one receive antenna. ML decision metric is used in both the schemes. The receiver in OSTBC scheme decodes the symbols one by one while the decoding for the rate one quasi-orthogonal code is done for pairs of symbols [9].

TAS (8,4;1) scheme is considered for performance comparison with scheme implementing (8x1) QSTBC. The simulation result is presented for independently identical distributed (i.i.d.) channels. Four channels which had higher SNR as compared to others were selected out of 8 antennas. The channels are modeled as Rayleigh fading channels and channel coefficients as complex zero mean Gaussian random variables with unit variance. The perfect channel knowledge at the receiver and partial channel knowledge at the transmitter is assumed [10]. Based on SNR selection

criteria, the “best” four transmit antennas are selected by finding the four highest euclidian norm values of the columns of channel matrix. The symbols encoded by applying QSTBC are transmitted through selected transmit antennas. Signal transmission follows the mathematical expression given by equation (1). The results are obtained by averaging 10,000 different channel realizations.

Figure 1. BER vs. SNR for SISO, MISO and MIMO

systems

Figure 2. BER vs. SNR plot for QSTBC and OSTBC with

4 transmit antennas

Figure 3. BER vs. SNR of QSTBC with and without TAS

As shown in Figure 1, the performance of transmit

antennas diversity scheme employing two transmit and single receive antenna is better than no diversity scheme. Further the performance can be enhanced by using two receive antennas instead of one. It is observed from Figure 2 that at low SNR the performance of QSTBC is better than that of the OSTBC, but is poor at high SNR. This is due to the fact that the slope of the BER-SNR curve depends on the diversity. Although rate one QSTBC code starts from a


62

better point in the BER-SNR plane but a code with full diversity benefits more from increasing the SNR. The performance comparison of multiple input single output (MISO) transmission schemes using QSTBC with TAS and without TAS is represented in Figure 3. The scheme implementing QSTBC on 8 transmit antennas is compared with TAS where 4 antennas are selected out of eight. As per the simulation results represented in Figure 3, the performance of TAS scheme is better than without TAS. The reason for better performance of TAS scheme can be the closed loop approach that allows to assign the transmit power onto the antennas with the least path attenuations (i.e. highest path gain).

4. Performance Evaluation of Diversity Scheme in Multi-Hop Ad Hoc Networks The diversity schemes discussed in the previous section and implemented on ad hoc networks are SISO, MISO with two transmit and one receive antenna and TAS ( 8,4;1).

Extensive simulations for the proposed schemes have been carried out using GloMoSim, which is designed using a layered approach similar to the OSI seven layer network architecture [12]. The main configuration parameters for setting up a scenario are defined in CONFIG.IN file. We considered 75 mobile nodes placed randomly within 1500 x 1500 m2 area that follows random waypoint mobility model (RWMM). In this model, node randomly selects a destination from the physical terrain, and moves with a speed uniformly chosen between a minimum and maximum speed. After the node reaches its destination, it stays there for a specified pause time, which is considered to be zero in our study to find performance in a worst case scenario. The packet reception model is based on BER. This parameter looks up BER values in the SNR-BER file. These BER files for each of the schemes are obtained through simulations. A summary of the salient simulation parameters is provided in Table I.

Table I: Salient Simulation Parameters

Traffic model is constant bit rate (CBR) with each source generating 4 packets/second. The data rate considered is 2 Mbps while the size of each data packet is 512 bytes. Randomly chosen source destination pairs are used for simulation. For example, 15 source destination pairs (SDPs) among 75 nodes specifies that there are 15 source and 15 destination nodes. Hence, out of total of 75 nodes, 30 nodes are engaged in data transfer, however all the 75 nodes will help in routing or forwarding, etc. The time specified for the simulation is 30 minutes. Five runs with different seed values have been conducted. The data collected is then averaged over these runs. The performance metrics chosen for study are defined as: Packet delivery ratio (PDR): The ratio of data packets delivered to the destinations and data packets originated by the CBR sources. Packets Dropped: This parameter gives the total number of packets dropped due to unavailability of routing path. Energy Consumption (EC): It is the sum of energy consumed by each node in successful and unsuccessful transmissions, reception, overhearing and idle listening. This value is normalized over the total number of packets delivered. This parameter is significant in a way that some or all of the nodes in a network may rely on batteries or other exhaustible means for their energy [11]. For these nodes, the most important system design optimization criteria may be energy conservation.

Figure 4 shows the PDR for different number of SDPs for different diversity schemes. It is observed that PDR increases as we increase diversity. For each value of SDP, PDR is better for scheme with two antennas as compared to scheme with no diversity. Further performance of TAS in terms of PDR is best among all three schemes presented here. For each of these schemes PDR decreases with increase in number of SDPs. The transmissions from more number of sources may increase packet collisions.

Figure 4. Packet delivery ratio vs. No. of SDPs

PARAMETER VALUE MOBILITY MODEL RANDOM-WAYPOINT

MOBILITY-WP-PAUSE 0 seconds

MOBILITY-WP-MIN-SPEED 1 m/sec

MOBILITY-WP-MAX-SPEED 39 m/sec

RADIO-TYPE RADIO-ACCNOISE

RADIO-FREQUENCY 2.4 GHz

RADIO-RX-SENSITIVITY -91.0 dB

RADIO-RX-THRESHOLD -81.0 dB

DATA RATE 2 Mbps

MAC PROTOCOL 802.11 DCF

ROUTING PROTOCOL AODV


63

Figure 5. Normalized energy consumption (in

mWhr) vs. No. of SDPs

Figure 5 presents the energy consumption for each of these schemes. The total energy consumption by the nodes is normalized over the total number of packets delivered. It is observed that energy consumption per packet delivered is less in TAS as compared to MISO scheme, which in turn is less than SISO scheme. For all practical purposes we require nodes which consume less power, as they rely on batteries.

Figure 6. Packets dropped due to unavailability of routing

path

The number of packets dropped is least in case of TAS, which further increases with increase in SDPs, as is evident from Figure 6. This result again favours the application of MIMO systems in ad hoc networks.

5. Conclusion The simulation results for different type of coding scheme i.e. QSTBC and OSTBC shows that the rate one QSTBC code starts from a better point in the BER-SNR plane but OSTBC because of full diversity benefits more from increasing SNR. Practically for the SNR value range obtained in the real transmission scenario, the performance of QSTBC is better. TAS helps in increasing the potential performance of a system with minimal added hardware complexity as compared to simple diversity schemes. Diversity of the TAS system is same as the system with no antenna selection. The QSTBC with TAS outperforms no antenna selection scheme. We have compared the performance of SISO, MISO and TAS schemes in multi-hop ad hoc networks. The diversity gain offered by MIMO antennas due to reduced BER on a communication link for a given SNR value, is translated to a better performance at the higher layers. We observed improvement in performance in

terms of PDR, normalized EC in TAS scheme as compared to other two schemes. Further, we found that there is a substantial decrease in number of packets dropped due to unavailability of routing path. Overall, it is observed that TAS in multi-hop ad hoc networks has shown considerable improvement in performance as compared to other two schemes. It can further be concluded from this result that the congestion in network is less when transmit diversity schemes are applied.

References [1] J. Paulraj, D. Gore, R. U. Naba and H. B¨olcskei,” An Overview of

MIMO communications - A Key to Gigabit Wireless”, Proceedings of the IEEE, vol. 92, issue 2, pp. 198-218, 2004.

[2] V. Tarokh, H. Jafarkhani, and A.R. Calderbank. “Space-time block codes from orthogonal designs”, IEEE Trans. on Information Theory, vol. 45, issue 5, pp. 1456-1467, 1999.

[3] “A simple transmitter diversity scheme for wireless communications”, EEE Journal on Selected Areas in Communications, vol. 16, pp. 1451 -1458, 1998.

[4] H. Jafarkhani. “A quasi-orthogonal space-time block code”, IEEE Transactions on Communications, vol. 49, issue 1, pp. 1-4, 2001.

[5] L. Zheng & D.N.C. Tse. “Diversity and multiplexing: A fundamental tradeoff in multiple antenna channels”, IEEE Transactions on Information Theory, vol. 49, issue 5, pp. 1073-1096, 2003.

[6] F. Molisch & M. Z. Win. “MIMO systems with antenna selection”, IEEE Microwave Magazine, vol. 5, issue 1, pp. 46-56, 2004.

[7] Gorokhov, D.A. Gore & A.J. Paulraj. “Receive antenna selection for MIMO spatial multiplexing: theory and algorithms”, IEEE Transactions on Signal processing, vol. 51, issue 11, pp. 2796-2807, 2003.

[8] H. Boche and E. Jorswieck. “Universal Approach for Performance Optimization in Multiuser MIMO Systems”, European Trans. on Telecommunications, vol. 18, issue 3, pp. 217-233, 2007.

[9] E. Jorswieck and H. Boche. “Outage Probability in Multiple Antenna Systems”, European Trans. On Telecommunications, vol. 18, no. 3, pp. 287-304, 2007.

[10] T. Gucluoglu and E. Panayirci. “Performance of Transmit and Receive Antenna Selection in the Presence of Channel Estimation Errors”, IEEE Communication Letters, vol. 12, issue 5, pp. 371-373, 2008.

[11] Ramakrishna, Saswat Chakrabarti, and Debashish Datta. “Impact of contention resolution algorithms on the performance of IEEE 802.11 DCF based MAC protocol in a mobile ad hoc network”. In Proceedings of National conference on Communications (NCC-2005), pp. 78-81, IIT, Kharagpur, India, 2005.

[12] L. Bajaj, et.al. “Glomosim: A Scalable Network Simulation Environment”, MILCOM, 1999.

[13] C. Ramakrishna, s. Chakrabarti, and D. Datta. “Mobile Ad Hoc Networking: Overview, Applications and Challenges”. In Proceedings of the National Conferenece on Intelligent Systems and Networks (ISN- 2004), Jagadhri, Haryana, India, 2004.


64

Identifying, Measuring, and Implementing Cultural Difference in Cross-Cultural User Interface Design

Jasem M. Alostath1, Laila R. Abdulhadi Abdullah 2

The Public Authority for Applied Education and Training (PAAET), College of Business studies,

PAAET, Kuwait, P.O.Box 23167, Safat, 13092, Kuwait [email protected] [email protected]

Abstract: This paper introduces the Cross-Use experiment, which aims to evaluate the mapping between website design elements and cultural attributes using a user-in-context evaluation approach. This is done by developing three UI designs, and applying them to 63 local participants from the case study cultures (UK, Egypt, and Kuwait). The experiment was conducted using the developed prototypes, which was able to classify cultures differently, and highlighted those design markers that affects cultural differences in the design of e-banking websites. This is based on user preferences and usability. Finally, the cross-cultural usability tool in a form of Pattern Language was developed to show how the various forms of evidence relating to cultural usability can be made more accessible to designers.

Keywords: Usability, Cross-cultural design, Pattern Language, User-Centred Design.

1. Introduction Many cross-cultural design evaluations use existing websites designs in identifying cultural design differences. However, these design evaluations are not supported with a cultural model, or adopts cultural models that are not design oriented in interpreting design based on culture [5]-[8]. In our research of Culture-Centred Design (CCD) we have conducted design evaluations based on the identified subjective cultural attributes (CA) that characterize similarities and differences within and between user groups of different nationality of the cultural model that were developed based on HCI design [9][13]. The most important advantage of this new approach is that the results of the analysis provide the designer with sufficient information to generate new websites that are more sensitive to culture and genre variability. However, the designs generated are not guaranteed to be optimal. This is because: (1) the existing websites that form the basis of the analysis may not have been well designed from the cultural point of view, (2) the claims from the cultural-design mapping from which designs are generated may be insufficient to determine a unique design decision, and (3) the design analysis that is undertaken does not provide any important information on design aspects such as usability [9]. Our solution to this problem is in the CCD methodology [9], which uses the quantitative design analysis results to develop a number of possible prototype websites that will be culturally adapted to some degree. Then a rigorous user testing approach is used

to decide between the alternatives (further details about the CCD method see Alostath [9]).

2. Cross-cultural Design Claims In our earlier studies [9], an HCI-oriented model was developed based on the cultural four models of Hall [15], Victor [16], Hofstede [17] and Trompenaars [18]. This model was used with many design evaluation approaches to interpret the results of many of the existing designs, such those studies are conducted by Barber and Badre [5], which uses Cultural Markers approach, or website audit approach conducted by Smith et al. [8]. The results of both type of design evaluation and the interpretation using the HCI cultural model generates cultural design claims that was scoped based on national culture [14][15]. These cultural design claims are presented in a matrix structure called Culture-user Interface Design Matrix (CIDM) as shown in Figure 1, and an example of the actual cultural design claims are presented in Table 1.

Figure 1. The CIDM matrix: different levels




65

Figure 1 presents the four levels of the CIDM matrix. Level 1 presents culture using the HCI cultural model (HCICM), while level 4 shows the design level by presenting the cultural design claims that show design variability across-cultures. Any CA in level 1 will be linked to an abstract level of design in level 2 to show what UI aspect this CA is linked to. This can be achieved by the Cultural Profile (CP) results presented here when a new culture is introduced [13]. The CP presents cultural differences based on CAs’ values. These values help in predicting cultural design claims later in level 4. Then, the concrete mapping of culture and design is shown in level 3, at which level it shows the cultural design claims that are forming the links between culture and design. Each of these claims that present the claim description, the associated CAs, and finally, the references and results that support the design rationale are presented in level four (see Table 1). The references and design rationale column helps in keeping the design dynamic by reporting all the evidence that was either identified from the literature, obtained through design evaluation or user testing results. This makes the model dynamic and not fixed to specific understanding with loose connections to the context-of-use.

Table 1: Sample of the cultural design claims table Claim Code

Claim Description Associated Cultural

Attributes

References (Rationale)

C1 Users from low verbal oriented1 cultures need less textual detail2 information than high verbal oriented1 cultures, which are prone to presenting (or expected) higher need for textual2 detailed information.

[I1] Information Amount*, [I3] Information Speed

Fogg, 2003; Dormann and Chisalita, 2002; Alexakis, 2001; Husmann, 2001; and Sasse, 1997.

Legend * Primary Cultural cause 1 Cultural attribute variation 2 Related design change

3. Cross-Use Experiment: Method and Process The experiment design involves three national cultures, using three user interfaces for simple and complex tasks (3*3*2 mixed design). The independent variables of the cultural factors were manipulated using three designs and are shown using the Latin Square design to counterbalance order effects [1]. The prototype used in this experiment was developed from scratch by the researchers based on the results of the design analysis. The three websites developed have one user interface design for each culture that maximizes the cultural and genre attributes appropriate for that culture (see Table 2 and Figure 2, 3, and 4). In addition, for each of the interfaces developed design alternative with content that is appropriate for each of the other cultures

being tested is also included. This is done by exploiting the XML technology. XML usually used to display different data across different UI platforms (e.g. Computer UI, mobile interface and others). Here, it is used to display different cultural data into HTML file, and this is based on users’ culture. For example, the website optimized for Kuwait contained a crescent moon image, which is a highly regarded religious symbol. In one of the alternative websites this crescent moon was replaced with a Christmas tree image, in this way we were able to produce a website optimized for Kuwaiti culture that could be meaningfully shown to UK participants (see Figure 2, 3 and 4). This website differed from the website presented in design-C for the UK culture because in that design there were no religious symbols. We thus generated a total of nine website prototypes, three of which are optimized for a given culture as shown in Table 2. The others were non-optimal culturally but were meaningful to one of the cultures.

Table 2: The nine versions of UI designs

Design Culture Design-A Design-B Design-C Kuwait AK BK CK UK AU BU CU Egypt AE BE CE Legend K-Kuwait, U-UK, and E-Egypt

Figure 2. Design AK, religious oriented model presented for

Kuwaiti participants

Figure 3. Design AU, religious oriented model presented for

UK participants


66

Figure 4. Design CU , neutral oriented model presented for

UK participants

The above three figures show design AK and AU, which are presented in Figure 2 and Figure 3 to show how religious symbols, for example, vary across-cultures. Also design CU, which is presented in Figure 4, shows how design-C is a religiously neutral design.

3.1 Variables and Participants 84 user variables are measured in this experiment. Fourteen variables are required to collect participants’ demographic information. Of the remaining 70 variables, 58 are the users’ subjective valuations of interface properties (e.g. text, images, and others) that are thought to have a cultural impact. The remaining 12 variables are used for evaluating each group of tasks (simple and complex tasks). Each group has six variables, of which four measure usability and two measure culture and trust compliance. These six variables are repeated for each task group. These 12 questions are aimed at building a usability factor that can be used to determine: (1) at the high level, the most usable design for each of the studied cultures, and (2) at the lower level, the design markers (DMs) that improve usability from the 58 DMs. The experiments were conducted with 21 participants from each culture (Kuwait, UK, and Egypt). Participants were selected based on their ability to use the computer, internet, speak English and were given financial incentive.

3.2 Procedure and Materials

Figure 5. The procedure of Cross-cultural experiment

(Cross-Use)

The Cross-Use experiment procedure consists of seven stages as shown in Figure 5. In the first stage, participants were informed about the three experimental sessions, objectives and procedure, and were required to sign the consent form. This is followed by the second stage, where each participant receives two 3-digit personal account codes and a password that allows them to run the experiment process and perform the online transactions required. In the third stage, a questionnaire of 28 questions is administered; each question included one or more images of a DM relevant to one of the design claims being investigated. The aim is to obtain an initial understanding of the participants’ expectations before interacting with the e-banking prototype. In the fourth stage (Task performance evaluation), the participant starts to perform six tasks, which are divided into two task groups (simple and complex tasks). Each group contains three tasks, the first three are for information inquiry and the other three are for performing transaction tasks. During tasks performance a think a loud method is applied to determine user perceptions and expectations, and these with user's actions are recorded using Morae™ recording tool. Upon completion of the three tasks, a comparison questionnaire is administered to rank the tasks. After each of the three tasks, participants answer the six design comparison questions, which compare the three designs in terms of usefulness, ease of use, frustration, satisfaction, culturally related issues and the most trustable design. The aim of this stage is to obtain the most usable design and what are the DMs that make a design usable for a particular culture. In the fifth stage, the participants were presented with several design layouts, and transactions processes necessary to explain the question, and were asked design-specific questions to rank several cultural design claims (30 questions presented in a forced-choice comparisons as well as 5-point Likert scale questions). The aim of this stage is to measure users’ experience after their interaction with different interface designs and performing different types of tasks. The final stages are used to wrap-up the experiment by collecting participants demographic data and ending with a thank you message. The experiment uses a Pentium Centrino 1.5 MHz laptop with 15” TFT screen, and regular mouse. The experiment was executed from the local web-server running on the same computer. In addition, a reasonable resolution (320 x 240 pixels) webcam was connected to the computer to record the participants’ facial expressions using Morae™ tasks recording tool (see www.techsmith.com). The Morae™ tool records all users’ actions similar to an actual usability lab. This local and remote evaluation technique allows evaluating designs for more cultures and genres in a wider context with less cost and time. In addition, this tool is able to capture qualitative and quantitative data (see Figure 5) that will enrich the analysis with a concrete evidences that supports cross-cultural design decisions.

http://www.techsmith.com)


67

3.3 Objectives and Hypotheses The objective of the Cross-Use experiment is to substantiate the cultural design claims [9][12][14], which have been substantiated earlier in design evaluations approaches [9]. This experiment further substantiates these claims based on user-in-context evaluation, and aims to provide two types of results. These are related to the user preferences, and usability for the selected design, and design markers. User preferences refer to the results based on a comparison made by the user between two or more UIs or on specific aspects of those designs. In contrast, usability is assessed by performing real tasks, and then both objective (e.g. time to perform a task) and subjective (e.g. satisfaction with task) outcomes are measured. The results of users’ preferences and usability are also useful in deciding whether the design preferences are a good indicator for usability. In order to test these objectives, several analysis methods were conducted, to examine the validity of the following hypotheses:

H1: When given a choice between a website designed for a different target culture and one designed for their own target culture, users will prefer the website designed for their own culture.

H2: Websites that have been designed for a particular target culture (e.g. Kuwait, or Egypt, or UK) using the developed cultural design claims will produce better usability results when tested by members of that particular target culture.

H3: Using Discriminant Analysis (DA), it is possible to identify specific or aggregated DMs that are the main contributors to the observed user preferences and usability improvement.

In this study, the DA and Chi-Square statistical analysis methods were used to analyse the questionnaire data, which involves a 189 observations -- 63 observations for 3 designs. The DA is used to show the most important or interpretive independent variables, which discriminate the dependent variable or affect it [11], while the Chi-square is used to determine whether the groupings of cases on one variable are related to the groupings of cases on another variable [2].

4. The Cross-Use experiment findings The aim of the Cross-Use experiment is to present the important DMs that were identified by users’ preferences, and usability. This can be determined by two analyses, which are concerned with the ability of the developed user interface designs to classify the cultures differently, and the identification of those DMs that play a significant role in causing these differences. The key factors in this analysis are usability and preferences.

4.1 Cross-cultural design preferences Study hypothesis (H1) predicted that when creating designs that are in accordance with cultural design claims [9], these designs are able to generate culturally sensitive designs. The data collected from the experiment were used in this analysis to classify the three cultural groups of users according to their preferences for the identified cultural designs. DA was performed with national culture as the dependent variable, and the DMs as independent variables.

The results of this analysis confirmed hypothesis H1 (see Figure 6 and Table 3). This indicates the ability of the website designs that adopted the cultural design claims to design for different cultures to capture users’ different preferences. The DMs that cause the cultural preference differences among specific national cultures resulting from the above DA test are shown in Table 3.

Table 3. Partial summary table for the user preferences DMs

CA Claim Design markers KU

EG

U K

Related Question

Relationship Metaphors

R6, R7

C16 Religious Metaphors (Design A)

M M L B2a (*)

National Metaphors (Design B)

M H M B2b (*)

Neutral Metaphors design

(Design C)

H H H B2c

Navigation tools

T4 C21 Drop-down Menu (complex

navigation)

H M H A1a (*)

Tree-view (complex

navigation)

L M L A1b

Sense of security

Legend CA is refer to the cultural attribute code identified in the HCI-cultural model [see 10] - Low (L): <2.49; Medium (M)=2.50..3.49; High (H): >3.49 - (*) DM identified to be significant (p<.001) based on both the DA with Univariate ANOVA tests - No sign indicates the DM was significant based on DA (p<.001) but not significant across cultures based on the Univariate ANOVA test (p<.001). - Claim (C16): High racial tendency oriented cultures (relationship) are expected to show high use of religious and/or national symbols in the design more than low racial tendency oriented cultures, which tend to show neutral symbols.

20100-10-20-30-40

Function 1

10

5

0

-5

-10

Func

tion

2

UK

Egyptian

Kuwaiti

Group CentroidUKEgyptianKuwaiti

Participant nationality

Figure 6. Canonical Discriminant Functions plot: visualizing how the two functions discriminate between cultural groups

by plotting the individual scores for the two functions.

4.2 ross-cultural design usability In this section, an investigation of a good representative score for the cultural usability factor is conducted. Then, two


68

types of analysis are performed. The first analysis uses a Chi-square test, and the second uses DA. The first analysis tells whether or not there is a relation between national cultures and design usability. The second analysis helps in classifying designs according to cultural usability and DMs, and identifying the DMs that are used to improve usability for each culture.

4.2.1 Culture and usability relation The aim of this analysis is to identify the design differences affecting usability among the three cultures, based on the usability factor. Here, attempts are made to find if there are any relationships between national cultures and design usability. If there are any, then the DMs that are affecting usability across these cultures are investigated. The study hypothesis (H2) predicts that when creating designs for cultures based on the cultural design claims and design investigation results (presented in [9]), such as design (A) for Kuwait, design (B) for Egypt, and design (C) for UK, such designs are expected to show better usability results by members of those particular cultures in their own cultural designs. Based on study hypothesis (H2), two issues need to be verified: the first issue is in determining whether a relation exists between culture and usability, which was verified using a Chi-Square test. Then, the second issue is determining the usability improvement that occurs frequently within the targeted cultural design, which was verified using a DA test. As for the existence of a relation between the design usability (represented by the usability factor) and the national cultures, the following hypothesis was defined: Hypothesis: There is a relation between national cultures and designs’ usability (dependent)

19.0% 19.0%

38.1%

14.3%

76.2%

61.9%

19.0%

4.8%

47.6%

0.0%

10.0%

20.0%

30.0%

40.0%

50.0%

60.0%

70.0%

80.0%

Kuwaiti Egyptian UK

Pa rticipant Na tionality

Parti

cipa

nt c

hoic

e (u

sabi

lity

fact

or)

Design-ADesign-B

Design-C

Figure 7. The distribution graph for the usability scores according to culture and design

A Chi-squared analysis shows that there is a significant relation between national culture and design usability (χ2=19.08, df = 4, Sig. < 0.001). In Figure 7, certain website designs are found to be more usable by certain national cultures is shown. In validating hypothesis (H2), which

predicted that websites that have been designed for a particular target culture (e.g. Kuwait, or Egypt, or UK) using the cultural design claims will produce better usability results when tested by members of that particular target culture. Figure 7 shows a clear tendency for high usability by Kuwaiti participants in using their cultural design (design-A), but there is an exception to the hypothesis for Egypt and UK. Egyptian participants show high usability in using design-A, while UK participants have a usability score that is split between design-B and design-C. To further investigate the cause of this unexpected result, in the following section, the DA is used to identify which specific variables were affecting usability scores for each of the cultures.

4.2.2 The classification of the three designs using DA test

DA was performed with usability factor as the dependent variable, and the studied CMs (58 variables) as independent variables. This test provides two types of result. The first result is the classification of the three designs (A, B, and C) based on the usability factor for each case study culture to determine the usability level on different designs (see Figure 8, Figure 9, and Figure 10).

100500-50-100

Function 1

20

10

0

-10

Fu

ncti

on 2

Design C

Design B

design A

Group CentroidDesign CDesign Bdesign A

Usability FactorParticipant nationality: Kuwaiti

Figure 8. Visual graphs showing the two functions used

to classify three designs based on usability factor for the Kuwaiti culture

100500-50-100-150-200

Function 1

80

60

40

20

0

-20

Func

tion

2

Design C

Design B

design A


Usability FactorParticipant nationality: Egyptian

Figure 9. Visual graphs showing the two functions used

to classify three designs based on usability factor for the Egyptian culture


69

7550250-25-50

Function 1

60

40

20

0

-20

Func

tion

2

Design C

Design B

design A


Usability FactorParticipant nationality: UK

Figure 10. Visual graphs showing the two functions used to

classify three designs based on usability factor for the UK culture

Table 4: Partial summary table for cultural usability DMs

CA Claim Design marker KU E

G UK

Relationship Metaphors

R6, R7

C16 National Metaphors (Design B)

H†

Navigation tools T4 C21 Drop-down Menu

(complex navigation)

H† H†

Tree-view (complex navigation)

L†

Drop-down field (complex navigation)

H† H†

Free-search (complex navigation)

H†

Legend † This symbol indicates that this DM affects usability for this particular culture (presenting a cultural-usability design). The result of this indicator is determined by performing DA.

The second result is in identifying the DMs, which cause usability improvements among specific national cultures as shown in Table 4. The DA results shows that the total validity of the proposed model is 100% for observations, which indicates that all cases were adequately categorized in all cultures. In addition, the visual graphs produced by the DA [9] show a divergence between the design type centroid points (see Figure 8, Figure 9, and Figure 10), which primarily discriminate between UK, Kuwaiti and Egyptian cultures. However, the design classification based on usability factor across cultures shows that design-A seems not to discriminate between Kuwaiti and Egyptian cultures. This confirms the results shown in Figure 7, which stresses that at the cultural usability level, Kuwaiti and Egyptian participants show some similarities in usable DMs. This

indicates that, based on usability, Kuwait and Egypt could share design-A and that the UK site (design-C) should be redesigned to have cultural DMs from design-B, in addition to design-C DMs. Thus, study hypothesis (H2) is partially confirmed for Kuwaiti culture. However, to be sure of this conclusion we need to look at the DA results in more detail in order to determine which particular design factors were causing these usability effects. This will enable us to determine how to fine-tune the designs and modify the identified cultural design. The specific details of the DMs that affect these changes are identified and discussed in Table 4. As can be seen from the summary DA results shown in Table 4, there is a clear tendency to identify specific DMs that are the main contributors to the observed participants’ usability. Hence, H3 is confirmed for identifying the DMs for usability. This indicates the ability of the DA to identify the DMs that affect usability. These DMs are used as user-in-context based evidence in supporting or contradicting the cultural design claims. Reviewing the complete list of the usability DMs (see [9]) indicates that the shared DMs and cultures based on the cultural usability factor shows that there are more shared cultural usability DMs between Kuwait and Egypt, followed with Kuwait and UK. However, between Egypt and UK, there are no shared DMs. Again this confirms the relation between Kuwaiti and Egyptian cultures discussed earlier in sections 4.1 and 4.2.1. In addition, the DMs related to preferences and usability levels, the analysis shows that the identified DMs for preferences are higher than usability (see [9]). Furthermore, some usability markers appear to be different from preferences related DMs.

5. Extending the Cross-Use evidence-based guidelines with pattern language

The above sections provided evidence-based guidelines that are usable by cross-cultural users. However, there is clear evidence that cross-cultural User-Interface designers are seeking more concrete design examples that capture the user context to guide their work [19]. As a result, pattern language [20]-[24] is adopted in this research to model user experiences and provide a common language between multi-disciplinary design team members, to make available more concrete and situation dependent solutions [25][26]. This can be achieved by integrating pattern language into the Cross-Use tool to capture proven successful design knowledge in terms of the problem context and situation [27]. The Cross-Use tool can be used to support these patterns, and the patterns are delivered to designers, which guide them to design for cross-cultural user interfaces. In the following section, the adopted Pattern Language model is described and presents the way integrated with the Cross-Use experiment results. In addition, the developed tool to communicate pattern language to designers is described and a working example is shown.


70

5.1 Cross-cultural User-Interface Design Pattern Template

The use of pattern design differs in different study fields [20]-[22] . However, in HCI, there is consensus on the three main fields of a pattern template. These are: problem, context and solution, without any one of which pattern cannot be constructed [28][23]. The purpose of the pattern language use in HCI is deliver a usable example of user interface design and within this research scope is to deliver usable solutions for multiple cultures design [24]. As a result the developed pattern model described in the next section have 11 fields on the top of the three main fields adopted to meet the objectives of this research. Each of the developed patterns will have the following fourteen fields and is described as follows: Title: Meaningful, easy to remember, and unique name for the pattern. Rank: Pattern design degree of confidence. the Evidence-based Theory ranking system was adopted. This ranking system is based on six levels, where level five has the strongest evidence level and zero has no evidence (see http://www.usability.gov/guidelines/3 for more details). Culture/Country: National culture or country the pattern solution intended to develop or tested for. Genre: Application domain. Problem: The problem statement of the pattern in a question format that needs to be resolved. Context: Description of the context of use, user, task being performed and environment of pattern being applied. Usability Principle: Usability principle used in the UID solution considered by this pattern, to help designer to understand the usability concept used by this pattern [25][26]. Forces: The conflicting interests as opposing. Solutions: Captured design solution that balances the various interests (or forces) in a useful way. Rationale: “This section describes how pattern actually works, why it works, and why it is good” [25] based on usability. (How and Why the usability is improved). Diagram: Illustration of the solution using a diagram or any illustration tools. Empirical result and test scenarios: The test scenario process of the empirical study and its results are reported in this field, which will provide with a rationale method describing how the decision of the solution was made. Known uses (Examples): This section show the pattern has been used successfully in an application. Result Context: show pattern links with other patterns that will solve sub-problems related the one in the same pattern. The design pattern framework is adopted to develop the tool that will help cross-cultural designers. This is presented in the following section, where the structure of the Cross-Use tool is shown and described.

5.2 The structure of the Cross-Use tool The Cross-Use tool is based on the Compendium tool (See www.compendiuminstitute.org for more information about

the tool). Compendium was developed to support the QOC method [29]. Compendium is designed to help capture actual design questions (Q) asked during design meetings. This design question is associated with options (O) of alternative solutions to the design question. Finally, the Criteria (C) are generated, which represents the strengths and weaknesses of each option based on a set of criteria (Dix et al., 1998, 2003). The Compendium tool represents the QOC concepts using a graphical representation of typed nodes and arcs [30][31]. In some cases, there are designs that show different evidence. This leads to an argument node, which extends the QOC concept. This QOC concept is adopted to link culture and design variables. This link is represented in the form of questions that show different design options arising from the cultural design claims. These claims generate different design options depending on the cultural differences described by the claim. For every culture, the CP study [9] provides the initial evidence in support of the various design options. Then, several CM , Website Audit (WA) [9], and Cross-Use experiment (see section 3) provide additional evidence which is added to the visual representation (see Table A-1 in appendix A). The evidence can support (Pros) or oppose (Cons) the initial design claim. Furthermore, on some occasions, different results show contradictions, and in this case, the contradiction is represented using the argument node. The outcome of the Cross-Use tool is presented in a map shown in Figure 10. This map shows the visual representation of the Culture-Centred Design (CCD) method [9][14] used to link the concepts of culture and design. This map presents an example that is used here to illustrate the tool’s structure. In Figure 10, the first level presents the culture attributes concepts of this research. Each culture attribute has relations with user interface design aspects of the UI language, Metaphors, Mental Models, Navigation, Interaction and Appearance as shown in level 2. This relation is presented in the form of a design claim such as the C16 node. Each design claim shows the design options for each of the case study cultures as shown in level 3, where the expected culture design is indicated by the link supported by the CP results. For example, for Kuwait design option 1 is the predicted design as indicated from the CP, while design option 2 is the predicted design for UK. In level 4, the results of the studies conducted in this research are represented by three types of evidence. These are: pros indicated by ‘+’, cons indicated by ‘-‘, and argument indicated by ‘+/-‘. The pros indicator represents a response in favour of a design option, the cons indicator represents a response against a design option, and finally argument represents a general argument, usually in response to a design option. These results represent the supporting or opposing results to various design options provided.

http://www.usability.gov/guidelines/3

http://www.compendiuminstitute.org


71

Figure 10. The visual map showing the relation between culture and design concepts (This figure is for illustration only and not based on real data)

As shown in Figure 10, the CA and UI design aspect, and how these two factors affect each national culture are represented as question nodes as follows: Q1. CA node question: “How does the Religious Relationship Symbols (R6) CA affect the User-Artefact relationship for the selected design or function?” Q2. Metaphor node question: “What is the effect of the Religious Relationship Symbols (R6) CA on the selection of the design metaphor (text or images)?” Q3. Culture (KU) node question: “What is the effect of the religious metaphors on the Kuwaiti culture users?” Then, the possible answer to Q3 will provide different design options for design claim C16. For example, the Kuwaiti (KU) culture in level 4 shows design option 1 representing the expected solution for claim C16 as indicated by the CP results. In this example, the CM and WA results show some argument in which some results appear to be in favour of the selected design, but some others are against the selected design. This argument needs more robust evidence on which design should be chosen. Here, the Cross-Use experiment (Exp.) results present this robust evidence, which in this example supports the expected design indicated by the CP. This support makes this design claim become an evidence-based guideline for the Kuwaiti culture based on the guideline process. However, if the evidence contradicts the design claims, the guidelines are compared with the existing websites to identify their shortcomings. These are then represented as design recommendations. Figure 10 shows another scenario for the UK culture, where design option 2 is the expected design. However, the design

evaluation of CM and WA and the user-in-context evaluation are all supporting design option 3, these results are shown as opposing evidence presented by the cons sign to the predicted design option. In this case, design option 3 is adopted to develop the guideline and redesign the original claim. The new claim becomes C16* as shown in Table A-1 by the redesign claim arrow. Figure 10 also shows that each design claim as shown by C16 could be linked to many CAs. The map presented in Figure 10 shows the basic structure of the developed tool, which is presented here as one hierarchal structure level to help illustrate the example. However, the actual structure of the map comprises many maps that are hyper-linked. The following section shows a real example of the developed tool.

5.2.1 The Cross-Use Tool Example The developed tool presented in the previous section presents the culture and design relation in a visual representation using the Compendium tool. An example of the developed tool is presented diagrammatically in Figure 11, 12, 13 and 14, which shows the visual representations for the Cross-cultural design tool, with the results of the CP, CM, WA and Cross-Use data.

Figure 11. The high level visual representation of

Authoritativeness (R12) CA for the case


72

Figure 12. Visual representation for cultural attribute R12

and claim C35 detailed level for Kuwait culture

Figure 13. Visual representation for cultural attribute R12

and claim C35 detailed level for UK culture

Figures 11, 12, and 13 present a worked example of claim C35 that relates to the Authoritativeness (R12) CA identified in CIDM, which demonstrates the relationship between this CA and metaphor design aspect. Figure 1 represents the high level question for the CA (R14). This question is applicable to all three cultures. In Figure 12 and Figure 13, detailed levels of the cultural design relation are presented, which shows how Authoritativeness affects design based on claim C35 [9]. Claim C35 affects the Mental Model (MM) design aspect, which suggested two design options. The first option was applicable to Kuwait culture (see the rectangle in Figure 12), which indicates high adherence to traditional designs, and the second option was applicable to UK culture (see the rectangle in Figure 13), which indicates high adherence to novel or innovative designs. These two design options at this level are supported with CP and literature references as shown by using a book metaphor inside the rectangle in Figure 12 and Figure 13. The development of the WADT (see Table 5) and the results presented in the above example show the ability of the CIDM when placed at the centre of the evaluation process to identify cultural design differences, which confirms the prediction of H1.

Table 5: Partial example of the Website Audit Design Template (WADT)

Cultural Attributes

Associated Design Claim(s)

Design Marker Testing Method Adopted design layout

Compare design layout in artifacts design with different design layout of different genre types to determine what the adopted design layout type is

Authoritativeness (R12)

C35

Genre design consistency

Search for the consistency of studied culture artifacts with the genre design markers identified in cultural markers study [9]

Conducting the previous steps completes the design of the Cross-Use tool, which at this level presents the concepts of culture and design, the links between the concepts through maps, the design claims with initial hypothesis, different results to support or oppose claims, and the decision making related to the context-of-use. Finally, using the publishing technique of the Compendium tool, the developed Cross-Use tool could be generated in the form of web-based design guidelines presented as circles in Figure 14 .


73

Figure 14. Snapshot of the Cross-Use tool in the form of

maps (shows the design claim C16 for the Egyptian culture) The Cross-Use tool website is in process to be published. This website presents the method, concepts, process and the presentation of the Cross-Use tool. The final tool is currently finalized and will be given to Cross-cultural designers for evaluation. In the future research, the evaluation of the tool will be reported.

6. Discussion and conclusion The Cross-Use data analysis was presented through two models. The first model is the cultural preferences model, which consists of the high level classification and DMs of cultural preferences (as shown in Section 4.1), and the second model is the cultural usability model, which consists of the high level classification and DMs of cultural usability (as shown in Section 4.2). Both models have different concepts that require various analysis techniques, which produce diverse results and significance levels. The cultural preferences model concept was to identify whether the participants’ preferences for using the three designs are different, where the experiment shows there are significant differences. This proves that the experiment designs were able to classify cultures based on participants’ preferences for the DMs, which at one level substantiates the experiment design and on the other level shows that there are cultural design differences. In addition, this model shows that a high number of the identified DMs are culturally preferred, which indicates that most of the DMs can be differentiated based on participants’ preferences. The next challenge here was to see whether the usage of culturally preferred DMs in local designs improves local design usability. This led to the development of the second model, which covers usability and was referred to as the cultural usability model. The cultural usability model was developed based on how the user performs the assigned six

tasks (see Section 2 and 3.1), where the usability factor was developed to discriminate between the studied cultures. Based on this model, several issues were identified. The first issue shows that there is a high relation between culture and design usability using the three designs. This indicates that the three designs were able to identify a relation between culture and usability, which shows that at the classification level culture preferences are able to make usable designs. However, based on the most usable design related to culture, the results show that the Egyptian culture reflects design-A as the most usable design compared to the earlier expectation, which is design-B. In addition, the UK participants shared both design-C and design-B as they are the most usable designs (as shown in Figure 7). Therefore, the cultural DMs based on usability are not the same as the cultural design claims. These findings motivate the investigation of cultural usability DM. Earlier, design preferences and usability were discussed to determine their differences. Then, during the experiment evaluation, these two issues were tested using a process to evaluate users. The question here is whether the websites that have been designed based on user cultural preferences are necessarily presenting usable design. The answer to this question helps in recognizing the sensitivity of the approach in collecting data that provides results to help in delivering usable design. The study of Evers and Day [3] uses the culturally extended Technology Acceptance Model (TAM), which uses the usability variables such as usefulness, ease of use, and satisfaction to determine the UI acceptance. They use questionnaires to collect users’ preferences. Their study indicates that design preferences affect interface acceptance across cultures. In the Cross-Use experiment, the general view of the design classification based on the usability factor for each culture shows higher differences on cultural preferences than usability (see [9]). This proves that participants prefer design differently, but when they use the design, it shows more differences in usability than originally expected. This highlights the complementary usage of the user-in-context evaluation in determining the usable cultural DMs. Many website developers and evaluators use methods that assess user preferences aiming to create usable design. For example, the Cultural Markers [5], Website Audit [8], and user evaluation [10] using questionnaire based tools only are not sufficient in understanding and identifying the appropriate usability requirements. According to the results of Cross-Use experiment, as can be seen from Table 3, which presents user preferences CMs, and Table 4, which presents usability CMs, the comparison between the two markers indicates that the number of the identified markers in each type is different, and the identified markers based on preferences are not necessarily identified based on usability and vice-versa. The cultural usability model identifies fewer DMs than in the cultural preferences model. These prove that not all of the preferred DMs are necessarily usable DMs. Furthermore, the cultural usability DMs show that there are some DMs that are not shown to be preferred by the participants but are statistically proven to improve usability (e.g. Tree-view navigation DM in claim C21, as


74

shown in Table 3 and Table 4). This suggests that research based on design preferences does not necessarily present the effects of usability as indicated by Constantine and Lockwood [4]. As a consequence, the results of such studies linking participants’ preferences to design can be doubted, and this also affects the investigation of existing website design, as both adopt the same results. Therefore, the results obtained from users’ preferences and usability should scale differently in supporting cultural design claims and in the later stages of the development of cultural design guidelines. This conclusion strengthens the research results as they are obtained by evaluating both the cultural preferences and usability DMs. For the future research a detailed inspection method are expected to be used to analyse these results together with results of earlier research studies, which aims at developing evidence-based cultural design tool . The final step to determine how the various forms of evidence relating to cultural usability can be made more accessible to designers. The Pattern Language framework was developed with hyper-link patterns. Currently, Compendium tool was used as the platform for the Cross-Use tool because it is freely available on the internet and it provides all of the features we required to capture and browse the hyper-linked structure of the CIDM and the developed patterns. The tool is in its early stages of development to encapsulate the results of our studies of the Kuwaiti, Egyptian and UK cultures, and further work is required to make it more mature.

References [1] S. MacKenzie (2002). "Research Note: Within-subjects

vs. Between-subjects Designs:Which to Use?" [http://www.yorku.ca/mack/RN-Counterbalancing.html]. Toronto, Ontario, Canada, 2002.

[2] J. Pallant, "SPSS Survival manual," New York, USA, McGraw, 2005.

[3] V. Evers and D. Day, "The Role of Culture in Interface Acceptance." Human-Computer Interaction, Interact'97, London, 1997.

[4] L. Constantine and L. Lockwood, "Software for Use: A Practical Guide to the Models and Methods of Usage-Centred Design," New York, Addison-Wesley, 1999.

[5] W. Barber and A. Badre, Culturability: "The Merging of culture and usability," proceedings of the 4th Conference of Human Factors and the Web, 1998.

[6] A. Marcus, "User Interface Design and Culture. Usability and Internationalization of Information Technology," N. Aykin. Mahwah, New Jersey, Lawrence Erlbaum Associates, Inc, 2005, pp 51-78.

[7] P. Bourges-Waldegg and S. A. R. Scrivener, "Applying and Testing an Approach to Design for Culturally Diverse User Groups." Interacting with Computers, vol. 13, pp 111-126, 2000.

[8] A. Smith, L. Dunckley, et al., "A process model for developing usable cross-cultural websites," Interacting with computers Vol. 16, pp 63-91, 2004.

[9] J. Alostath, "Culture-Centred Design: Integrating Culture into Human-Computer Interaction," Doctoral Thesis, The University of York, UK, 2006.

[10] V. Evers, "Cultural Aspects of User Interface Understanding: An Empirical Evaluation of an E-Learning website by International User Groups," Doctoral Thesis, the Open University, 2001.

[11] N. Brace, R. Kemp, and S. Rosemary. "SPSS for psychologists: A guide to data analysis using SPSS for Windows," Mahwah, N.J., L. Erlbaum Associates: vii, p. 287, 2003.

[12] M. B. Rosson and J. M. Carroll, "Usability Engineering: Scenario-Based Developement of Human-Computer Interaction," USA, 2002.

[13] J. Alostath, J. and P. Wright, "Integrating Cultural Models into Human-Interaction Design," Conference on Information Technology in Asia (CITA2005), Kuching, Sarawak, Malaysia, 2005.

[14] J. Alostad and A. Khalfan, "Cross-Use: Cross-Cultural Usability User Evaluation-in-context," 12th International Conference, HCI International 2007. Beijing, China, July 22-27, 2007.

[15] E. T. Hall, "Beyond Culture," New York, An Anchor Book, 1989.

[16] D. A. Victor, "International business Communications," New York, NY, HarperCollins, 1992.

[17] G. H. Hofstede, "Cultures and organizations: software of the mind," New York, McGraw-Hill, 1997.

[18] A. Trompenaars and C. Hampden-Turner, "Riding the waves culture understanding cultural diversity in Business," London, Nicholas Brealey Publishing, 2001.

[19] S. Henninger,. "A methodology and tools for applying context-specific usability guidelines to interface design," Interacting with computers vol. 12, pp 225-243, 2000.

[20] C. Alexander, "A Timeless Way of Building," New York, Oxford University Press, 1979.

[21] C. Alexander, S. Ishikawa, M. Jacobson, I. Fiksdahl-King, and S. A. Angel, "A Pattern Language," New York, 1977.

[22] C. Lawson and S. Minocha, "Guidelines versus Design Patterns for Cultural Localisation," Proceeding of the Second British Computer Society HCI and Culture Workshop on "Culture and HCI: Bridging Cultural and Digital Divides", University of Greenwich, 2003.

[23] D. K. Van Duyne, J. A. Landay, and J. Hong, "The Design of Sites: Principles, and Processes and Patterns for Crafting a Customer-Centred Web Experience," Addison Wesley. 2002.

[24] M. J. Mahemoff and L. J. Johnston, "The Planet pattern Language for Software Internationalisation," PLoP1999, 1999.

[25] M. V. Welie and H. Taetteberg, "Interaction Patterns in User Interfaces," PLoP2000 Conference, 2000.

[26] M. V. Welie, G. C. V. d. Veer, and A. Eliens, "Patterns as Tools for User Interface Design," International Workshop on Tools for Working with Guidelines, pp. 313-324, 7-8, Biarritz, France, 2000.

http://www.yorku.ca/mack/RN


75

[27] J. Alostath and P. Wright. "Pattern Language: Towards A tool for Cross-Cultural User Interface Development," Conference on Information Science, Technology Management, CISTM2004, Alexandria, Egypt, 2004.

[28] L. Barfield, W. V. Burgsteden, et al, "Interaction Design at the Utrecht School of the Arts," SIGCHI Bulletin, 1994

[29] A. MacLean, R. M. Young, V. M. Bellotti, and T. P. Moran, "Questions, Options, and Criteria: Elements of Design Space Analysis. Design Rationale: Concepts, Techniques and Use," T. P. Moran and J. M. Carroll. New Jersy, Lawrence Erlbaum, 1996.

[30] B. Shum, "Hypermedia Support for Argumentation-Based Rationale: 15 Years on from gIBIS and QOC," Rationale Management in Software Engineering. e. a. Dutoit. Berlin, Springer-Verlag, pp. 111-132, 2006.

[31] Shum, S. B. and N. Hammond, "Argumentation-Based Design Rationale: What Use at What Cost?" International Journal of Human-Computer Studies vol. 40(4), pp. 603-652, 1994.

Authors Profile

Jasem M. Alostad received the B.S. degree in Computer Science from Western Kentucky University in 1990, M.S. degree in Software Engineering from Manmouth University in 1996 and PhD degree in Human-Computer Interaction (Computer Science) from The University of York (UK) in 2007. Currently, he is assistant professor in the Computer Science Department at the College of Business

Studies (PAAET). His research interest is in cross-cultural user interface design and usability.

Laila R. Abduihadi Abdullah received the B.S. degree in Computer Science from Richmond College, London ,UK in 1991 and M.S. degree in Computer Information Systems from Bradely University, US in 1995. Currently, she is lecturer in the Computer Science Department at the College of Business Studies (PAAET). Her research interest is in Adaptive Assessment Systems in Learner Centered

Education, and Human-Computer Interaction focusing on usability issues.


76

Appendix A Table A-1. Claims and guidelines development process results for claim C16

1 Design Claim C16 CA R6, R7 2 Design Marker Artefact initial image Experiment Questions B2a, B2b, and B2c Design (A) Show religious metaphors (B2a) 3 Design (B) Show national metaphors (B2b) Design (C) Neutral design (do not present any religious or national metaphors) (B2c) 4 Cultural

Profile (CP) Hypothesis: High racial tendency cultures (e.g. Kuwait and Egypt) are expected to show higher use of relationship symbols than in lower racial tendency cultures (e.g. UK). Expectations: Based on religious metaphors cultural attribute (R6) the Kuwaiti and Egyptian participants are expected to show high preferences for religious metaphors (see design-A), and UK participants are expected to show low preferences for religious metaphors (see design-C). In the national metaphors related CA (R7), Egyptian participants are expected to show high preferences for national metaphors (see design-B), and UK and Kuwaitis are expected to show medium preferences for national symbols (see design-B).

5 Cultural Markers

(CM)

The cultural markers study shows high use of religious and national metaphors in the Kuwaiti websites, low in Egyptian and rarely used in UK websites. Note that in this study the two items of religious and national metaphors are treated as one design marker called religious and national design marker.

6 Website Audit (WA)

KU: Frequent use of religious symbols and medium use of national symbols EG: Frequent use of national symbols with less use of religious symbols. UK: Infrequent use of religious or national symbols.

7 Cross-Use Experiment

Evaluating participants according to C16 scenario shows the following results:

Design A* Design B* Design C KU 3.14 3.28 (UK, EG) 4.04 EG 2.71 3.61 (KU) † 4.29 UK 1.23 2.47 (KU) 4.52

* identified to be significant (P<.001) based on a DA with Univariate ANOVA test based on user preferences † indicates the use of the design marker in the associated culture and genre (e-banking) shows a usability improvement (See Table 8-6).

8 Decision Making

Comparing the experiment results with CP, CM and WA shows:

Religious symbols CP CM WA Exp KU û (H) û (H) ü (M) M

EG û (H) û (L) û (N/A) M

UK ü (L) ü (N/A) ü (N/A) L ü Conforms with experiment result û Contradicts experiment result

National symbols CP CM WA Exp

. KU ü (M) -- ü (M) M

EG ü (H) -- ü (H) H

UK ü (M) -- û (N/A) M ü Conforms with experiment result û Contradicts experiment result

9 Redesign Claim C16*

- Change the use of Religious metaphors to medium for both KU and EG national cultures.

10 Design Recommendations

- UK websites should include a medium level of national symbols (or metaphors). - A design recommendation to KU and EG designers to medium use of religious

metaphors. Local Design View Global Design View

11 Religious Metaphors: KU: Medium focus on religious metaphors (design-A) EG: Medium focus on religious metaphors (design-B) UK: Low (or no use) avoid using religious metaphors as possible (design-C) National Metaphors: EG: High focus on national metaphors (Design-B†) KU: Medium focus on national metaphors (Design-A) UK: Low focus on national metaphors (Design-C)

Design-C

Legend † This symbol indicates the use of the related design marker in the associated culture and genre (e-banking) shows a usability improvement (using DA based on the usability factor) KU: Kuwait; EG: Egypt; UK: United Kingdom; CA: Cultural Attribute; CM: Cultural Markers; WA: Website Audit; Cross-Use: Cross-cultural usability experiment


77

Design and VLSI implementation of Fuzzy Logic Controller

Balwinder Singh 1, Rajneesh Goyal 2, Rakesh Kumar 3 and R.P.P Singh 4

1,3Centre for Development of Advanced Computing (CDAC), Mohali, India (A scientific Society of Ministry of Comm. &Information Technology, Govt.of India)

[email protected]

2Electronics and Communication Department, Jagannath Gupta Institute of Engg. & Technology, Jaipur, India

2Electronics and Communication Department, Sri Sai institute of Engineering and technology, Pathankot India

Abstract: Fuzzy systems implemented in hardware can operate with much higher performance than software implementations on standard micro controllers. Implementation of a fuzzy logic controller on an FPGA using VHDL for motor control is presented in this paper. The hardware implementation of fuzzy logic controller (FLC) on FPGA is very important because of the increasing number of fuzzy applications requiring highly parallel and high speed fuzzy processing. This paper describes the hardware implementation of two inputs (error and change in error), one output fuzzy logic controller using VHDL. Keywords: Fuzzy Logic Controller, Hardware implementation, VHDL, Field Programmable Gate Array.

1. Introduction Fuzzy logic is a problem-solving- control system methodology that lends itself to implementation in systems ranging from simple, small, embedded micro- controllers to large, networked, multi-channel PC based data acquisition and control systems. It can be implemented in hardware, software, or a combination of both. The hardware/software implementation on traditional processors takes longer processing time, while hardware implementations by dedicated. Fuzzy processors run faster and exhibit more compactness. [1] The advantage of the fuzzy logic control is that we don't need to measure the parameters of the model. Hence, many complex systems can be controlled without knowing the exact mathematical model of the plant. In the conventional based design there are five stages used but in fuzzy based design there are three stages used. So design development cycle is reduced and time to market of fuzzy design system is fast then conventional design shown in figure 1. Faster Fuzzy Logic reduces the design development cycle, simplifies design complexity, improves time to market and also improves control performance that

simplifies implementation and reduces hardware costs. [8, 9].

Figure 1. Comparison of Conventional and Fuzzy Design In section II, we will discuss the some prior work related to fuzzy logic controller, section III will give the theoretical concepts of fuzzy logic controllers, the discuss the stepper motors membership functions. Finally in section V will give a discussion on the simulation and synthesis results of the fuzzy controlled stepper motor.

Understand physical System and control

Requirement

Develop a linear model of the plant, sensor and Actuator

Determine a simplified controller from control timer

Develop an algorithm for controller

Simulate, debug and implement the design

Understand physical System and control Requirement

Simulate, debug and implement the design

Design the controller Using Fuzzy Rules

Conventional Design Fuzzy Design



78

2. Prior Work Many researchers studying about Fuzzy Logic Controller, which described the different type of methods for fuzzification, rule base, de-fuzzification. Some researchers control the motor using fuzzy logic. S. Singh [4] introduces the position control of a brush less DC servo motor using fuzzy logic controller is implemented on an FPGA. In this controller monitors and maintain voltage, current and torque along with the direction of rotation. Shabiul Islam [6] has described a Fuzzy Logic Controller (FLC) algorithm for designing an autonomous mobile robot controller (MRC). The controller enables the robot to navigate in an unstructured environment and that avoid any encountered obstacles without human intervention. In [7], Daijin Kim proposed a fuzzy logic controller (FLC) on a Reconfigurable field-programmable gate array (FPGA) system. Mort et al [15], describes the modeling and control of a high speed manufacturing machine and the response of the system using a conventional PI controller is also compared with that of a fuzzy logic controller.

3. Fuzzy Logic Controller The Basic block diagram fuzzy logic controller (figure 2) consists of three-sub blocks. Fuzzification, Inference Engine, De-fuzzification. These are sub blocks are explained in this section with help of detailed figure 3 of FLC

Figure 2. Fuzzy Logic Controller

3.1 Fuzzification

The first component in the FLC is the fuzzifier that transforms crisp inputs into a set of membership values in the interval [0, 1] in the corresponding fuzzy sets. The membership function shapes are typically triangular, trapezoidal or exponential.

Figure 3. Fuzzy Logic Controller In this paper, triangular membership function is used. The triangular membership function for error and change of error with 8-bit center points in the interval [00, FF] and with 3-bit fuzzy labels NB (negative big), NM (negative medium), NS (negative small), ZO (zero), PS (positive small), PM (positive medium) and PB (positive big) for error and change in error. (Figure 4 a)

Figure 4(a). MFs for error

Figure 4(b). MFs for change in error For an error input signal e, NS and ZO fuzzy subsets are active and the membership value for the two fuzzy subsets within that region depend on the point at which the input line cut the two slopes. (Figure 4 b)

3.2 Inference Engine The degree of membership is determined in the fuzzification stage. The next step is to create rules to decide what action should be taken in response to the given set of degree of membership function. The process of determining the fuzzy output from the input fuzzy sets is called inference evaluation. The process of inference evaluation is accomplished by using the knowledge base. As inputs are received by the system, the rule base is evaluated. There standard fuzzy operator which can be used to define a rule are “AND”, “OR” and “NOT”. The “AND” and “OR” fuzzy operators are best used for rules with multiple antecedents. MIN-MAX Operator The antecedent (IF X AND Y) blocks test the inputs and produce control actions. The consequent (THEN Z) blocks of some rules are satisfied while others are not. The control actions are combined to form logical sums. These control actions feed into the inference process where each output member function's firing strength (0 to 1) is determined. The inference mechanism, proposed by Mamdani, uses the maximum and minimum as basic operators. The MAX-MIN


79

method tests the magnitudes of each rule and selects the highest one. A rule usually takes the form of IF-THEN statement as follow:

IF x is A AND y is B THEN z is C “AND” is a fuzzy operator, which is the minimum operation between the two antecedents. There is also an implied “OR” operation between successive rules when more than one rule is involved with the same output. The linguistic rule, IF (x1 is A1 AND y1 is B1) OR (x2 is A2 AND y2 is B2) THEN z is C, can be implemented by taking the maximum value of all the results with the same consequent block.[2]

Figure 5. MIN-MAX Evaluator

3.3 De fuzzification The result for the current rule evaluation is then sent to the multiplier and the accumulator l (figure 3) of the de fuizzification block. The accumulator2 adds each of the results from multiplier. The sum in the accumulator2 will be used as the numerator and sum in the accumulator l will be used as the denominator for the division. The multiplier multiplies the result of the min-max evaluator by a factor representing the weighted strength of the ith output membership function. The factor is obtained from the linguistic variables of output membership function. The multiplied result is accumulated in accumulator2. The multiplication is then repeated for the weight of the second output and the result is accumulated and so on. The multiplier part is implemented by using shift and adds technique, the results are accumulated in accumulator2 and the results of rule evaluation are summed using accumulator 1. During the defuzzification process, each fuzzy output is multiplied by its corresponding singleton position. The sum of this product is divided by the sum of all fuzzy output to obtain the result of the final output.

Where wi, is weighted strength of i-th membership function from rule-base stage and zi is the output membership function center points. [4]

4. Fuzzy Logic Controller For Stepper Motor Many applications related to positioning systems are being implemented with stepper motors. It has some applications in Robotics, Computer peripherals, Industrial servo quality drivers and so on. One of the main advantages of stepper motors is the strong relation between electrical pulses and rotation discrete angle steps. The application area for fuzzy control is really wide, there are many possible controller structures, some differing significantly from each other by the number of inputs and outputs, or less significantly by the number of input and output fuzzy sets and their membership functions forms, or by the form of control rules, the type of inference engine, and the method of defuzzification In the proposed FLC, total of 3 outputs of the motor which two of them are 3-bit in size and one of them is 4-bit in size. The 3-bit output is to control the horizontal and vertical motor movement and the 4-bit output is tilt motor movement. Each of the bits represents the degree and the direction of the rotation. There are total number of 17 movement is designed for the output which each of the represented by the output bit. If all the output bits are ‘0’, it represents the home position (Position 1) or the reset position which indicates the original position of the motor. The bit and the movement it represents are shown is the Table1.

Table 1: Direction and Degree of Rotation Each of the position of the output is represented by a singleton

membership function.


80

Figure 6. Rule Matrix

weighted average. This is done by multiplying fuzzy output obtained from the rules evaluation with its corresponding singleton value, then sum of this value is divided by the sum of all fuzzy output obtained from the rules evaluation. The result from this calculation is the final single output, which can be used to control the motor movements.

5. Simulation and Synthesis Result The design of fuzzy logic controller is implemented using VHDL. Synthesis process has performed using Xilinx tools [14] for synthesizing the compiled VHDL design codes into gate level schematics. The VHDL codes are synthesized for converting into RTL view of the FLC architecture as shown in figure 7. The Technology mapping has chosen in this project from Spartan 3E(xc3s1600) with FG484 package and a speed grade of -4. The synthesized schematic is also simulated to ensure the synthesized design functions. The simulation results for the fuzzy controlled stepper motor are shown in Figure 8.

Table2: Device utilization summary

Number of Slices 1610 out of 1920 83%

Number of Slice Flip Flops

1416 out of 3840 36%

Number of 4 input LUTs

1826 out of 3840 47%

Number of bonded IOBs

81 out of 173 46%

Number of MULT18X18s

6 out of 12 50%

Number of GCLKs 1 out of 8 12%

Minimum period: 16.476ns Maximum Frequency: 60.694MHz Total memory usage is 516524 kilobytes

Figure 7. RTL view of FLC

Table 3: Comparison output between PID [4] and Fuzzy

Logic Controller

Figure 8. Simulation results of Fuzzy Logic Controller

Parameter PID Controller

Fuzzy Logic Controller

Rise Time(Tr) 415.38ms 400.00ms

Settling Time (Ts)

538.46ms 100ms

%Overshoot 0% 16.0%


81

6. Conclusion This paper presents an approach for the implementation of a fuzzy logic controller for Stepper motor on an FPGA using VHDL. This paper presents the implementation of a fuzzy logic controller for a Stepper motor on a Xilinx Sparten III FPGA using VHDL. The implementation of the fuzzy logic controller is very straightforward by coding each component of the fuzzy inference system in VHDL according to the design specifications. The design of the FLC is highly flexible as the membership functions and rule base can be easily changed. By simply changing some parameters in the codes and design constraint on the specific synthesis tool, one can experiment with different design circuitry to get the best result in order to satisfy the system requirement. The FLC can also be used for control purposes in other applications.

References [1] Oscar Montiel, Roberto Sep´ulveda, Patricia Melin,

Oscar Castillo, Miguel A. Porta, Iliana Marlene Meza, “Performance of a Simple Tuned Fuzzy Controller and a PID Controller on a DC motor” In Proceedings of the IEEE Symposium on Foundations of Computational Intelligence, pp 531 - 537, 2007

[2] S.P. Joy Vasantha Rani, P. Kanagasabapathy and A. Sathish Kumar, “Digital Fuzzy Logic Controller using VHDL”, In Proceedings of the IEEE Indicon Conference, Chennai pp 463- 466,2005.

[3] PhilipT. Vuong, Asad M. Madni and Jim B. voung. “VHDL Implementation for a Fuzzy Logic Controller” , BEITechnologies,Inc. 2005

[4] Sameep Singh and Kuldip S. Rattan “Implementation of a Fuzzy Logic Controller on an FPGA using VHDL”, IEEE. (2003)

[5] Bohdan Butkiewicz “About Robustness of Fuzzy Logic PD and PID Controller under Changes of Reasoning Methods” , Institute of Electronic Systems, Warsaw University of Technology , 2002

[6] Md. Shabiul Islam, Md. Anwarul Azim, Md. Saukat Jahan, Masuri Othman, “Design and Synthesis of Mobile Robot Controller” using Fuzzy” , IEEE International Conference on Semiconductor Electronics (ICSE 2006), vol.2 Proc., Kuala Lumpur, Malaysia,pp 825-829, 2006

[7] Daijin Kim, “An Implementation of Fuzzy Logic Controller on the Reconfigurable FPGA System”, IEEE transactions on industrial electronics, vol. 47, no. 3, june 2000

[8] A.M.Ibrahim, "Fuzzy logic for Embedded Systems and Applications", Elsevier Science, 2004

[9] Zadeh L. A. “Fuzzy algorithms”, Information and Control Volume 12, Issue 2, February 1968, Pages 94-102

[10] L. A. Zadeh, “The concept of a linguistic variable and its applications to approximate reasoning – I”, Information Sciences, vol. 8, pp. 199-249, 1975.

[11] M. McKenna and B. M.Wilamowski, "Implementing a Fuzzy System on a Field Programmable Gate Array", in Proc. Internat. Joint Conf. Neural Networks, Washington, DC, USA, pp. 189—194, 2001.

[12] T. Hollstein, S.K. Halgamuge and M. Glesner: "Computer-Aided Design of Fuzzy Systems Based on Generic VHDL Specifications", IEEE Trans. on Fuzzy Systems, vol. 4, no. 4, pp.403417, Nov. 1996.

[13] J. Bhasker, VHDL Primer, Addison Wesley Longman Singapore Pvt Ltd, third edition.

[14] Xilinx ISE 9.2i Software Manuals www.xilinx.com/itp/xilinx92/books/manuals.pdf

[15] Mort, N. Abbod, M.F. Linkens, D.A. “Comparative study of fuzzy DC servo motors and stepper motors for mechatronic systems”, IEEE Colloquium on Innovations in Manufacturing Control Through Mechatronics, pages 6/1-6/5, 1995

http://www.xilinx.com/itp/xilinx92/books/manuals.pdf


82

Progressive Image Compression with Ridgelets and Vector Quantization

A.Vasuki1, P.T.Vanathi2

1Department of Electronics and Communication Engineering,

Kumaraguru College of Technology, Coimbatore – 641006. Tamilnadu. India. [email protected]

2Department of Electronics and Communication Engineering,

PSG College of Technology, Coimbatore – 641004. Tamilnadu. India. [email protected]

Abstract: Wavelets are good at isolating discontinuities at edge points, but cannot capture the smoothness of edge segments. The Finite Ridgelet Transform (FRIT) has been proposed to provide a sparse expansion for functions that have line singularities in 2D. It is based on the Finite Radon Transform (FRAT) that maps a line singularity into a point singularity. The Discrete Wavelet Transform (DWT) is applied on the Radon domain to capture the point singularity. The resulting FRIT is orthonormal, non-redundant and invertible. The multilevel FRIT (MLFRIT) is a variation of the conventional FRIT that divides the input image into blocks and applies the orthonormal FRIT on each block to produce approximations and details. This process can be iteratively done on the approximations to produce a subband decomposition similar to DWT. The resulting data structure shows that MLFRIT is suitable for progressive compression of images. Based on MLFRIT and Vector Quantization (VQ), a novel progressive image compression algorithm has been proposed. In this method, the orthonormal MLFRIT is performed on the input images and the resulting ridgelet coefficients are grouped appropriately and subjected to vector quantization. The proposed compression algorithm results have been presented extensively for a two level MLFRIT.

Keywords: Image Compression, Progressive, Ridgelets, Vector Quantization

1. Introduction In the human visual system, the receptive fields in the visual cortex are characterized as being localized, oriented, and bandpass. Therefore, a computationally efficient image representation should be based on a local, directional, multi-resolution expansion. Natural images contain intrinsic geometrical structures that are key features in visual information. The geometry of natural surfaces is complex and multi-scale. Image representation using separable, orthonormal bases are not optimized since they do not account for geometric regularity of images. Moreover, a major challenge in capturing geometry and directionality in images comes from the discrete nature of the data; typically the input is sampled images defined on rectangular grids. Non-linear wavelet transforms are flexible in representing images since they change the filtering directions according to the image features, thus achieving more energy compaction for sharp features. An efficient transform must be able to capture the essential features of the image with

few coefficients. The transform must be shift-invariant and account for geometrical structure in the image. The basis elements should contain elongated shapes with different aspect ratios and they must be oriented along different directions. The representation must have negligible redundancy. Wavelets in two dimensions that are separable are good at isolating the discontinuities at edge points, but cannot capture the smoothness of the edge segments. In addition, wavelets can capture only limited directional information. Therefore, more powerful representations are needed in higher dimensions.

The FRIT has been proposed to overcome the weakness of wavelets in higher dimensions and captures edges in images with few coefficients. The FRIT provides a sparse expansion for functions that have line singularities in 2D. It is based on the FRAT that maps a line singularity into a point singularity. The DWT is applied on the Radon domain to capture the point singularity. The resulting FRIT is orthonormal, non-redundant and invertible. The FRAT gives projection sequences along different directions and the DWT is applied on these projections, leading to ridglet coefficients. The multilevel FRIT is a variation of the FRIT [1] that is suitable for progressive compression of images. MLFRIT divides the input image into blocks and applies the orthonormal FRIT on each block to produce approximations and details. This process can be iteratively done on the coarse image (approximations) to produce a subband decomposition similar to DWT. Based on MLFRIT and VQ [2][3], a novel progressive image compression algorithm has been proposed. In this method, the orthonormal MLFRIT is performed on the images and the resulting ridgelet coefficients are grouped appropriately and subjected to VQ. The proposed compression algorithm results have been presented extensively for a two level MLFRIT. This algorithm can be extended for images with larger size and higher levels of decomposition greater than two. Since, ridgelet compression is inherently suitable for images with straight edge segments it results in sparser representation for those types of images. Ridgelets take advantage of the image geometry, thus paving the path for efficient compression. D.L.Donoho [4] has produced a sparse representation for data compression by a non-linear approximation method




83

using orthornormal ridge functions. The same author [5] has explored the relationship between orthonormal ridgelets and true ridge functions in another paper. M. Do and M. Vetterli [6] have presented a finite implementation of the orthonormal ridgelet transform and numerical results as applied to representation of images and have shown that ridgelets are specially adapted for straight line singularities. In their paper, E.J. Candes and D. L. Donoho [7] present a review of the work on continuous ridgelet transform, ridgelet frames, orthonormal ridgelet bases, ridgelets and edges. The paper by F. Matus and J. Flusser [8] makes simple computational studies of the finite Radon transform and its application to image compression. In the work by A.J.Flesia et al, [9] a large family of orthonormal bases have been constructed using ridgelet packets and they have been applied to images. D.L.Donoho and A.J.Flesia [10] have applied ridgelet transform for digital data. They have made a study of the analysis and synthesis using true ridge functions. E.J.Candes [11] has made a quantitative study of the properties of estimation by finite linear combination of ridgelets. The representation of arbitrary functions as ridge functions has been extensively studied and numerical experiments have been conducted to illustrate their practical performance. Zhen Yao and Nasir Rajpoot [12] have developed a novel content-based image signature for authentication using ridgelet transform that is robust to content-preserving manipulations like compression and allows progressive authentication. A comparative study of image compression using curvelet, ridgelet and wavelet transforms has been done by M.S. Joshi , R.R. Manthalkar and Y.V. Joshi [13]. Section 2 enumerates the basics of finite ridgelet transform and Section 3 deals with the concepts of multilevel FRIT. Section 4 elaborately explains the proposed compression algorithm and discusses the results obtained. Section 5 gives the conclusion and future work.

2. Finite Ridgelet Transform The FRAT is applied on the input image to produce projection sequences along different directions. The DWT is applied on these projections leading to finite ridgelet coefficients. The continuous Radon transform is given by: dxtxxxftR

Rf ))sin()cos(()(),( 21

2−+∫= θθδθ (1)

The FRAT is adapted from the continuous Radon transform for application to discrete images. The FRAT is defined as summations of image pixels over certain sets of lines. Let Zp = {0, 1, 2, ….p-1} be a finite field with modulo-p operations defined on the set; ‘p’ is a prime number. The FRAT of a real function ‘f’ on the finite lattice Zp2 is defined as:

∑==∈ lkLji

f jifp

lkFRATlkr,),(

),(1),(),( (2)

where Lk,l denotes the set of points that make up a line in the 2pZ lattice. FRAT treats the input image as one period of a

periodic image; therefore it exhibits a wrap-around effect. In the FRAT domain, energy is best compacted if the mean

is subtracted from the image before taking the transform.

The factor p

1 is used for normalization. In Euclidean

geometry, a line in the 2pZ plane is uniquely represented by

its slope or direction ‘k’ and its intercept ‘l’. There are (p2+ p) lines defined in this way and every line contains ‘p’ points. Moreover, two lines of different slopes intersect at exactly one point. For any given slope there are ‘p’ parallel lines that completely cover the plane 2

pZ . Therefore, the

FRAT is redundant, as given by the equation below:

∑ ∑ ==−

= ∈

1

0 ),( 20),(1),(

p

l Zji p

jifp

lkr (3)

In each direction, there are only (p-1) independent FRAT coefficients. Those coefficients at (p+1) directions together with the mean value total (p+1)(p-1)+ 1 = p2 independent coefficients in the finite Radon domain. The Finite Back-Projection operator (FBP) is defined as the sum of the Radon coefficients that pass through a given point.

2

),(),(),,(1),(

,

pPlk

r Zjilkrp

jiFBPji

∈∑=∈ (4)

The back propagation operator computes the inverse of FRAT; the transform matrices of FRAT and FBP are transpose of each other.

The FRAT, the basic building block in FRIT, has a wrap-around effect that is eliminated by a special ordering of the coefficients. The FRIT is the application of the DWT on the slices of the FRAT that have been ordered:

dttRtbaCRTR

fbaf ),()(),,(2

, θψθ ∫= (5)

The continuous ridgelet transform of the integrable, bivariate function f(x)is given by:

dxxfxbaCRTR

baf )()(),,(2

,,∫= θψθ (6)

where the ridgelets in 2D are defined from a wavelet-type function in 1D as,

−+

= −

abxxaxba

θθψψ θ

sincos)( 212/1,, (7)

where ‘a’ is the scaling factor and ‘b’ is the shift parameter. The ridgelet function is constant along the lines,

ttanconssinxcosx =+ θθ 21 (8) The separable continuous wavelet transform is given by,

dx)x(f)x()b,b,a,a(CWTR

bbaaf ∫=2 21212121 ψ (9)

where the 2D wavelets are tensor products of 1D wavelets as specified below:

)x()x()x( bababbaa 21 22112121ψψψ = and,

−

= −

abta)t( /

b,a ψψ 21 (10)

The wavelets and ridgelets are related by,

positionlinescale

positionposcale

Ridgelets

Wavelets

−

−

→

→

,

int,

ψ

ψ

Therefore, wavelets are effective for point singularities and ridgelets are effective for line singularities. In 2D, points and lines are related via the Radon transform; thus the


84

wavelet and ridgelet transforms are linked via the Radon transform. The FRIT is suitable for images with straight edges but not optimal for images with curves. It can be implemented with fast algorithms.

3. Multilevel FRIT In the conventional FRIT, the number of directions is nearly equal to the size of the input image and the basis functions have long support extending over the entire image. In MLFRIT, the basis functions have smaller support and the number of directions is also reduced. Let the input image be of size n x n, where n = p1.p2.p3…..pJ.q and the ‘pi’ are prime numbers. The input image of size n x n is divided into non-overlapping subimages of size p1 x p1. The number of subimages obtained is n1 x n1, where n1 = n/p1 and also n1 = p2.p3.…..pJ.q. The FRIT of each subimage produces one mean value and (p1

2 – 1) detail coefficients. The number of mean values produced is equal to the number of subimages; therefore, the mean values form a coarse approximation of the original image of size n1 x n1.

The above decomposition can be applied again on the coarse image (mean values of first decomposition). At the second level, the subimage size is p2 x p2 and the number of subimages produced is n2 x n2, where n2 = p3…..pJ.q or n2 = n1/p2. This decomposition can be done iteratively up to level ‘J’. The basis functions are orthogonal within their block, and also orthogonal with the constant function of the block as well as with other blocks. Therefore, MLFRIT is an orthonormal transform. The MLFRIT coefficients can be grouped based on their scales and directions to obtain a subband-like decomposition. The results of progressive compression with MLFRIT, decomposing the image into two levels is presented in Table 1 for the test image ‘object’.

The results show that level 1 (L1) detail coefficients are required for picking up the straight edges in the images. The level 2 (L2) mean gives the average value of all the blocks at the second level decomposition. The level 2 detail coefficients are responsible for reconstruction of the smooth regions of the image. The PSNR values of the reconstructed images reflect the percentage of coefficients included for the reconstruction and the subband in which the coefficients are present. Including details at level 1 is sufficient for reproducing the edges in the image and details at level 2 for the smooth regions of the image. It is obvious that a small percentage of coefficients are only required to pick up edge information in images. Adding more details to the reconstruction improves the PSNR and visual quality of the image progressively.

4. Proposed Algorithm A progressive image compression algorithm is proposed based on the concept of MLFRIT and VQ. The FRIT is applied on the input image whose size is chosen as 255 x 255, i.e., n = 255. The parameter ‘n’ is factored as, n = p1.p2.q (11)

where p1 = 17, p2 = 5 and q = 3. The input image is divided into subimages of size 17 x 17, with a total of 15 x 15 non-overlapping blocks, as shown in Figure 1.

Figure 1. Input image of size 255x255 divided into

subimages of size 17x17, with a total of 15x15 subimages The FRIT is applied on each 17 x 17 block producing one

mean value and 16 x 18 detail coefficients for each subimage. This first level of decomposition produces a coarse approximation of size 15 x 15 and detail coefficients of size 240 x 270, shown in Figure 2(a) and (b). Within each 17 x 17 block, the directional projections are transformed using a one dimensional wavelet transform, with four levels of decomposition for each direction. The second level of MLFRIT is applied on the coarse approximation by dividing it into blocks of size 5 x 5, shown in Figure 3, with a total of 3 x 3 non-overlapping blocks. The second level decomposition produces an approximation of the original image of size 3 x 3 (one mean value for each block), shown in Figure 4(a). Each block is transformed into 4 x 6 detail coefficients, with a total size of 12 x 18 coefficients for the second level details (Figure 4(b)). The

wavelet decomposition can be done up to a maximum of 2 levels for each block of 5 x 5.

(a) (b)

Figure 2. First Level Decomposition, (a) Approximation coefficients, one mean value for each block, 15x15 values (b) Detail coefficients, 16x18 coefficients for one block,

15x15 detail blocks, 240x270 coefficients

17x17

17x17

… 17x17

17x17

17x17

… 17x17

. . .

. . .

…

. . .

17x17

17x17

… 17x17

…

15

15 240

270

…


85

The coefficients of MLFRIT can be grouped into three: (i) mean values of level 2 decomposition representing the coarse approximation of the input image, (ii) detail coefficients of level 2 and (iii) detail coefficients of level 1. VQ is appliedon each group of coefficients by designing separate codebooks for each of them. The number of images in the training set used for codebook design is 10. In the proposed compression algorithm, the mean values of level 2 are grouped into vectors of size 1 x 1, since it contains the average value of the input image which is the most important information.

The number of coefficients produced is a total of 3 x 3 = 9. The codebook size used for level 2 mean is 64. Four codebooks h ave been designed for level 1 details, one codebook for each subband, as shown in Figure 5. In every block, each subband is divided into three vectors of the respective size. Each codebook has been designed for a size of 1024, requiring 10 bits for the index. Figure 6 shows the level 2 codebook, code vectors of size 1 x 6 in each subband, with two subcodebooks. The subcodebook size is 128 with a total codebook size of 256, requiring 8 bits for each vector. Similarly, the design has been carried out for

Coefficients included

PSNR (dB)

Reconstructed Images

L2 Mean

L2 Details 0% L1 Details 0%

1.93

L2 Mean


2.67

L2 Mean


9.87

L2 Mean


14.76

L2 Mean


21.23

L2 Mean


30.62

L2 Mean


241.27

L2 Mean


25.01

Coefficients included

PSNR (dB)

Reconstructed Images

L2 Mean


30.35

L2 Mean


36.59

L2 Mean


46.41

L2 Mean


6.91

L2 Mean


8.73

L2 Mean


9.66

L2 Mean


9.93

L2 Mean


9.97

Table 1: Progressive Reconstruction of Images in MLFRIT


86

vector sizes of 1 x 3 and 1 x 2. But, for codevector sizes of 1 x 3 and 1 x 2, the number of subcodebooks is 4, with two subcodebooks for each subband. This will lead to an improvement in quality of the reconstructed image as shown in Figure 7.

Figure 3. First Level Approximation, 15x15 block divided

into subblocks of size 5x5, total of 3x3 subblocks (a) (b)

Figure 4. Second Level Decomposition, (a) Approximation coefficients, one mean value for each

block, 3x3 values (b) Detail coefficients, 4x6 coefficients for one block, 3x3 detail blocks, total of 12x18 coefficients

Figure 5. Level 1 codebooks and vectors

Figure 6. Level 2 codebook and vectors

The reconstructed images with level 2 mean and progressively including level 1 details (level 2 details not included) is shown in Figure 8. The results show that level 1 details are sufficient for picking up the edges in images, even though the PSNR values are low. Figure 9 shows the reconstructed images with level 2 mean and details, not including level 1 details. The reconstructed images show that level 2 details are required for the smooth regions of the image. Figure 10 shows the results of progressive image reconstruction including level 2 mean, level 2 details and progressively including level 1 details. Similarly, Figure 11 shows the reconstruction by progressively including level 2 details, with level 2 mean and level 1 details included.

(a) (b) (c) Figure 7. Reconstructed images, codebook size - level 2 mean : 64, level 1 details : 1024 x 4, (a) level 2 details – code vector size = 1x6, codebook size : 256, bpp = 0.4052, PSNR = 17.26 dB (b) level 2 details - code vector size 1x3, codebook size : 512, bpp = 0.4108, PSNR = 22.07 dB (b) level 2 details - code vector size 1x2, codebook size : 1024, bpp = 0.4168, PSNR = 23.12 dB

(a) (b) (b) (d)

Figure 8. Image reconstruction with level 2 mean included, level 2 details not included, subbands of level 1 details progressively included : (a) Subband 4, bpp = 0.1, PSNR = 2.32 dB, (b) Subband 4 & 3, bpp = 0.2, PSNR = 2.45 dB, (c) Subband 4, 3 & 2, bpp = 0.3, PSNR = 2.56 dB, (d) Subband 4, 3, 2 &1, bpp = 0.4, PSNR = 2.63 dB, Codebook for mean of level 2 (size : 64), details of level 2 (size : 256) and details of level 1 (size : 1024x4) are kept constant

5x5

5x5

5x5

5x5

5x5

5x5

5x5

5x5

5x5

3

3

18

12

18

16

8x6 vector

4x6 vector

2x6 vector

1x6 vector

1x6 vector

Codebook 1

Codebook 2

Codebook 3

Codebook 4

Subcodebook 2

Subcodebook 1

6

4

1x6 1x6

1x6

1x6


87

Figure 9. Image reconstruction by progressively including level 2 details, level 1 not included (a) Reconstruction with Mean and Subband 2 of level 2 details, bpp = 0.003, PSNR = 5.39 dB, (b) Reconstruction with Mean, Subband 2 and 1 of level 2 details, bpp = 0.005, PSNR = 9.39 dB, Codebook for mean of level 2 (size : 64), details of level 2 (size : 256)

and details of level 1 (size : 1024x4) are kept constant.

(a) (b) (c) (d)

Figure 10. Image reconstruction with mean and level 2 details included, progressively including subbands of level 1 details, (a) Subband 4, bpp = 0.1, PSNR = 12.29 dB, (b) Subband 4 & 3, bpp = 0.2, PSNR = 13.72 dB, (c) Subband 4, 3 & 2, bpp = 0.3, PSNR = 15.53 dB, (d) Subband 4, 3, 2 &1, bpp = 0.4, PSNR = 17.26 dB, Codebook for mean of level 2 (size : 64), details of level 2 (size : 256) and details of level 1 (size : 1024x4) are kept constant.

(a) (b)

Figure 11. Image reconstruction with mean and level 1 details included, progressively including subbands of level 2 details, (a) Subband 2, bpp = 0.4030, PSNR = 7.15 dB, (b) Subband 2 & 1, bpp = 0.4052, PSNR = 4.30 dB, Codebook for mean of level 2 (size : 64), details of level 2 (size : 256) and details of level 1 (size : 1024x4) are kept constant.

5. Conclusion and Future Work The Multilevel FRIT decomposes the input image into approximations and details iteratively in a manner similar to DWT. Based on MLFRIT and VQ, a novel progressive compression algorithm has been proposed for efficient lossy compression of images. The ridgelet coefficients resulting from different levels of MLFRIT are subjected to vector quantization based on the level and subband in which they

are present. The proposed algorithm results for a two level MLFRIT validates the suitability of the algorithm for progressive compression at low bit rates. This algorithm can be extended to images with larger size and higher levels of decomposition greater than two. The codebook size also determines the quality of the reconstruction. The overhead resulting from the storage required for codebooks is the disadvantage of this method. Since, ridgelet compression is inherently suitable for certain types of images, especially for images with straight edge segments, it results in sparser image representation compared to other transforms. This work can be extended further by applying multistage VQ with channel coding for packet-based transmission of images through communication channels.

References

[1] Minh N. Do, Martin Vetterli, “The Finite Ridgelet Transform for Image Representation”, IEEE Transactions on Image Processing, 12 (1), pp. 16-28, 2003.

[2] Robert M. Gray, “Vector Quantization”, IEEE Acoustics, Speech and Signal Processing Magazine, 1(2), pp. 4-29, 1984. [3] Y. Linde, A. Buzo, R.M. Gray, “An Algorithm for

Vector Quantizer Design”, IEEE Transactions on Communications, COM-28 (1), pp. 84 – 95, 1980.

[4] D.L. Donoho, “Orthonormal Rdigelets and Linear Singularities”, SIAM Journal of Mathematical Analysis, 31 (5), pp.1062 – 1099, 2000.

[5] D.L. Donoho, “Ridge Functions and Orthonormal Ridgelets”, Journal of Approximation Theory, 111 (2), pp. 143 – 179, 2001.

[6] Minh N. Do, Martin Vetterli, “Orthonormal Finite Ridgelet Transform for Image Compression”, In Proceedings of IEEE International Conference on Image Processing (ICIP), pp. 367 – 370, 2000.

[7] E.J. Candes, D.L. Donoho, “Ridgelets: a key to higher-dimensional intermittency”, Phil. Trans. R. Soc. Lond. A., 357(1760), pp.2495-2509, 1999.

[8] F. Matus, J. Flusser, “Image Representations via Finite Radon Transform”, IEEE Transactions on Pattern Analysis and Machine Intelligence, 15 (10), pp. 996 – 1006, 1993.

[9] A.J. Flesia, H. Hel-Or, A. Averbuch, E.J. Candes, R.R. Coifman, D.L. Donoho, Digital Implementation of Ridgelet Packets, Beyond Wavelets, J. Stoeckler and

G.V.Welland eds. Academic Press, 2001. [10] D.L. Donoho, A.J. Flesia, “Digital Ridgelet Transform

Based on True Ridge Functions”, Beyond Wavelets, 10, pp. 1 – 30, 2003.

[11] E.J. Candes, “Ridgelets : Estimating with Ridge Functions”, Annals of Statistics, 31(5), pp. 1561 – 1599, 2003.

[12] Zhen Yao, Nasir Rajpoot, “Radon / Ridgelet Signature For Image Authentication”, In Proceedings of IEEE International Conference on Image Processing (ICIP), pp. 43 – 46, 2004.


88

[13] M.S. Joshi, R.R. Manthalkar, Y.V. Joshi, “Image Compression Using Curvelet, Ridgelet and Wavelet Transform, A Comparative Study”, ICGST International Journal on Graphics, Vision and Image Processing, 8 (3), pp. 25 - 34, 2008.

Authors Profile

A.Vasuki She obtained her B.E. degree in Electronics and Communication Engineering from PSG College of Technology in the year 1989. She obtained her Master’s degree in Applied Electronics from Coimbatore Institute of Technology in 1991. She has published over 25 papers in journals and conferences and her

research interest is in the field of Image Compression. She is currently working as Asst. Professor in the Dept. of ECE, Kumaraguru College of Technology, Coimbatore, India.

P.T.Vanathi She obtained her Bachelor’s degree in Electronics and Communication Engineering and Master’s degree in Computer Science from PSG College of Technology in the year 1985 and 1991 respectively. She has a Ph.D in the area of Speech Coding and has so far published over 30 papers in Journals

and 60 papers in National and International Conferences. Her areas of interest are VLSI Design and Speech Recognition and she has over 23 years of teaching experience. She is currently working as Asst. Professor in the Dept. of ECE, PSG College of Technology, Coimbatore, India.

vol1 no3

Documents

Transcript of vol1 no3