PARAMETER OPTIMIZATION IN DESIGN OF A RECTANGULAR MICROSTRIP PATCH ANTENNA USING ADAPTIVE NEURO-FUZZY INFERENCE SYSTEM TECHNIQUE

Transcription

1 Internatonal Journal on Techncal and Physcal Problems of Engneerng (IJTPE) Publshed by Internatonal Organzaton of IOTPE ISSN IJTPE Journal September 2012 Issue 12 Volume 4 Number 3 Pages PARAMETER OPTIMIZATION IN DESIGN OF A RECTANGULAR MICROSTRIP PATCH ANTENNA USING ADAPTIVE NEURO-FUZZY INFERENCE SYSTEM TECHNIQUE K.V. Rop 1 D.B.O. Kondt 2 H.A. Ouma 3 S.M. Musyok 1 1. Department of Telecommuncaton and Informaton Engneerng, Jomo Kenyatta Unversty of Agrculture and Technology, Narob, Kenya, vkrop@gmal.com, smusyok@yahoo.com 2. Faculty of Engneerng, Multmeda Unversty, Narob, Kenya, onyango_d@yahoo.com 3. Department of Electrcal and Informaton Engneerng, Unversty of Narob, Narob, Kenya, houma@eee.org Abstract- Modern wreless systems are placng greater emphass on antenna desgns for future development n communcaton technology because the antenna s a key element n the overall communcaton system. A Mcrostrp Antenna s well suted for wreless communcaton due to ts lght weght, low volume and low profle planar confguraton whch can be easly conformed to the host surface. In ths paper, an optmzaton method based on adaptve neuro-fuzzy nference system (ANFIS) for determnng the parameters used n the desgn of a rectangular mcrostrp patch antenna s presented. The ANFIS has the advantages of expert knowledge of fuzzy nference system (FIS) and the learnng capablty of artfcal neural network (ANN). By calculatng and optmzng the patch dmensons and the feed pont of a rectangular mcrostrp antenna, ths paper shows that ANFIS produces good results that are n agreement wth Ansoft HFSS 13.0 smulaton results. Keywords: Mcrostrp Antennas (MSAs), Adaptve Neuro-Fuzzy Inference System (ANFIS), Fuzzy Inference System (FIS), Artfcal Neural Networks (ANNs). I. INTRODUCTION The szes and weghts of varous wreless electronc systems (e.g. moble handsets) have rapdly reduced due to the development of modern ntegrated crcut technology. In many wreless communcaton systems, there s a requrement for low profle antennas. These antennas are less obstructve and n addton, snow, ran, or wnd has less effect on ther performance [1]. Mcrostrp patch antennas are an example of low profle antennas. Mcrostrp antennas (MSAs) antennas are used n for example, hgh performance arcraft, spacecraft, satelltes, and mssles, where sze, weght, cost, performance, ease of nstallaton, and aerodynamc profle are constrants. These attractve features have ncreased MSA popularty and applcaton and stmulated greater research effort to understand and mprove ther performance. MSAs however have lmtatons n terms of bandwdth and effcency; all mposed by the very presence of the delectrc substrate [2]. Often, MSAs are also referred to as patch antennas because of the radatng elements (patches) photo-etched onto the delectrc substrate. The radatng patch may be square, rectangular, crcular, trangular, and any other confguraton. In ths paper, the rectangular mcrostrp patch antennas are consdered. In the past, analytcal and numercal methods have been used to desgn mcrostrp patch antennas. The analytcal methods, based on some fundamental smplfyng physcal assumptons regardng the radaton mechansm of antennas, are the most useful for practcal desgns as well as for provdng a good ntutve explanaton of the operaton of MSAs. However, these methods are not sutable for many structures, n partcular, f the thckness of the substrate s sgnfcant. The numercal methods are mathematcally complex, and stll cannot make a practcal antenna desgn feasble wthn a reasonable perod of tme. They also, requre strong background knowledge, and have tme-consumng numercal calculatons whch need very expensve software packages [3]. Recently, many papers have reported varous mproved methods used n desgnng of mcrostrp patch antennas ncludng the use of varous forms of artfcal ntellgence [2]. In ths paper, a method based on the adaptve neuro-fuzzy nference system (ANFIS) s presented to effectvely calculate and optmze the patch dmensons of a coaxal probe fed rectangular mcrostrp patch antenna. Many papers that have been wrtten on the same feld have concentrated on optmzng only one feature (e.g. resonant frequency, patch length/wdth, or feed ponts etc.) n the desgn of MSA. In ths paper, a new artfcal ntellgence based method s presented for calculatng and optmzng the three mportant features n a desgn of an MSA; the patch length, patch wdth, and the feed pont. The fnal results are then valdated wth the smulaton results usng Ansoft HFSS antenna smulaton software. 16

2 II. OVERVIEW OF MICROSTRIP PATCH ANTENNAS In ts most basc form, a mcrostrp patch antenna conssts of a radatng patch on one sde of a delectrc substrate and a ground plane on the other sde as shown n Fgure 1. The bottom surface of a thn delectrc substrate s completely covered wth metallzaton that serves as a ground plane [4]. Fgure 1. Structure of a rectangular mcrostrp patch antenna The essental property for consderaton n any antenna desgn s Drectvty, Gan, Bandwdth, and Effcency. Also, dependng upon ther geometry, MSAs can produce dfferent polarzatons. The common and typcal types of polarzaton nclude the lnear (horzontal or vertcal) where the path of the electrc feld vector s back and forth along a lne and crcular (rght hand or the left hand) polarzaton where the electrc feld vector reman constant n length but rotates around n a crcular path [5]. MSAs can be fed by a varety of methods whch are classfed nto two categores; contactng and noncontactng. The most popular contactng feed technques used are the mcrostrp lne fed and coaxal probe fed, whle the most popular non-contactng feed technques are aperture couplng fed and proxmty couplng fed [6]. Coaxal probe fed (shown n Fgure 2) s the most common feed technque used n the desgn of mcrostrp patch antennas due to ts low spurous radaton, easy fabrcaton, and easy nput mpedance matchng, and was thus used n ths work. Fgure 2. Dagram of a coaxal probe feed technque III. RECTANGULAR MICROSTRIP PATCH ANTENNA DESIGN METHODOLOGY A. Desgn Specfcatons The rectangular mcrostrp antenna s made of a rectangular patch wth dmensons wdth (W) and length (L) over a ground plane wth a substrate thckness (h) and delectrc constant ( ε r ) as shown n Fgure 1. There are numerous substrates that can be used for the desgn of mcrostrp antennas, and ther delectrc constants are usually n the range of 2.2 < ε r < 12. The steps followed n the desgn of rectangular MSAs as dscussed n [1] [3] [4, 7, 11] are as follows; The Patch Wdth (W) for effcent radaton s gven as; vo 2 W = (1) 2fr εr + 1 where, W s the Patch wdth, v o s the speed of lght, f r s the resonant frequency, and ε r s the delectrc constant of the substrate The Effectve Delectrc Constant ( ε reff ) - Due to the frngng and the wave propagaton n the feld lne, an effectve delectrc constant ( ε reff ) must be obtaned from Equaton (2). 1 2 εr + 1 εr 1 h εreff = W (2) where, ε reff s the effectve delectrc constant and h s the heght of the delectrc substrate The Effectve Length ( L eff ) for a gven resonance frequency f r s gven as; c Leff = (3) 2 fr εreff The Length Extenson ( Δ L ) s gven as; W ( εreff + 0.3) h ΔL = 0.412h (4) W ( εreff 0.258) h The Patch Length (L) L = Leff 2ΔL (5) The Bandwdth (BW) ( εr 1) W h BW % = 3.77 *100% 2 ε L (6) λ r o where, λ o s the wavelength n free space. The Feed Co-ordnates - Usng coaxal probe-fed technque, the feed ponts are calculated as; Y = W /2 (7) f L x f = (8) 2 εreff where, Y f and X f are the feed co-ordnates along the patch wdth and length respectvely 17

3 The Plane Ground Dmensons - It has been shown that MSAs produces good results f the sze of the ground plane s greater than the patch dmensons by approxmately sx tmes the substrate thckness all around the perphery [13]. L = 6 h + L (9) g Wg = 6 h + W (10) where, L g and W g are the plane ground dmensons along the patch length and wdth, respectvely. B. Archtecture of Adaptve Neuro-Fuzzy Inference System (ANFIS) ANFIS network s organzed nto two parts lke fuzzy systems. The frst part s the antecedent and the second part s the concluson, and the two are connected together by rules toform a network. The ANFIS archtecture conssts of fve layers namely; fuzzy layer, product layer, normalzed layer, de-fuzzy layer, and summaton (total output) layer as shown n Fgure 3 below. In the fgure, a crcle ndcates a fxed node, whereas a square ndcates an adaptve node [3, 8, 9, 14]. Fgure 3. Archtecture of an ANFIS Assume that the fuzzy nference system under consderaton has two nputs x and y and one output z. Based on a frst-order Sugeno model, a typcal rule set wth two fuzzy f-then rules can be expressed as; Rule 1: If x s A 1 and y s B 1, then f 1 = p 1 x + q 1 y + r 1 (11) Rule 2: If x s A 2 and y s B 2, then f 2 = p 2 x + q 2 y + r 2 (12) where, A 1, B 1, A 2 and B 2 are fuzzy sets, p, q and r ( = 1, 2) are the coeffcents of the frst-order polynomal lnear functons. The fve layer of Fgure 3 are as follows. Layer 1 s the Fuzzy Layer, n whch x and y are the nput of nodes A 1, B 1, and A 2, B 2, respectvely. A 1, B 1, A 2, and B 2 are the lngustc labels used n the fuzzy theory for dvdng the membershp functons. The membershp relatonshp between the output and nput functons of ths layer can be expressed as; O1, = μa ( x), = 1,2 (13) O1, = μ ( y), j = 1,2 (14) j B j where O 1, and O 1, j denote the output functons, whereas μ A and μ B j denote the membershp functons, respectvely. Layer 2 s the Product Layer that conssts of two nodes labeled Prod. The output w 1 and w 2 are the weght functons of the next layer. The output of ths layer (O 2, ) s the product of the nput sgnal and s defned as; O2, = w = μa ( x) μb ( y),, j = 1,2 (15) j Layer 3 s the Normalzed Layer whch calculates the rato of the th rule s frng strength to the sum of the entre rule s frng strengths. The output s denoted as O 3, and s defned n Equaton (16). w O3, = w =, 1,2 w1 w = (16) + 2 Layer 4 s the De-fuzzy Layer whose nodes are adaptve. p, q and r denote the lnear parameters whch are also called consequent parameters of the node. The de-fuzzy relatonshp between the nput and output of ths layer can be defned as Equaton (17), where O 4, denotes the Layer 4 output. O4, = w f = w( px+ q y+ r) (17) Layer 5 s the Total Output Layer whose sngle node s labeled as Σ. The output of ths layer denoted as O 5, s the total of nput sgnals. The results can be expressed as; wf O5, = w f = (18) w Substtutng Equatons (3)-(16) nto Equatons (3)-(18) yelds f = wf 1 1+ w2f2 (19) ANFIS uses Sugeno FIS model. The Sugeno fuzzy model provdes a systematc approach to the generaton of fuzzy rules from a set of nput-output data pars. The ANFIS systems also employs the use of hybrd learnng algorthm by combnng the least-squares method (LSM) and the back-propagaton (BP) algorthm for ts learnng and tranng process. Durng the learnng process, the premse parameters n the Layer 1 and the consequent parameters n the Layer 4 are tuned untl the desred response of the FIS s acheved [7]. The hybrd learnng algorthm s a two-step process. Frst, whle holdng the premse parameters fxed, the functonal sgnals are propagated forward to Layer 4, where the consequent parameters are dentfed by the LSM. Then, the consequent parameters are held fxed whle the error sgnals, the dervatve of the error measure wth respect to each node output, are propagated from the output end to the nput end, and the premse parameters are updated by the standard BP algorthm. Ths process s repeated untl the results are deemed satsfactory or once t reaches a specfed epoch number [10]. The tranng data presented to ANFIS for tranng (estmatng) membershp functon parameters should be fully representatve of the features of the data that the traned FIS s ntended to model. To ensure that the data sets are representatve, the testng data sets are also ncluded n the system. 18

4 C. Applcaton of ANFIS n the Desgn of a Rectangular Mcrostrp Patch Antenna As dscussed above, ANFIS uses a set of data for tranng of ts network. There are two types of data generators (measurements and smulatons) for antenna applcatons. The selecton of a data generator depends on the applcaton and the avalablty of the data generator. In ths paper, the ANFIS model shown n Fgure 3 wth the nputs substrate heght (h), resonant frequency (f r ), and, delectrc constant ( ε r ) and the outputs patch wdth (W t ), patch length (L t ), and the feed pont along the wdth and length (Y f, X f ) respectvely, llustrates how parameters used n the desgn of rectangular mcrostrp patch antenna are optmzed. The ANFIS can smulate and analyze the mappng relaton between the nput and output data through a learnng algorthm so as to optmze the parameters used n desgn of mcrostrp antennas. The tranng and test data sets used n ths work have been obtaned from both smulatons and prevous expermental works whch are documented n varous refereed journals. ε r ε r Fgure 4. ANFIS Model for Desgn of Rectangular MSA As llustrated n Fgure 4, the ANFIS model contans four stages. In the frst stage, resonant frequency, delectrc constant, and substrate heght are used n optmzng the patch wdth (W) of the antenna. 90 and 18 data sets were used for tranng and testng respectvely. The membershp functons (MFs) for the nput varables f r, ε r, and h, are 4, 3, and 3 respectvely. The number of rules s then 36 (4x3x3) and the number of epochs s specfed as 700. In the second stage, the antenna patch length (L) s optmzed. The three nput varables used n frst stage are mantaned wth the addton of the optmzed patch wdth (W t ) as an nput varable, therefore, varables f r, ε r, h, and W t were used as nputs wth L as the output varable to be optmzed. 90 tranng data sets and 16 testng data sets were used n ths stage. The MFs for the nput varables f r, ε r, h, and W t are 4, 2, 2, and 4 respectvely thus, the number of rules s 64 (4x2x2x4) wth the number of teratons specfed as 700. The thrd stage n the ANFIS model s used for optmzng the feed pont (Y f ) along the patch wdth. In ths stage, the number of epochs s specfed as 600 wth 90 testng data sets and 15 testng data sets used. The varables f r, W t, and L t are used as nputs wth the MFs as 3, 4, and 4 respectvely. Ths gves the number of rules as 48 (3x4x4). Fnally, the nput varables f r, W t, and L t are used n optmzng the feed pont X f along the path length of the antenna. Wth 90 testng data sets and 15 testng data used, the number of teratons was specfed as 600. The nput varables f r, W t, and L t were each allocate the MFs values as 3, 4, and 4 respectvely, makng the number of rules as 48 (3x4x4). The nput output relatonshps are llustrated n Fgures 5 and 6. IV. RESULTS AND DISCUSSIONS By usng ANFIS mplemented n MATLAB platform, tranng and testng of data sets was carred out. Also usng Ansoft HFSS software, smulaton was carred out to generate the values of varous rectangular MSA parameters namely patch wdth, patch length, and feed ponts. Fnally ANFIS optmzed data were valdated wth the Ansoft HFSS smulated data and the results tabulated and plotted. Fgure 5. Input output relatonshp for 1st (patch wdth) and 2nd (patch length) ANFIS model, respectvely 19

5 Fgure 6. Input output relatonshp for 3rd (feed pont Y f ) and 4th (feed pont X f ) ANFIS model, respectvely Table 1. A representatve of nput varables, ANFIS output, and HFSS smulated data Input Varables ANFIS Output Ansoft HFSS f r (MHz) E r h (mm) W (mm) L (mm) Y f (mm) X f (mm) W (mm) L (mm) Y f (mm) X f (mm) Fgure 7. ANFIS output values for rectangular MSA wth varous delectrc constants and subtrate heghts 20

6 Table 1 shows a representatve of the data sets obtaned by usng both ANFIS (MATLAB) and Ansoft HFSS software. The Ansoft smulated results therefore valdates that ANFIS calculates and optmzes the parameters used n buldng a rectangular MSA effectvely. The error dfference between the ANFIS and Ansoft HFSS results for the patch wdth (average of ) and patch length (Average of ) s mnmal showng how effectve ANFIS can be n producng accurate results. However, the error dfference wth the feed ponts was sgnfcantly large. From the lterature, there are no exact formulas for calculatng the feed pont, and therefore tral and error method s usually used n locatng the pont wth proper mpedance matchng [11] [12]. Ths mght have therefore caused a bgger error margn for the feed pont locaton. Fgure 7 plots varous optmzed parameter n relaton to the resonant frequency. Varous substrate materals wth varous delectrc constants and heght were used n the tranng and the outputs show that the hgher the frequency (wth other factors held constant) the smaller the antenna sze. Wth the emphass on FR4 substrate materal wth a delectrc constant of 4.3 and substrate heght of 1.6mm at 2GHz, valdaton was carred out by usng Ansoft. Fgure 8. 3D gan dagram Fgures 8 and 10 show the gan of the desgned antenna and the nput mpedance respectvely as generated by Ansof HFSS. The hghest gan pont s n the z-drecton. Fgure 9 shows that the return loss s lowest at about 1.95GHz wth a bandwdth of about 25MHz Return Loss Patch_Antenna_ADKv1 ANSOFT Curve Info db(st(1,1)) Setup1 : Sw eep db(st(1,1)) Freq [GHz] Fgure 9. Return loss dagram Fgure 10. Dagram showng the nput mpedance V. CONCLUSIONS The representatve optmzed parameters computed by usng ANFIS presented n ths paper for rectangular mcrostrp patch antenna are lsted n Table 1 ncludng the smulated data form Ansoft HFSS. From the results, t s clear that ANFIS produces good results n comparson wth HFSS software. It can be clearly seen from Table 1 that most results are n good agreement. The agreement shown n these results supports the valdty of the ANFIS model proposed n ths paper. To further prove the valdty of the proposed ANFIS model, the practcal mplementaton of the modeled antenna s to be carred out. A lot of research n ths feld has been wth the use of artfcal ntellgent technques n calculatng the desgn parameters of varous patch antennas. However, not much of ths has been drected to optmzng the feed pont. The error margn wth the feed pont s a subject of further research amng to reduce the same. It also needs to be emphaszed that better results may be obtaned from the ANFIS ether by choosng dfferent tranng and test data sets from the ones used n ths paper or by supplyng more nput data set values for tranng. Ths work therefore shows that ANFIS beng fast and accurate can be used to effectvely desgn MSAs and other related work. 21

7 REFERENCES [1] I. Sngh, et al, Mcrostrp Patch Antenna and ts Applcatons: A Survey, Int. J. Comp. Tech. Appl., Vol. 2, pp , [2] K. Guney, N. Srkaya, Adaptve Neuro-Fuzzy Inference System for Computng of the Resonant Frequency of Crcular Mcrostrp Antennas, Aces Journal, Vol. 19, No. 3, pp , [3] N. Turker, F. Gunes, T. Yldrm, Artfcal Neural Desgn of Mcrostrp Antennas, Turk J. Elec Engn, Tubtak, Vol. 14, No. 3, pp , [4] C.A. Balans, Antenna Theory - Analyss and Desgn, John Wley & Sons Inc., 2nd Edton, pp , [5] R.A. Saeed, K. Sabra, Desgn of Mcrostrp Antenna for WLAN, Journal of Appled Scences, Vol. 5, No. 1, Asan Network for Scentfc Informaton, pp , [6] B. Mlovanovc, M. Mljc, A. Atanaskovc, Z. Stankovc, Modelng of Patch Antennas usng Neural Networks, Telskks, Serba and Montenegro, IEEE, pp , [7] K. Guney, N. Sarkaya, Adaptve Neuro-Fuzzy Inference Systems for Computaton of the Bandwdth of Electrcally Thn and Thck Rectangular Mcrostrp Antennas, Electrcal Engneerng Journal, Sprnger- Verlag, Vol. 88, pp , [8] N.K. Kasabov, Foundatons of Neural Networks, Fuzzy Systems, and Knowledge Engneerng, The MIT Press Cambrdge, Massachusetts London, England, pp and , [9] J.S. Roger Jang, ANFIS: Adaptve-Network-Based Fuzzy Inference System, IEEE Transactons on Systems, MAN, and Cybernetcs, Vol. 23, No. 3, pp , [10] S.N. Svanandam, S. Sumath, S.N. Deepa, Introducton to Fuzzy Logc usng Matlab, Sprnger- Verlag Berln Hedelberg, pp , [11] Z.I. Dafalla, W.T.Y. Kuan, A.M. Abdel Rahman, S.C. Shudakar, Desgn of a Rectangular Mcrostrp Patch Antenna at 1GHz, RF and Mcrowave Conference, Subang, Selangor, Malaysa, pp , [12] V.V. Thakare, P.K. Snghal, Analyss of Feed Pont Coordnates of a Coaxal Feed Rectangular Mcrostrp Antenna usng MLPFFBP Artfcal Neural Network, ICIT 5th Internatonal Conference on Informaton Technology, pp. 1-6, [13] A.B. Mutara, R. Refant, Rachmansyah, Desgn of Mcrostrp Antenna for Wreless Communcaton at 2.4 GHz, Journal of Theoretcal and Appled Informaton Technology, Vol. 33 No. 2, pp , 30th Nov [14] A. Barat, S.J. Dastgheb, A. Movaghar, I. Attarzadeh, An Effectve Fuzzy Based Algorthm to Detect Faulty Readngs n Long Thn Wreless Sensor Networks, Internatonal Journal on Techncal and Physcal Problems of Engneerng (IJTPE), Issue 10, Vol. 4, No. 1, pp , March [15] B. Behnam, K. Rezapour, A. Nkranjbar, A.D. Taft, Artfcal Intellgent Modelng of the B-Fuel Engne, Internatonal Journal on Techncal and Physcal Problems of Engneerng (IJTPE), Issue 11, Vol. 4, No. 2, pp , June BIOGRAPHIES K. Vctor Rop receved hs B.S.T. n Electroncs Engneerng from Unversty of Eastern Afrca, Baraton n He s currently pursung the M.Sc. degree n Telecommuncaton Engneerng at Jomo Kenyatta Unversty of Agrculture and Technology (JKUAT). He has worked n varous capactes n telecommuncaton ndustry and he currently works as an Electrcal Engneer/Consultant. Hs research nterests are n the area of artfcal ntellgence, smart antennas, and audo Vsual and vdeo conferencng technologes. He s a member of KSEEE (Kenya). Domnc B.O. Kondt was born on July 22, 1950 n Kocha, Homa-Bay County, Kenya. He receved hs M.Eng. degree n Electrcal Engneerng from Tottor Unversty, Japan n 1991 and Ph.D. degree n Electroncs and Computer Engneerng from Indan Insttute of Technology (IIT), Roorkee, n He s currently a Professor and the Drector of Center for Sustanablty and Development Studes, Multmeda Unversty of Kenya (MMU). He s the former Char of Kenya Socety of Electrcal & Electronc Engneers (KSEEE) and the founder Dean of Faculty of Engneerng, MMU. He has authored several papers n refereed journals and conference proceedngs. In 2003, he receved the best paper award n New Delh, Inda, for hs paper publshed n the Insttuton of Electroncs & Telecommuncatons Engneers (IETE) journal. He has been a revewer of Rado Scence Geophyscal Socety Journal (Washngton DC, USA), The New Measurements and Instrumentaton Journal (USA), WSEAS journals and conference proceedngs (Athens, Greece), ENMA conference proceedngs (Blbao, Span), SAIEEE Afrca Research Journal (South Afrca), and JAGST journal (Narob, Kenya). He has been cted n Marqus WHO S WHO n Scence and Engneerng and Cambrdge Bblographc Socety, England. Hs research nterests are n computatonal electromagnetcs, numercal technques for wavegudes, conductng screens and mcro-strp lnes and apertures as pertans to EMC/EMI and bomedcal engneerng. He s a member of IEEE (USA), WSEAS (Greece) and KSEEE (Kenya), and Assocate Member of IEK (Kenya). 22

8 Internatonal Journal on Techncal and Physcal Problems of Engneerng ( IJTPE), Iss. 12, Vol. 4, No. 3, Sep Heywood Absaloms Ouma was born on January 23, He receved hs B.Sc. degree (Hons) n Electrcal Engneerng from Unversty of Narob, Kenya, n 1988, M.Eng. n Telecommuncaton Engneerng from the Unversty of Technology, Sydney (UTS), Australa, n 1993, and Ph.D. n Engneerng from the Kanagawa Insttute of Technology (KAIT), Kanagawa, Japan n He s currently a Senor Lecturer and the Char of Department of Electrcal and Informaton Engneerng at the Unversty of Narob, Kenya. He has publshed several papers n refereed journals and conference proceedngs. Hs research nterests nclude genetc algorthm applcatons n mage processng, real- He s a member of the Insttute of Electrcal and tme mage processng and bometrc mage processng. Electronc Engneers (IEEE), and IEEE Computer Socety. Stephen M. Musyok graduated the D.Eng. degree n Electroncs from Tohoku Unversty, Japan n He s currently a Senor Lecturer and the Char of Department of Telecommuncaton and Informatonn Engneerngg at Jomo Kenyattaa Unversty of Scence and Technology (JKUAT), Kenya. He has worked for Japan Atomc Research Insttute (JAERI) as a Research Fellow, NEC Corporaton (Japan) as a Development Engneer, and Sangkyo Corporaton (Japan) as a Senor Engneer. He has publshed several papers n refereed journals and conference proceedngs. 23