A New Dual Channel Speech Enhancement Approach Based on Accelerated Particle Swarm Optimization (APSO)

Transcription

1 I.J. Intellgent Systems and Applcatons, 24, 4, - Publshed Onlne March 24 n MECS ( DOI:.585/jsa A New Dual Channel Speech Enhancement Approach Based on Accelerated Partcle Swarm Optmzaton (APSO) K.Prajna Dept. of Electroncs and Communcaton Engneerng, Andhra Unversty, Inda E-mal: prajnakunche@yahoo.com G.Sas Bhushan Rao Dept. of Electroncs and Communcaton Engneerng, Andhra Unversty, Inda E-mal:sasgps@gmal.com K.V.V.S.Reddy Dept. of Electroncs and Communcaton Engneerng, Andhra Unversty, Inda E-mal: konalavs@yahoo.com R.Uma Maheswar Dept. of Electroncs and Communcaton Engneerng, Vgnan Insttue of Informaton Technology, Inda Abstract Ths research paper proposes a recently developed new varant of Partcle Swarm Optmzaton (PSO) called Accelerated Partcle Swarm Optmzaton (APSO) n speech enhancement applcaton. Accelerated Partcle Swarm Optmzaton technque s developed by Xn she Yang n 2. APSO s smpler to mplement and t has faster convergence when compared to the standard PSO (SPSO) algorthm. Hence as an alternatve to SPSO based speech enhancement algorthm, APSO s ntroduced to speech enhancement n the present paper. The present study ams to analyze the performance of APSO and to compare t wth estng standard PSO algorthm, n the contet of dual channel speech enhancement. Objectve evaluaton of the proposed method s carred out by usng three objectve measures of speech qualty SNR, Improved SNR, PESQ and one objectve measure of speech ntellgblty FAI. The performance of the algorthm s studed under babble and factory nose envronments. Smulaton result proves that APSO based speech enhancement algorthm s superor to the standard PSO based algorthm wth an mproved speech qualty and ntellgblty measures. Inde Terms Dual Channel Speech Enhancement, Partcle Swarm Optmzaton, Accelerated Partcle Swarm Optmzaton (APSO) I. Introducton The goal of the speech enhancement s to mprove the qualty and/or ntellgblty of the speech degraded by addtve nose. The problem of enhancng speech sgnal whch s degraded by addtve nose has been wdely studed n the past and stll an actve feld of research. Many gradent based algorthms are developed for dual channel speech enhancement. Most commonly used algorthms are the Least Mean Squares (LMS) and Recursve least mean squares (RLS) []-[2]. Stochastc optmzaton based adaptve flterng s an alternatve for gradent based algorthms. Nowadays stochastc and heurstc algorthms are becomng powerful for solvng the nose reducton problems. These optmzaton algorthms are ndependent of system structure and they do not drectly affect the parameter update. Partcle swarm optmzaton s one of the recent algorthms n the class of optmzaton methods [4]. It s nspred by the behavor of swarms of nsects, schools of fsh, flocks of brds, etc. Several studes are carred out on PSO and there ests almost two dozens of PSO varants [5]. The stochastc optmzaton technque of PSO has been studed for adaptve flterng [6]. PSO s ntroduced to speech enhancement applcaton n 2 by Laleh Badr Asl and Vahd Majd [7]. They also studed an mproved partcle swarm optmzaton based speech enhancement algorthm n the same year [8]. Another varant of PSO called aseual reproducton based PSO s studed by L.Badr and M.Geravanchzadeh n 2 [9]. Hybrd Partcle swarm optmzaton s ntroduced to speech enhancement n 2 []. Varous heurstc approaches have been adopted by researchers for dual channel speech enhancement so far, for eample Genetc algorthm []-[2], Partcle Swarm Optmzaton and many of the varants of PSO. Copyrght 24 MECS I.J. Intellgent Systems and Applcatons, 24, 4, -

2 2 A New Dual Channel Speech Enhancement Approach Based on Accelerated Partcle Swarm Optmzaton (APSO) These algorthms are based on the concepts found n nature.they have become feasble as a consequence of growng computatonal power. Some algorthms gve a better soluton than the others.hence, searchng for new heurstc algorthm for adaptve nose cancellaton s an open problem. In standard PSO, ntalzaton of veloctes may requre etra nputs. A smpler varant s the accelerated partcle swarm optmzaton (APSO) [3], whch does not need to use velocty at all and can speed up the convergence n many applcatons. Hence, n the present paper an attempt has been made to study the effectveness of APSO optmzaton technque, n speech enhancement applcaton. Ths work ams to present the APSO algorthm as a better approach to fnd more qualtatve solutons than SPSO for adaptve nose cancellaton. The rest of the paper s organzed as follows; the basc PSO called Standard PSO (SPSO) s eplaned n secton 2. The proposed speech enhancement algorthm s eplaned n secton 3. Secton 4 descrbes the objectve measures used for the evaluaton of the algorthms, secton 5 deals wth the smulaton results and fnally conclusons are gven n secton 6. The phenomenon of lnearly decreasng nerta weght was proposed by Sh, as the followng equaton T t w ( w n w end ) w T end (2) where T s the mamum teratons and t s the current teraton, w and w are the ntal and fnal n nerta weghts respectvely. For many applcatons, to solve the premature convergence n PSO best values for w and w are set as.9 and.4. n end At a new poston durng teratons, the poston vector of a partcle s updated by t t t v t end (3) t s the change of teraton. The velocty where and poston update for a partcle n two dmensonal search space s shown n Fg.. II. Revew of Partcle Swarm Optmzaton In the standard PSO [4], all n partcles nteract and form dfferent trajectores durng the search process. Each partcle has a poston vector, and a velocty vector. The poston of the partcle represents a possble soluton to the optmzaton. The partcles are ntally generated randomly n the search space. At each teraton partcle moves wth two components; a determnstc component and a stochastc component. That s, each partcle s attracted toward the poston of the current global best g and ts own best locaton n hstory, whle at the same tme t has a tendency to move randomly. For a partcle wth poston vector v ts velocty at a new tme step s updated as t t t v w v. t 2 g and velocty () where and 2 are two random vectors drawn from a unform dstrbuton [,]. Here and are called the learnng parameters or acceleraton coeffcents, and w s the nerta weght, used to mantan the momentum of the partcle. Fg. : Illustraton of partcle movement n 2-dmentonal search space III. Proposed Algorthm for Speech Enhancement 3. APSO The movement of a swarmng partcle conssts of two major components: a stochastc component and a determnstc component. Each partcle s attracted towards the poston of the current global best g and ts own best locaton n hstory, whle at the same tme t has a tendency to move randomly. There s a current best for all n partcles at any tme t durng p teratons. The standard partcle swarm optmzaton uses both the current global best g and the ndvdual Copyrght 24 MECS I.J. Intellgent Systems and Applcatons, 24, 4, -

3 A New Dual Channel Speech Enhancement Approach Based on 3 Accelerated Partcle Swarm Optmzaton (APSO) best to update the partcle poston. The reason for usng the ndvdual best s to ncrease the dversty n the search space. However, ths dversty can be acheved by usng some randomness. Subsequently, there s no compellng reason for usng the ndvdual best, unless the optmzaton problem of nterest s hghly nonlnear and multmodal. A smplfed verson whch could accelerate the convergence of the algorthm s to use the global best only. Thus, n the APSO, the velocty vector at teraton t s generated by a smpler formula [4] t v t v n g t (4) where n s a random vector n the nterval of {,}. The update equaton for poston vector s smply t t t v (5) The rate of convergence of algorthm can further be ncreased by updatng the locaton of partcle n a sngle step as follows t t. g n (6) The typcal values for APSO are α from. to.4 and β from. to.7, though α =.2 and β =.5 can be taken as the ntal values for most unmodal objectve functons. It s worth pontng out that the parameters α and β should n general be related to the scales of the ndependent varables and the search doman [4]. 3.2 APSO to Speech Enhancement The structure of the Dual channel speech enhancement system s shown n Fg.2. In dual channel speech enhancement, t s assumed that nosy speech sgnal sn nose sgnal z s present n one channel and the reference rn s present n the second channel. F s the acoustc path between these two sgnals. The transfer functon Fz derved by usng an adaptve flter of the acoustc path s z W z. In the present paper, the adaptve flter W s modeled usng the APSO algorthm. Most of the nonlnear systems are recursve n nature. Hence models for real world systems are better represented as IIR systems. Based on ths consderaton, for our practcal smulaton, we model an IIR flter as an acoustc path between the two channels of a dual channel enhancement system. Fg. 2: Block dagram of dual channel speech enhancement system An nfnte mpulse response (IIR) flter s a recursve flter n whch the present output depends on prevous outputs. The dfference equaton of IIR flter s shown below: L M (7) y n a n b y n where a and b are the coeffcents of the flter and M( L) represents the order of the flter. The transfer functon of the gven as H z A B z z L a z M b z th M order IIR flter s (8) The prmary nput sgnal of the dual channel system d n whch conssts of clean speech sgnal and nose sgnalb n s made avalable to the adaptve flter. The characterstcs of the adaptve flter are then modfed so that the output of the flter y n resembles d n as close as possble. The man complcaton whch IIR flters ntroduce s that they depend nonlnearly on ther coeffcents.ths can be overcome by applyng an optmzaton technque such as Partcle swarm Optmzaton (PSO) or Accelerated PSO (APSO). Usng optmzaton technque several possble collectons of IIR coeffcents are chosen and could see what error each produces. Based on those results, new ponts are chosen to test, and contnue untl all of the ponts have clustered together and swarm n a small area. Input speech sgnals are segmented nto frames. In stochastc optmzaton based speech enhancement t s Copyrght 24 MECS I.J. Intellgent Systems and Applcatons, 24, 4, -

4 4 A New Dual Channel Speech Enhancement Approach Based on Accelerated Partcle Swarm Optmzaton (APSO) requred to defne an objectve functon to evaluate the ftness of each partcle. Here, the objectve functon s defned as the average error between the nosy speech and estmated nose sgnal n each frame. The epresson for the ftness functon J s gven as N J (9) L k 2 d( k) y ( k) where L s length of frame and y (n) s the output agent. Here th of flter W z for W z s desgned by usng APSO. Each partcle n the swarm s consdered as a canddate soluton whch represents a set of coeffcents of adaptve flterw z. After some teratons, W zgves the best soluton when the ftness value s mnmum.e. J s mnmum. The nosy sgnal y n s estmated by convolvng the nose reference r n wth the error sgnal. Then enhanced frame s obtaned by subtractng the estmated nose sgnal from nosy speech. To mplement the APSO n speech enhancement applcaton, t s requred to set some parameters. They are number of teratons ( t ma ), number of partcles ( n p ) and acceleraton constants gven n (6).Flow chart for APSO s gven n Fg.3. Start Set number of teratons, no of partcles, and Generate ntal poston and veloctes randomly for all Fnd global best for t = Whle t < t ma Fnd best n whole searchng process Fnd actual gbest at teraton t For =, < n p, Calculate partcle velocty Proposed algorthm: APSO to Speech Enhancement Step. Create a populaton of partcles; here each partcle s a set of coeffcents of adaptve flterw z. Step 2. Evaluate the objectve functon for each partcle poston by usng equaton (9) Step 3. Fnd the current best locaton Step 4. Move the partcles nto new locatons accordng to the equaton (6) Step 5. Go to step 2 untl stop crtera satsfed End Calculate partcle poston Evaluate objectve functon for the poston Fnd actual poston for each partcle IV. Objectve Measures The objectve measures that are consdered for the evaluaton of the proposed algorthm are Sgnal to Nose Rato (SNR), Perceptual Speech Qualty Measure (PESQ) and Fractonal Artculaton Inde (FAI). 4. SNR SNR s the most common measure of performance that has often been used to evaluate the enhancement algorthms. SNR s defned as [6] Fg. 3: Flow chart of APSO n yn 2 S log () N 2 n where dstorted speech. n s the clean speech and n y s the The dfference between the SNR of enhanced speech and that of nosy speech s the measure of the Improved SNR (SNRI). Copyrght 24 MECS I.J. Intellgent Systems and Applcatons, 24, 4, -

5 Mean Square Error A New Dual Channel Speech Enhancement Approach Based on 5 Accelerated Partcle Swarm Optmzaton (APSO) SNR of enhanced speech sgnal s calculated by equaton (). 4.2 PESQ To compute the PESQ measure, frst the clean sgnal and the degraded sgnals are level equalzed to a standard lstenng level and these sgnals are allowed through a flter havng the response smlar to a standard telephone handset. The tme delays are corrected by algnng the sgnals and processng through an audtory transform lke BSD to obtan the loudness spectra. The dfference n the loudness spectra of clean sgnal and degraded sgnal s computed and averaged over tme and frequency and s termed as dsturbance [5]-6]. As n Hu and P.Lozou [7], PESQ score s computed as a lnear combnaton of the average dsturbance value D and the average asymmetrcal dsturbance values nd A and s gven by PESQ 4.5 adnd a2a nd () where nd a. and a2. 39 PESQ score ranges from.5 to 4.5. Hgher values represent better qualty accordng to the ITU-T Recommendatons P.862 standard. 4.3 FAI Ths measure s based on the prncple that the ntellgblty of speech depends on the proporton of spectral nformaton that s audble to the lstener and s computed by dvdng the spectrum nto 2 bands (contrbutng equally to ntellgblty) and estmatng the weghted average of the sgnal-to-nose ratos (SNRs) n each band. FAI s computed based on the weghted average of the proporton of the nput SNR transmtted by the nose-suppresson algorthm [8] n each band and s gven by FAI where M k W M k k W fsnr k k (2) W k denotes the weghtng functons or bandmportance functons appled to band k and M s the total number of bands used and fsnr denotes the fracton of the nput SNR transmtted by the nose reducton algorthm Table : Smulaton parameters for SPSO Algorthm Parameter Value SPSO N 3 Iteratons 7 k.5.5 w n.9 w end.4 Samples for each frame alpha=.3,beta=.5 alpha=.2,beta=.4 alpha=.2,beta= Iteratons Fg. 4: Effect of acceleraton coeffcents on APSO Copyrght 24 MECS I.J. Intellgent Systems and Applcatons, 24, 4, -

6 6 A New Dual Channel Speech Enhancement Approach Based on Accelerated Partcle Swarm Optmzaton (APSO) Table 2: Epermental condtons for APSO Algorthm Parameter Value APSO n p 3 t ma Samples for each frame 32 APSO than SPSO. Tme doman waveforms of nosy speech, clean speech and the speech sgnals enhanced by APSO and SPSO are shown n Fg.6. From these waveforms t can be clearly notced that the sgnal enhanced usng APSO algorthm resembles close to the clean speech compared to the sgnal enhanced by SPSO. The performance of the proposed algorthm s also studed n terms of ntellgblty by evaluatng the objectve measure called FAI. Graphcal representatons for the mprovement n objectve measures SNRI, PESQ and FAI for babble nose condton are shown n Fg.7 and Fg.8 and Fg.9 respectvely. Results ndcate that the ntellgblty of speech enhanced by APSO s better than the sgnal enhanced by SPSO. V. Smulaton Results Fve clean speech sentences are selected randomly from NOIZEUS database [9] for the smulaton. Nose references are taken from the NOISEX-92 database [2]. The nosy speech sgnal s obtaned by addng the clean speech sgnal to the nose reference modfed by a transfer functon F (z). So, F (z) s the acoustc path between the two nput sgnals. The flter F(z) used n our smulaton s a second order IIR flter and s defned as follow F ( z ) (3) 2.2z.36z The adaptve flter W(z) s gven as W ( z ) 2 p2z p3z p (4) where swarm. p s the j th j dmenson of th agent n the The stochastc optmzaton algorthm called APSO Algorthm s used to determne the weghts of the z adaptve flter W.The nput nosy sgnals are segmented nto frames of 2ms. Each frame conssts of 32 samples. The smulaton parameters chosen for APSO and SPSO are shown n Table and Table 2 respectvely. The effect of dfferent acceleraton coeffcents on convergence of APSO s shown n Fg.4. From ths fgure t s observed that acceleraton coeffcents of =.3 and =.5 gves the better solutons for our problem. Magntude response of acoustc path s shown n Fg.5.The SNR levels of the nput nosy sgnal for babble and factory type nose s set at -, and 5dB. Results are averaged over tral runs. Table 3 gves the comparson for the SPSO algorthm and APSO usng an objectve measure called mproved SNR (SNRI) n babble nose condton. Results show that SNRI s sgnfcantly mproved for Fg. 5: Magntude and frequency response of acoustc path Table 3: Improved SNR of APSO and SPSO Input SNR for babble Nose(dB) 5 - Algorthm Improvement n SNR (db) SPSO 6.6 APSO 9.3 SPSO 8.2 APSO 2.7 SPSO APSO Copyrght 24 MECS I.J. Intellgent Systems and Applcatons, 24, 4, -

7 Improved SNR(dB) FAI Value PESQ score A New Dual Channel Speech Enhancement Approach Based on 7 Accelerated Partcle Swarm Optmzaton (APSO) Input SNR (db) SPSO APSO Fg. 8: Graphcal representaton of mprovement n PESQ score for babble nose nput SPSO APSO..5-5 Input SNR (db) Fg. 9: Graphcal representaton of mprovement n FAI measure for babble nput nose Table 4: Evaluaton of algorthms n factory nose condton Fg. 6: Tme doman waveforms of (a) speech enhanced by SPSO (b) APSO enhanced sgnal (c) clean speech and (d) babble nose reference Input SNR (db) SPSO APSO Fg. 7: Graphcal representaton of mproved SNR under babble nose condton Input SNR for Factory Nose(dB) 5 - Algorthm SNR FAI value PESQ score SPSO APSO SPSO APSO SPSO APSO Copyrght 24 MECS I.J. Intellgent Systems and Applcatons, 24, 4, -

8 Mean Square Error 8 A New Dual Channel Speech Enhancement Approach Based on Accelerated Partcle Swarm Optmzaton (APSO).4.2 APSO SPSO Iteratons Fg. : Convergence of MSE for APSO and SPSO The objectve measures are computed by segmentng the sentences usng 3-ms duraton Hammng wndow wth 75% overlap between adjacent frames. The convergence of SPSO and APSO algorthm s shown n Fg. From ths fgure, t can be clearly notced that APSO fnds more qualtatve soluton than SPSO wth mnmum ftness value (mnmum mean square error for APSO). The spectrograms for speech sgnals are shown n Fg.. The power spectral denstes of the sgnals enhanced by APSO and SPSO are compared n Fg.2. From ths fgure t can be nferred that the sgnal estmated wth APSO based approach s close to clean speech when compared wth the SPSO based approach. Table 4 shows the performance evaluaton of both the algorthms under factory nose condton. From ths table t can be clearly observed that APSO speech enhancement algorthm outperforms than the SPSO based enhancement algorthm. Fg. : Spectrograms of (a) sgnal enhanced by SPSO (b) Enhanced sgnal of APSO (c) Clean speech and (d) Babble nose reference sgnal VI. Concludng Remarks Ths paper deals wth the problem of adaptve nose cancellaton n speech enhancement usng stochastc and heurstc optmzaton strateges. In ths present work, a recently developed stochastc and heurstc optmzaton algorthm called Accelerated partcle Swarm Optmzaton (APSO) s ntroduced to speech enhancement applcaton. Its performance s compared wth another approach whch s based on the well known heurstc algorthm called Standard PSO algorthm. From the smulaton results (Tables 3 and 4) t can be concluded that the performance of APSO algorthm s better when compared to SPSO wth respect to the speech qualty and ntellgblty, as there s an mprovement n the correspondng objectve measures SNR, PESQ and FAI values. Copyrght 24 MECS I.J. Intellgent Systems and Applcatons, 24, 4, -

9 Power/frequency (db/rad/sample) Power/frequency (db/rad/sample) Power/frequency (db/rad/sample) A New Dual Channel Speech Enhancement Approach Based on 9 Accelerated Partcle Swarm Optmzaton (APSO) Welch Power Spectral Densty Estmate From the convergence analyss of SPSO and APSO (Fg.) t s notced that APSO has faster convergence than SPSO and could fnd the best qualty soluton for adaptve flter coeffcents n our problem of actve nose control. Hence, APSO may effectvely enhance the nosy speech by mprovng the qualty of speech compared to the estng SPSO based speech enhancement. It can also be concluded that APSO could reduce the back ground nose dstorton more effectvely than SPSO, as t has mproved ntellgblty and PESQ scores Normalzed Frequency ( rad/sample) (a) Clean Normalzed Frequency ( rad/sample) Welch Power Spectral Densty Estmate (b) APSO Welch Power Spectral Densty Estmate Normalzed Frequency ( rad/sample) (c ) SPSO Fg. 2: Comparson for the power spectral denstes of (a) the clean speech, sgnals enhanced by (a) APSO and (b) SPSO at -db nput SNR References [] B. Wdrow, and S. Stearns, Adaptve Sgnal Processng, Englewood Clffs, NJ: Prentce Hall, 985 [2] T.Ueda, H suzuk, Performance of Equalzers Employng a Re-tranng RLS Algorthm for dgtal moble rado communcatons, 4th IEEE Vehcular Technology Conference, 99, pp [3] P.Mars, J.R.Chen, and R. Nambar, Learnng algorthms: theory and applcatons n sgnal processng, control and communcatons, CRC Press, Boca Raton, FL, 996 [4] R.C.Eberhart, and J.Kennady, A new optmzer usng partcles swarm theory,proceedngs of the Sth Internatonal Symposum on Mcro Machne and Human Scence, Nagoya, Japan, IEEE Press, Pscataway,NJ,995, pp [5] Kruscnsk, D.J. and Jenkns, W.K., Adaptve Flterng Va Partcle Swarm Optmzaton, Proceedngs of 37 th Aslomar Conf on Sgnals, Systems, and Computers, November 23. [6] D. J. Krusensk and W. K. Jenkns, Desgn and Performance of Adaptve Systems Based on Structured Stochastc Optmzaton Strateges, Crcuts and Systems Magazne, Vol. 5, No., 25, pp 8 2. [7] Laleh Badr Asl and Vahd Majd Nezhad, Speech enhancement usng Partcle swarm optmzaton Technques, Internatonal conference on Measurng Technology and Mechatroncs Automaton, 2, pp [8] Laleh Badr Asl and Vahd Mjd Nezhad, Improved Partcle Swarm Optmzaton for Dual- Channel Speech Enhancement, Internatonal Conference on Sgnal Acquston and Processng, 2, pp 3-7. [9] L.Badr Asl and M.Geravanchzadeh, Aseual Reproducton based adaptve quantum partcle swarm optmzaton algorthm for dual channel speech enhancement, Internatonal conference on Copyrght 24 MECS I.J. Intellgent Systems and Applcatons, 24, 4, -

10 A New Dual Channel Speech Enhancement Approach Based on Accelerated Partcle Swarm Optmzaton (APSO) Informaton scence, Sgnal Processng and ther Applcatons, ISSPA, 2, pp [] Sna Ghalam Osgoue, and Masoud Geravan Chzadeh, Dual Channel Speech Enhancement based on a Hybrd Partcle Swarm Optmzaton Algorthm, 5th Internatonal Symposum on Telecommuncatons (IST 2), pp [] Whte, M.S. and Flockton,S.J., Chapter n Evolutonary Algorthms n Engneerng applcatons, Sprnger Verlog, 997, pp [2] Kumon, T.Iwasak, M.,Suzuk, T.Hashyama, T.;Matsu,N., Okuma,S., Nonlnear system dentfcaton usng Genetc Algorthm Industral Electroncs Socety, IECON th Annual conference of the IEEE- volume 4, Oct. 2, pp Vol.4. [3] X.S.Yang, Nature-Inspred Metaheurstc Algorthms, Lunver Press, 2nd edton, 2. [4] Yang, X. S., Deb, S., and Fong, S., Accelerated Partcle Swarm Optmzaton and Support Vector Machne for Busness Optmzaton and Applcatons, n: NDT2, CCIS 36, Sprnger,2, pp [5] A.R, J. Beerens,M.Holler and A. Hekstra, Perceptual evaluaton of speech qualty (PESQ) - A new method for speech qualty assessment of telephone networks and codecs, n Proc. IEEE, Int. Conf. Acoustcs Speech, Sgnal processng, 2;vol.2, pp [6] Phlpos C.Lozou, Speech Enhancement Theory and Practce, CRC press.27 [7] Y.Hu and P.Lozou, Subjectve comparson of speech enhancement algorthms, ICASSP Proceedngs. Toulouse, Franc., 26, pp [8] Phlpos C. Lozou and Janfen Ma, Etendng the artculaton nde to account for non-lnear dstortons ntroduced by nose- suppresson algorthms, J. Acoustc Socety of Amerca, 2, pp [9] [2] n/data/nose.html Authors Profles K.Prajna receved the B.Tech degree n ECE from Pragat Engneerng College and M.Tech degree n Radar and Mcrowave Engneerng from Andhra Unversty n 29. She s currently pursng Ph.D n the Department of ECE Andhra Unversty, Vsakhapatnam, Inda. Her area of research nterests ncludes sgnal processng, speech processng and ntellgent systems. Dr. G. Sas Bhushana Rao s presently workng as a Professor and Head of the Department n Dept. of Electroncs and Communcaton Engneerng, Andhra Unversty College of Engneerng. He has 28 years of eperence n Industry, Research and Teachng. He has 29 research publcatons n Natonal and Internatonal Conferences and Journals. He s a recpent of Dr. Survepall Radhakrshnan Best Academcan of the year 28, and best researcher award n the year of 27 from Andhra Unversty, Vsakhapatnam. Dr. K.V.V.S. Reddy s a former Professor of Electroncs and Communcaton Engneerng n Andhra Unversty College of Engneerng. He has publshed more than 7 journal and conference papers. He has 33 years of eperence n teachng and research besdes possessng 3 years of ndustral eperence. He s a Fellow of Insttute of Electroncs and Telecommuncaton Engneers (FIETE) and Lfe Member Socety for EMI/EMC Engneers, Inda. He has produced Ph.Ds and few more research scholars are actvely workng under hs gudance towards ther Ph.D. He has guded more than 7 M.E & M.Tech projects. Hs areas of research nterest are Communcaton systems, Sgnal Processng and Satellte Communcatons. R.Uma Maheswar receved the B.Tech degree n ECE from MVGR college of Engneerng and the M.Tech degree n VLSI System Desgn from Avanth Insttute of Engneerng & Technology, n 22. Presently Workng as Assstant Professor n Vgnan s Insttute of Informaton Technology, Vsakhapatnam. Prevously, she worked as Assstant Professor n Avanth College of Engneerng, whch s afflated to JNTU. She worked as IT- Assocate n IEG (Insttute for electronc Governance), Hyderabad for about one year. Her research nterest ncludes Sgnal Processng, Communcaton systems & Image processng. Copyrght 24 MECS I.J. Intellgent Systems and Applcatons, 24, 4, -