TABLE OF CONTENTS. Volume 3 Number 5 May G. Tatsis, C. Votis, V. Raptis, V. Christofilakis, P. Kostarakis 425

Transcription

1

2

3 Int. J. Communcatons, Network and System Scences, 010, 3, Publshed Onlne May 010 n ScRes ( TABLE OF CONTENTS Volume 3 Number 5 May 010 A/D Restrctons (Errors) n Ultra-Wdeband Impulse Rados G. Tatss, C. Vots, V. Rapts, V. Chrstoflaks, P. Kostaraks 45 Outage Performance of Opportunstc Amplfy-and-Forward Relayng over Asymmetrc Fadng Envronments S. Majh, Y. Nasser, J. F. Hélard 430 Measurements of Balun and Gap Effects n a Dpole Antenna C. Vots, V. Chrstoflaks, P. Kostaraks 434 The Performance Improvement of BASK System for Gga-Bt MODEM Usng the Fuzzy System K. H. Eom, K. H. Hyun, K. K. Jung 441 Analyss and Comparson of Tme Replca and Tme Lnear Interpolaton for Plot Aded Channel Estmaton n OFDM Systems D. L. Wang 446 ASIP Soluton for Implementaton of H.64 Mult Resoluton Moton Estmaton F. Tll, A. Ghorbel 453 Mcrostrp Low-Pass Ellptc Flter Desgn Based on Implct Space Mappng Optmzaton S. Tavakol, M. Zenadn, S. Mohanna 46 Partcle Swarm Optmzaton Based Approach for Resource Allocaton and Schedulng n OFDMA Systems C. K. Chakravarthy, P. Reddy 466 Interoperablty of Wreless Networks wth 4G Based on Layer Modfcaton D. Mahjabeen, A. H. M. Sayem, A. Ahmed, S. Rafque 47 Research on Access Network Intruson Detecton System Based on DMT Technology L. X. Wu, J. Zhan, Q. G. He, S. Y. He 477 Synchronzaton n Wreless Networks for Practcal MIMO-OFDM Systems M. K. Kyan, M. U. Ahmed, A. Loan 483 A Cross Layer Optmzaton Based on Varable-Power AMC and ARQ for MIMO Systems S. M. Tseng, W. S. Le 488

4 Internatonal Journal of Communcatons, Network and System Scences () Journal Informaton SUBSCRIPTIONS The Internatonal Journal of Communcatons, Network and System Scences (Onlne at Scentfc Research Publshng, s publshed monthly by Scentfc Research Publshng, Inc., USA. Subscrpton rates: Prnt: $50 per ssue. To subscrbe, please contact Journals Subscrptons Department, E-mal: sub@scrp.org SERVICES Advertsements Advertsement Sales Department, E-mal: servce@scrp.org Reprnts (mnmum quantty 100 copes) Reprnts Co-ordnator, Scentfc Research Publshng, Inc., USA. E-mal: sub@scrp.org COPYRIGHT Copyrght 010 Scentfc Research Publshng, Inc. All Rghts Reserved. No part of ths publcaton may be reproduced, stored n a retreval system, or transmtted, n any form or by any means, electronc, mechancal, photocopyng, recordng, scannng or otherwse, except as descrbed below, wthout the permsson n wrtng of the Publsher. Copyng of artcles s not permtted except for personal and nternal use, to the extent permtted by natonal copyrght law, or under the terms of a lcense ssued by the natonal Reproducton Rghts Organzaton. Requests for permsson for other knds of copyng, such as copyng for general dstrbuton, for advertsng or promotonal purposes, for creatng new collectve works or for resale, and other enqures should be addressed to the Publsher. Statements and opnons expressed n the artcles and communcatons are those of the ndvdual contrbutors and not the statements and opnon of Scentfc Research Publshng, Inc. We assumes no responsblty or lablty for any damage or njury to persons or property arsng out of the use of any materals, nstructons, methods or deas contaned heren. We expressly dsclam any mpled warrantes of merchantablty or ftness for a partcular purpose. If expert assstance s requred, the servces of a competent professonal person should be sought. PRODUCTION INFORMATION For manuscrpts that have been accepted for publcaton, please contact: E-mal: jcns@scrp.org

5 Int. J. Communcatons, Network and System Scences, 010, 3, do:10.436/jcns Publshed Onlne May 010 ( A/D Restrctons (Errors) n Ultra-Wdeband Impulse Rados Gorgos Tatss 1, Constantnos Vots 1, Vasls Rapts 1, Vasls Chrstoflaks 1,, Panos Kostaraks 1 1 Physcs Department, Unversty of Ioannna, Ioannna, Greece Semens Enterprse Communcatons, Enterprse Products Development, Athens, Greece E-mal: {gtatss, kvots, vrapts}@grads.uo.gr baslos.chrstoflaks@semens-enterprse.com, kostaraks@uo.gr Receved February 4, 010; revsed March 6, 010; accepted Aprl 0, 010 Abstract Ultra-Wdeband Impulse Rado (UWB-IR) technologes, although are relatvely easy n transmsson but they present dffcultes n recepton, n fact the recepton of such waveform s a qute complcated matter. The man reason s that n fully dgtal recever the receved waveform must be sampled at a rate of several GHz. Ths paper focuses on the mpact of the Analog to Dgtal (A/D) converson stage that s used to sample the receved waveform. More specfcally we focus on the mpact of the two man parameters that affect the performance of the Software Defned Rado (SDR) system. These parameters are the bt resoluton and the tme jtterng. The nfluence of these parameters s deeply examned. Keywords: UWB, Impulse Rado, ADC, Jtter Error, Quantzaton Error 1. Introducton UWB transmsson has recently receved great attenton n academa and ndustry for applcatons n wreless communcatons. A UWB system s defned as any rado system that has a 10 db fractonal bandwdth larger than 0% of ts center frequency, or has a 10 db bandwdth larger than 500 MHz. It s expected that many approaches used for short-range wreless communcatons wll be revaluated and a new ndustral sector wth hgh data rate wll be formed. Fully dgtal recever for UWB-IR requres the use of A/D converson and SDR technques as descrbed below. The RF waveform receved from the antenna s drectly dgtzed from the antenna va an A/D converson stage. Then the dgtal nformaton derved from the UWB waveform s handled and processed by a DSP. However, ths process, ntroduce new sgnal dstortons, due to the new uncertantes ntroduced, that are the jtter error and the quantzaton error. The latter comes exclusvely from the bt resoluton of the A/D converter, whle tme jtterng comes merely from the aperture jtter of the ADC, and from clock jtter of the samplng crcutry [1,]. In ths paper we examne the mpact of those two parameters on the bt error rate performance of an UWB-IR fully dgtal recever. In UWB-IR systems a pulse tran, consstng of very short pulses and occupyng very large spectrum [3], s transmtted. Several modulaton schemes are used such as B-phase, Pulse Poston, On-Off keyng etc. [4]. In ths paper we choose Bnary Pulse Poston Modulaton (BPPM). We consder transmsson through ndoor multpath envronment [5], n the presence of whte Gaussan nose. The performance of the system s evaluated by the bt error probablty (BEP) n terms of jtter and quantzaton nose. An expresson of BEP s derved and numercally results are presented.. Analog to Dgtal Converson Durng the A/D converson addtonal nose s produced at the output of the A/D converter due to two man reasons: Quantzaton and Jtter error. The frst s llustrated n Fgure 1 (a) and t s a result of the dfference between the analog, contnuous nput sgnal and the dgtzed output of the ADC. The fnte ADC resoluton gves the form of the stars-lke sgnal. If an ADC has a bt resoluton of N bts, t means that the output sgnal s coded at N dfferent bnary numbers, from 0 to N 1. Let assume that the nput sgnals peak to peak ampltude ( V pp ) s the same wth the ADC full-scale voltage range. Then the correspondng quantzaton step s Q V / N pp. An ampltude value at the nput s mapped to the nearest N bt bnary number and the

6 46 G. TATSIS ET AL. absolute dfference between nput-output can be from zero to Q /, thus the quantzaton error s from Q / to Q /. We assume that the nput sgnal can take any random value wthn a quantzaton step, wth equal probablty. Therefore the dstrbuton of quantzaton error s unform and ts probablty densty functon f ( x ), s shown n Fgure 1(b). Obvously, t has a mean of zero and t s easy to prove that the standard devaton of quantzaton error s Q / 1, as follows, Q Q q () Q 0 q 3 1 Q Q x f x dx x dx x dx Q Q Q 4 1 The second error that concerns our study s called jtter error and t s a result of the non nfnte tmng precson of the samplng procedure and the ADC mperfectons. The fact s that there s an uncertanty at the samplng tme whch causes an uncertanty at the nput voltage of the sgnal. Ths effect s shown n Fgure. Let the nput sgnal at an A/D converter be Vt (). We focus at the Ampltude tme t 1, correspondng at a multplcate of samplng perod. Due to jtter effect the sample taken by the ADC s the one at the tme t 1 t j, where t j s a random varable, assumng normally dstrbuted wth zero mean and standard devaton j. The correspondng voltage error s then, Vj V( t1tj) V( t1). By rewrtng ths expresson we have, Vt ( t) Vt ( ) V V t t V t t 1 j 1 j ( 1 j) ( 1) j t j For small t j we can approxmate the expresson n the brackets wth the frst dervatve of Vt () [6], and obtanng, dv () t ' Vj tj tjv t1 dt tt 1 3. Sgnal Model Descrpton ( ) The transmtted pulses have the form of a Gauss monocycle,.e. the frst dervatve of a standard Gauss pulse. Fgure 3 shows a schematc representaton of the BPPM modulated transmtted sgnal. The bt perod s T f (frame perod) and the tme offset represents the modulaton ndex. Tme s dvded nto frames, the perod of V(t) t j (1) () V j Tme (a) f (x) t 1 t 1 + t j t 1/Q Fgure. Jtter error effect. T f T f -Q/ Q/ x (b) Fgure 1. (a) Dgtzaton of an analog contnuous sgnal (dotted lne) to dscrete and quantzed sgnal (normal lne); (b) Unform dstrbuton f ( x ) of quantzaton error x Fgure 3. BPPM sgnalng.

7 G. TATSIS ET AL. 47 the frames s T f. We determne the symbol by ts poston wthn each frame. A logc 0 s a pulse at the begnnng of the frame, whle a logc 1 s delayed by a small amount of tme. The modulated pulses waveform s() t s expressed by Equaton 3 below, N 1 s() t Ebw( t jtf bj) (3) j0 where, wt () s the pulse shape (frst dervatve of a Gaussan pulse) normalzed to have total energy wt () dt 1, E b s the energy per bt, T f s the frame perod, b j s the j-th bt, s the BPPM modulaton ndex, N s the total number of transmtted pulses. w s related to the pulse wdth wth the relatonshp Tp w, where T p represents the wdth of the pulse. The modulaton ndex s chosen to satsfy the orthogonalty of the transmtted symbols,.e., wtwt () ( ) dt0. We choose greater than the pulse duraton,.e., Tp. 4. Theoretcal Analyss of Error Probablty In order to derve an expresson for the probablty of error, we consder the transmt and receve system model shown n Fgure 4. The transmtted sgnal, s() t, descrbed above, propagates through a multpath channel wth mpulse response ht (). Then t s converted from analog to dgtal usng an A/D converter. As mentoned above the nput sgnal s sampled at the ADC frequency and quantzed wth correspondng ADC resoluton. For the detecton of the symbols, a matched flterng technque s used. The matched flter s constructed by two correlators. The receved sgnal s correlated wth the expected symbols and the output s the dfference of those. The output s sampled every frame perod. We assume perfect channel estmaton and synchronzaton. From ths pont, the analyss contnuous for the frst frame perod,.e., 0 t Tf. We use vector notaton, whch represents the sampled versons of the sgnals. Fgure 4. System transmsson-recepton model. All the vectors has length, N f Tf fs, where T f s the frame perod and f s s the samplng frequency. The templates for the two symbols ( x0, x 1) are the transmtted symbols for 0 and 1 respectvely after the channel, n s Gaussan process, representng total addtve nose, wth a mean value of zero and a double sde power spectral densty, N /. 0 The channel mpulse response, correspondng to the IEEE a model [5] for ndoor multpath envronments, s gven by ht () ( ) L1 l 0 l t l, where L s the total resolvable channel paths, l, l are the gan coeffcent and tme delay, respectvely, for the correspondng l path. Thus the transmtter sgnal after the channel s expressed as follows, L 1 x() t ( s h)() t als( t l) (4) l 0 where (*) denotes convoluton. The receved dscrete sgnal r s gven n Equaton 5 below, r=xn n n (5) where, n N(0, n), n N0 / s the total addtve nose at the recever, n j s the nose vector due to jtter error and by usng Equaton. we have: n N(0, ), ' T ' jt j x x and n q s the nose term due to Q Q Q quantzaton nose.e. n q U (, ), q. 1 ' The dervatve x s calculated from Equaton 3 and Equaton 4, as follows, L 1 d d xt () ast l ( l) dt dt l 0 (6) N 1 L1 d Eb w( t jtf bj l) dt j0 l0 and by takng the dscrete (sampled) vector. The waveform wt () as mentoned before s the frst dervatve of a gauss monopulse, thus d wt () s the second derva dt tve of the pulse. To obtan an expresson for the probablty of error on symbol detecton, we must frst defne the decson metrc at the output of the correlators n Fgure 4 at tme t Tf. For smplcty we consder the transmsson of a 0 and n the same way we can derve an expresson for the 1. The decson metrc s then, D T r x -x x nn n x -x T j q 0 1 j T j q T T j q T T (0) R ( ) n n x0 -x1 n x0 -x1 x x -x nn n x -x T xx xx nn n x -x T R xx xx j q q j jt

8 48 G. TATSIS ET AL. and we obtan, D0 Rxx(0) Rxx( ) Nn N (7) q where, R ( ) s the autocorrelaton functon of vector xx T x at tme, N n n n x -x s a gaussan j 0 1 random process ncludng thermal nose and jtter nose T N N(0, ), ( ) x -x x -x,.e., n g1 g1 n jt T and N n x -x s the nose term due to quantze- q q 0 1 ton whch s a summaton of N f terms of unformly dstrbuted random varables. Because of the fact that N f s usually a suffcently large number we may use the central lmt theorem [7], and approxmate ths term wth a Gaussan process wth varance.e., g T N N(0, ), x -x x -x q g g q Therefore the decson metrc s a Gaussan r.v. wth mean R (0) R ( ) and standard devaton xx xx thus the probablty of error s expressed as follows: Rxx (0) Rxx ( ) Pe Q g1 g 5. Numercal Results, g1 g After the above analyss we calculate the error probablty numercally usng smulaton program to evaluate Equaton 8 and by averagng over 1000 channel realzatons correspondng to IEEE a model CM1. The parameters that used are: wdth of the pulses Tp 00 psec, modulaton ndex 1nsec, frame perod Tf 100 nsec, samplng frequency fs 0 GHz, yeldng a channel tme resoluton of 50 psec. Fgure 5 shows the (8) bt error probablty (BEP) as a functon of sgnal to nose rato wth 6 bt ADC resoluton and wth dfferent number of jtter standard devaton. We can see that jtter s a sgnfcant factor to the performance especally when nose has lower power. Beyond 10 db of sgnal to nose rato, jtter s the man cause of performance degradaton. Fgure 6 shows the error probablty as a functon of jtter standard devaton. On top of the graph we set the bt resoluton of A/D at 4 bts and the curves correspond to several sgnal to nose ratos. Agan we can see that n cases of hgher SNR, the error probablty has strong dependence on jtter. In the graph at the bottom, we set SNR to 10 db and we change the bt resoluton of ADC. It s nterestng to notce that an ncrease of bt resoluton more than 4 bts doesn t mprove performance. The dependence of error probablty of bt resoluton s shown BEP Resoluton - 4 bts Eb/No = db 10-5 Eb/No = 4 db Eb/No = 6 db Eb/No = 8 db Eb/No = 10 db 10-6 Eb/No = 1 db Eb/No = 14 db Jtter devaton (psec) E b/n 0 = 10 db 10 0 Resoluton - 6 bts BEP BEP σj = 0 psec σj = 1 psec σj = psec σj = 3 psec E b/n 0 (db) Fgure 5. Bt error probablty as a functon of sgnal to nose rato (E b /N 0 ) for dfferent values of jtter standard devaton (psec), wth 6 bt ADC resoluton bt bt 4 bt 8 bt Jtter devaton (psec) Fgure 6. Bt error probablty as a functon of jtter standard devaton (psec), varyng sgnal to nose rato, wth 4 bt ADC resoluton (top graph) and varyng ADC resoluton wth E b /N 0 =10dB (bottom graph).

9 G. TATSIS ET AL. 49 BEP E b/n 0 = db E b/n 0 = 4 db ADC bt resoluton E b/n 0 = 8 db x 10-4 E b/n 0 = 1 db ADC bt resoluton 7 8 Fgure 7. Bt error probablty as a functon of ADC bt resoluton wth 1psec jtter, and wth dfferent values of sgnal to nose rato (E b /N 0 ). BEP x n Fgure 7, wth jtter standard devaton at 1 psec and varyng SNR. In all cases there s a lmt at bt resoluton and t s obvous that a use of 4 bts s adequate to lead to a suffcent performance. 6. Conclusons In the present paper we have studed the mpact of the two parameters that affect the performance of the dgtzng stage. These parameters are the jtter error and the quantzaton error. The error probablty dependence fr- om both parameters was nvestgated and presented. Both of them are crtcal to error performance of Ultra- Wdeband Impulse Rado systems. Jtter error plays an mportant role especally when addtve nose s not very strong. Quantzaton error s also a sgnfcant factor for the BEP mprovement for bt resoluton below 4 bts. For more than 4 bts of ADC resoluton the mprovement s neglgble. From the above study n order to assure low BEP, the jtter must be kept as low as possble (-3 psec) and the ADC resoluton above 4 bts. 7. Acknowledgements Ths research project (PENED) s co-fnanced by E. U. - European Socal Fund (80%) and the Greek Mnstry of Development-GSRT (0%). 8. References [1] V. N. Chrstoflaks, A. A. Alexandrds, P. Kostaraks and K. P. Dangaks, Software Defned Rado Implementaton Aspects Related to the ADC Performance, Proceedngs of the 6th Internatonal Multconference on Crcuts, Systems, Communcatons and Computers, Crete, 7-11 July 00, pp [] R. H. Walden, Analog-to-Dgtal Converter Survey and Analyss, IEEE Journal on Selected Areas n Communcatons, Vol. 17, No. 4, Aprl 1999, pp [3] X. Shen, M. Guzan, R. C. Qu and T. Le-Ngoc, Ultra- Wdeband Wreless Communcaton and Networks, John Wley & Sons, Malden, 006. [4] I. Güvenc and H. Arslan, On the Modulaton Optons for UWB Systems, Proceedngs of IEEE Conference on Mltary Communcaton, Boston, Vol., 003, pp [5] J. R. Foerster, M. Pendergrass and A. F. Molsch, A Channel Model for Ultrawdeband Indoor Communcaton, Proceedngs of Internatonal Symposum on Wreless Personal Multmeda Communcaton, Yokosuka, October 003. [6] A. Fort, M. Chen, R. W. Brodersen, C. Desset, P. Wambacq and L. V. Besen, Impact of Samplng Jtter on Mostly-Dgtal Archtectures for UWB Bo-Medcal Applcatons, Proceedngs of IEEE Internatonal Conference on Communcatons, Glasgow, 4-8 June 007, pp [7] A. Papouls and S. U. Plla, Probablty, Random Varables and Stochastc Process, McGraw-Hll, New York, 00.

10 Int. J. Communcatons, Network and System Scences, 010, 3, do:10.436/jcns Publshed Onlne May 010 ( Outage Performance of Opportunstc Amplfy-and-Forward Relayng over Asymmetrc Fadng Envronments Abstract Sudhan Majh 1,,3, Youssef Nasser 1,,3, Jean Franços Hélard 1,,3 1 European Unversty of Brettany, Rennes, France Insttut Natonal des Scences Applquees, Rennes, France 3 Insttute of Electroncs and Telecommuncatons of Rennes, Rennes, France E-mal: {sudhan.majh, youssef.nasser, jean-francos.helard}@nsa-rennes.fr Receved February 16, 010; revsed March 1, 010; accepted Aprl, 010 Ths letter analyzes the outage probablty of opportunstc amplfy-and-forward relayng over asymmetrc and ndependent but non-dentcally dstrbuted (.n.d) fadng envronments. The work nvestgates the scenaros where cooperatve nodes are located at dfferent geographcal locatons. As a result, the dfferent sgnals are affected by dfferent.n.d fadng channels, one may undergo Rcan fadng dstrbuton and others may undergo Raylegh fadng dstrbuton. In ths letter, a lower bound of the outage probablty for varous asymmetrc fadng envronments s derved at hgh SNR by applyng the ntal value theorem. The analytcal model s valdated through Monte-Carlo smulaton results. Keywords: Outage Probablty, Opportunstc Relayng, Amplfy-and-Forward Relayng, Raylegh and Rcan Fadng Channels, Asymmetrc Fadng Channels, Independent and Non-Identcally Dstrbuted 1. Introducton Cooperatve relayng s a promsng technology for future wreless communcatons. It can beneft most of the leverages of multple nput multple output (MIMO) wthout usng the conventonal MIMO schemes [1]. Among the cooperatve technques, the opportunstc relayng, n whch only one relay (R) node forwards the source s (S) data to the destnaton (D) has shown ts effcency compared to other technques []. The outage performance of opportunstc amplfy-andforward (AF) relayng over a symmetrc fadng envronment s wdely nvestgated n [1,3,4]. However, n practce, cooperatve nodes are usually located n dfferent geographcal locaton envronments and at dfferent dstances wth respect to S and D. Therefore, one lnk could be ether n lne-of-sght (LOS) stuaton or n non-los (NLOS) stuaton. For example, the fxed relay nodes used for forwardng source s data to a specfc regon (e.g. tunnel, behnd the buldng) often use drectonal antenna, so the R-D lnk s usually n a LOS stuaton. However, we cannot assume such a stuaton n all transmsson envronments especally when D s n a deep shadowng regon wth respect to S. The outage performance analyss of opportunstc relayng for mxed and.n.d fadng envronments s, therefore, of practcal mportance. The asymmetrc fadng channel s ntroduced n [5]. However, the authors of ths work assume addtve whte Gaussan nose (AWGN) channel of the R-D lnk. In [6], an approxmaton of the outage performance over asymmetrc fadng channel,.e., Raylegh and Rcan, s gven. However, to the best knowledge of the authors, no closed-form expresson s provded. In ths letter, the analytcal model of the outage probablty of opportunstc AF relayng over asymmetrc and.n.d fadng envronments s gven. Then, the lower bound of the outage probablty for hgh SNR values s deduced and verfed through Monte-Carlo smulatons.. System Model and SNR Evaluaton In ths framework, we consder a general -hop AF relayng network consstng of S, m relays, R, 1,,...,m and D. We assume that D performs maxmum rato combnng at the recevng sde. The equvalent nstantaneous end-to-end sgnal-to-nose rato (SNR) for opportunstc

11 S. MAJHI ET AL. 431 AF relayng s gven as [3]: SNR d P h sr P h s sr s rd Nsr N s sd rd max N sd Ps hsr P s hrd P h N N rd 1 where hab represents the channel gan of the a-b lnk, P a s the power transmtted by the node a. As mentoned n [3], we assume that AWGN varance s N ab 1 / 0, a,b where 0 s proportonal to the system SNR. For smplcty reasons, we use dfferent notatons of the random varables of the dfferent fadng dstrbutons. For the Raylegh fadng, let ab Pa hab be the nstanttaneous sgnal power and for the Rcan fadng, the nstantaneous sgnal power s denoted as ab. The probablty densty functon (PDF) of ab and ab are expressed respectvely as: 1 x/ ab f x e ab Kab 1 ab ab ab f e I ab 0 ab ab K 1/ K 4Kab Kab 1 ab (1) () where, ab E{ ab} ab E{ ab} and K ab s the Rcan factor. E {.} holds for expectaton value. The upper bound of the nstantaneous SNR of (1) for the opportunstc AF relayng s defned as: (3) SNRub Ps h sd 0 max mn Ps h sr,p s h rd (4) Ths nstantaneous SNR value wll be used n the followng secton to evaluate the outage probablty. 3. Analyss of Outage Probablty In ths secton, we provde the lower bound of the outage probablty of opportunstc AF relayng for dfferent channels gven n Fgure Asymmetrc Channel I Theorem 1: If S-D lnk s Raylegh fadng channel and S-R and R-D lnks are Rcan fadng channels, then the lower bound of the outage probablty over asymmetrc channel I s: m 1 Ksr 1 K 1 rd I m1 pout Ksr Kr d 1 (5) m sd 1 sr e rd e Fgure 1. Dfferent asymmetrc fadng channels of a cooperatve network. sd s sd Proof: By usng P h, rd P s hrd sr P s hsr and n [4], the outage probablty over the asymmetrc channel I can be wrtten as: I I p Pr[ ] (6) out I where SNRub s derved to ub sd max, R 1 / 0, max max 1,,..., m and mn ξ sr,ξ rd. The cumulatve dstrbuton functon (CDF) of the random varable over.n.d s gven as: 11 [ sr ] 1 [ r d ] Ksr 1 F Pr Pr 1Q 1 K sr, sr Krd 1 Q 1 K rd, rd where Q 1, s the 1 st order Marcum Q-functon and the PDF of s obtaned by dfferentatng above as: Ksr 1 f K sr, f rd sr Krd 1 Q 1 K rd, f rd max over.n.d fad- The CDF of the random varable ng channel can be expressed as: F max 1 max s obtaned by df- and the correspondng PDF of ferentatng the above as: m ub sr (7) (8) F (9)

12 43 S. MAJHI ET AL. m max j 1 j 1 j m f f F (10), the Snce F 0 0 m 1th order dervatve of (10) at hgh SNR.e., at 0 as 0, can be wrtten as: m 1 m 1 max 0 1 m f m! f 0 (11) I The outage probablty gven n (6) s a CDF of ub, whch can be evaluated by usng the ntal value theorem (IVT) of the Laplace Transformaton (LT). The LT of the PDF of the random varable I ub can be expressed by usng Equaton 15 n [3] and, then (11), as: m 1 I 1 Lf ( ) f 0 f m m s m m! m f sd f s ub 1 sd 1 max 0 f and 1 (1) Snce 0 f 0 sd are constant wth respect I to the varable s, the PDF of ub s obtaned by applyng the nverse LT (ILT) on (1) as: m I m ub sd 1 f ( ) f 0 f 0 (13) We complete the proof by ntegratng (13) and substtutng the vale of 0 f 0. f and 3.. Asymmetrc Channel II sd Theorem : If S-D and S-R lnks are Rcan fadng channels and R-D lnk s Raylegh fadng channel, then the lower bound of the outage probablty over asymmetrc channel II s: K sr 1 1 K 1 m sd II 1 m pout K (14) sd Ksr m 1 1 sd e sr e rd Proof: For the asymmetrc channel II, we use sd Ps hsd, sr P s hsr and outage probablty can be expressed as: rd Ps hrd. The II II pout Pr[ ub ] (15) II where ub sd gmax gmax max g 1,g,...,gm and g mn sr, rd. The CDF of the random varable g over.n.d fadng channel can be wrtten as:, K 1 F Q K, F sr g sr rd sr (16) where F s the CDF of the random varable rd The correspondng PDF of g s expressed as: K 1 f Q K, f sr g 1 sr rd sr 1 F f rd sr rd. (17) Smlarly, by usng the IVT and the ILT, the PDF of can be derved as: II ub m II ub sd 1 m 0 g 0 f f f (18) By ntegratng (18), we complete the proof Asymmetrc Channel III Theorem 3: If S-D lnk s Rcan fadng channel and S-R and R-D lnk are Raylegh fadng channels, the correspondng lower bound of outage probablty s: p III out K m m sd K sd 1 sd e 1 sr rd m1 Proof: For the asymmetrc channel III, we use P h, sd s sd P h and sr s sr The outage probablty can be wrtten as: rd s rd (19) P h. III III p Pr[ ] (0) out III where ub sd max max max 1,,..., m and mn sr, rd. The correspondng PDF of the random varable s gven by: ub, 1 1 f F f F f (1) sr rd rd sr Agan by usng the IVT and ILT, the PDF of obtaned as: m III ub sd 1 m 0 0 III ub s f f f () By ntegratng (), we complete the proof. Smlarly, the outage probablty of other possble asymmetrc channels can be derved by usng the above procedure. The upper bound of the outage probablty of the opportunstc AF relayng can be derved smply by usng the above method and Equaton 8 n [7]. 4. Numercal Examples In ths secton, analytcal and Monte-Carlo smulaton results are presented. Snce the channels are.n.d, we set dfferent means for dfferent S-R/R-D lnks. In the

13 S. MAJHI ET AL. 433 Rcan fadng channel, the Rcan factor K ab s unformly dstrbuted n [,3] and the mean ab of the NLOS components are unformly dstrbuted n [0,1]. The LOS components are derved for a gven K ab and ab. It s clear from (5), (14) and (19) that the outage probablty over Rcan fadng channel s obtaned by Ksd substtutng 1/ sd Ksd 1e / sd n (5) and the outage probablty over Raylegh fadng channel s obtaned by substtutng Ksd 0 n (19). Fgure shows the lower bound of the outage probablty over the symmetrc and asymmetrc fadng envronments. Due to the presence of LOS sgnal, the outage performance over Rcan fadng channel outperforms all other scenaros. Inversely, due to the absence of drect sgnal, the Raylegh fadng channel has poorer outage performance than the other scenaros. The opportunstc relayng provdes better outage performance than wthout cooperaton. It mples that the outage performance of opportunstc relayng depends manly on cooperatve lnks (S-R and R-D lnks). For ths reason, asymmetrc channel I provdes better outage performance than the asymmetrc channel II and asymmetrc channel III due to the presence of LOS sgnal nboth S-R and R-D lnks. We also note that asymmetrc channel II provdes better outage performance than asymmetrc channel III. Snce S-D and R-D lnks undergo the same fadng n both scenaros, the LOS component exstng n S-R lnk of Outage probablty Rcan, Analytcal Raylegh, Analytcal Asymmetrc I, Analytcal Equaton (5) Asymmetrc II, Analytcal Equaton (14) Asymmetrc III, Analytcal Equaton (19) Smulaton SNR [db] Fgure. The outage probablty over asymmetrc channel I, asymmetrc channel II and asymmetrc channel III. Due to the hgh SNR approxmaton for the analyss, analytcal results converge wth Monte-Carlo smulaton results at medum and hgh SNR regme. scenaro II hghly mproves the outage performance. It s clear from the above dscusson that S-R s a domnatng lnk, therefore, t s better to localze the opportunstc relay node n LOS envronment wth respect to S n order to mprove the overall outage performance. Fnally, the Monte-Carlo smulaton results provded n Fgure shows that the analytcal outage probabltes are a tght bound at medum and hgh SNR regme. 5. Conclusons In ths letter, the outage performance of opportunstc AF relayng over asymmetrc and.n.d fadng envronments has been nvestgated. A lower bound of the outage probablty has been derved and valdated through Monte-Carlo smulaton results. We show that the outage performance s better when the relay s n LOS stuaton wth respect to the source rather than to the destnaton. 6. Acknowledgements The authors would lke to thank the European IST-FP7 WHERE project and the European Network of Excellence NEWCOM++ for support of ths work. 7. References [1] J. N. Laneman, D. N. C. Tse and G. W. Wornell, Cooperatve Dversty n Wreless Networks: Effcent Protocols and Outage Behavor, IEEE Transactons of Informaton Theory, Vol. 50, No. 1, 004, pp [] Y. Zhao, R. Adve and T. J. Lm, Outage Probablty at Arbtrary SNR wth Cooperatve Dversty, IEEE Communcatons Letters, Vol. 9, No. 8, 005, pp [3] Y. Zhao, R. Adve and T. J. Lm, Symbol Error Rate of Selecton Amplfy-and-For- ward Relay Systems, IEEE Communcatons Letters, Vol. 10, No. 11, 006, pp [4] A. Bletsas, H. Shn and M. Z. Wn, Cooperatve Communcaton wth Outage Optmal Opportunstc Relayng, IEEE Transactons on Wreless Communcatons, Vol. 6, No. 9, 007, pp [5] M. Katz and S. Shama, Relayng Protocols for two Colocated Users, IEEE Transactons on Informaton Theory, Vol. 5, No. 6, 009, pp [6] H. Suraweera, G. Karagannds and P. Smth, Performance Analyss of the Dual-Hop Asymmetrc Fadng Channel, IEEE Transactons on Wreless Communcatons Letters, Vol. 8, No. 6, 009, pp [7] K.-S. Hwang, Y.-C. Ko and M.-S. Aloun, Performance Analyss of Incremental Opportunstc Relayng over Identcally and Non-Identcally Dstrbuted Cooperatve Paths, IEEE Transactons on Wreless Communcatons, Vol. 8, No. 4, Aprl 009, pp

14 Int. J. Communcatons, Network and System Scences, 010, 3, do:10.436/jcns Publshed Onlne May 010 ( Measurements of Balun and Gap Effects n a Dpole Antenna Abstract Constantnos Vots 1, Vasls Chrstoflaks 1,, Panos Kostaraks 1 1 Physcs Department, Unversty of Ioannna, Panepstmoupols, Ioannna, Greece Semens Enterprse Communcatons, Enterprse Products Development, Athens, Greece E-mal: kvots@grads.uo.gr, baslos.chrstoflaks@semens-enterprse.com, kostaraks@uo.gr Receved February 1, 010; revsed March 15, 010; accepted Aprl 18, 010 In the present paper, desgn and analyss of a.4 GHz prnted dpole antenna for wreless communcaton applcatons are presented. Measurements on return loss and radaton pattern of ths antenna confguraton are ncluded n ths nvestgaton. The prnted dpole s combned wth the feedng structure of a mcrostrp va-hole balun and s fabrcated on an FR-4 prnted-crcut-board substrate. Two nevtable dscontnutes are ntroduced by ths antenna archtecture n the form of rght-angle bends n the mcrostrp feed lne and n the dpole s gap, respectvely. The mpact of mterng these bends n the reflecton coeffcent, resonance bandwdth and radaton pattern of antenna has been nvestgated by means of smulaton and experment. Keywords: Prnted Dpole, Integrated Balun, S-Parameters, Radaton Pattern 1. Introducton The mcrostrp antenna archtecture, n general, offers nherent narrow bandwdth and qute low gan. These lmtatons do not provde wde usage of these antennas n wreless applcatons systems. Besdes, the evoluton of wreless communcatons leads to more compact and small equpment that demandng antennas wth smaller sze and profle. Hence, the scentfc communty has started to nvestgate methods to mprove ths antenna archtecture and provde better qualty of servces on wreless communcaton systems. In ths way, many research actvtes were based on the prnted dpole antenna because t has low profle, smple structure and omndrectonal radaton pattern. In order to develop ths antenna confguraton a prnted dpole antenna wth ntegrated balun and mcrostrp lne as feedng structure was proposed [1-3]. Based on these consderatons, we desgn and fabrcate a.4-ghz prnted dpole antenna wth ntegrated mcrostrp balun. Ths antenna desgn offers all the advantages of prnted crcuts and the correspondng geometry characterstcs have been n detal studed and nvestgated [4-5]. In order to mprove the bandwdth and the gan of ths archtecture we study and nvestgate the mpact on varatons of the l and w geometrcal parameters on antenna performance. The frst corresponds to the rght angle bend n the mcrostrp balun and the second affects the dpole s arms. Detals of the structure and desgn process are presented n the next secton (Secton ). The correspondng smulated and measured results are presented and dscussed n Secton 3. The paper concludes n Secton 4.. Desgn Process and Structure The geometry and desgn parameters of the.4 GHz prnted dpole antenna are drawn n Fgure 1. It s a modfed antenna desgn that was ntroduced by the correspondng lterature [1-4]. Ths prnted dpole antenna was etched on Fr4 substrate wth thckness h = 1.5 mm and permttvty ε r = 4.4. The ground plane of the mcrostrp lne and the dpole strps were prnted at the bottom layer. A mcrostrp va-hole balun acts as an unbalance-to-balance transformer from the coaxal lne to the prnted dpole strp. The lengths of the dpole arm strps and the mcrostrp balun are approxmately a quarter wavelength [4-6]. Based on matchng technques theory, the ntegrated balun affects the current flow at each dpole arm, because of cancellaton of the current flow to ground on the outsde part of the outer conductor n correspondng coaxal lne. In fact, the balun confguraton ensures that the currents whch flow the dpole arms become qute dentcal wthout elmnatng the radaton effcency [6]. The structure parameters of the proposed dpole for.4 GHz frequency pont are lsted as follows:

15 C. VOTIS ET AL. 435 Dpole strps: length L 1 = 0.8 mm, wdth W 1 = 6 mm, dpole gap: g 1 = 3 mm; Mcrostrp balun: length L = 3 mm, L 3 = 16 mm, L 4 = 3 mm, L 5 = 3 mm, wdth W = 3 mm, W 3 = 5 mm, W 4 = 3 mm, gap g = 1mm; Va radus: r = mm; Ground plane: length L 6 = 1 mm, wdth W 5 = 17 mm; Sde of mcrostrp bend: l varable; Sde of dpole strp bend: w varable. Accurate dmensons of each part of dpole and ntegrated balun have numercally been computed and nvestgated [4,5,7]. These parameters were specfed to acheve desred performance of the prnted dpole antenna n the frequency of.4 GHz. Both smulated and expermental results ntroduce ths performance. The prototype prnted dpole antenna s shown n Fgure. Ths prototype dpole on whch the dscontnutes are not nvestgated (l and w unchanged) has frequency resonance at about.45 GHz and an effcent bandwdth for wreless applcatons at.4 GHz ISM band. An nterestng approxmaton for these dscontnutes Bottom Layer s proposed. As already mentoned, the rght angle bends at the mcrostrp feed lne and at the strp arms of the prnted dpole are nevtable dscontnutes and can cause degradaton n crcut performance. Ths s due to the fact that such dscontnutes ntroduce parastc reactances whch can lead to phase and ampltude errors, nput and output msmatch and possbly spurous couplng [8]. Based on dscontnutes forms, mcrowave engneerng theory has proposed thoughts to face the dffcultes. One approach n order to elmnate ths effect s to compensate the dscontnuty drectly, by chamferng or mterng the conductor. That way, the excess capactance at the bend s reduced [8]. It s generally known that the optmum value of the mter length depends on the characterstc mpedance and the bend angle. Ths s the purpose of the proposed procedure. The varaton s mpact of the correspondng parameters l and w on return loss, resonance bandwdth and radaton pattern have been nvestgated. About resonance bandwdth defnton s specfed as the frequency range n whch return loss s less than 10 db. These varatons on the values of the l and w parameters ntroduce geometrcal modfcatons n the prototype dpole archtecture so as to obtan the correspondng dfferent prnted dpole antennas. Therefore, based on the prototype antenna desgn (Fgure ) sxteen dfferent prnted dpole antennas have been desgned and mplemented. Each of them has dfferent value of l and w parameter. Both these values are ranged from 0 mm to 3 mm, respectvely. The return loss and radaton pattern of the correspondng prnted dpole antenna n each case were nvestgated. 3. Results and Dscusson Top Layer Fgure 1. Geometry of prnted dpole antenna wth ntegrated mcrostrp balun. Fgure. Prototype prnted dpole. The defned varatons of l and w parameters n the prototype prnted dpole cause an nterestng amount of smulated and measured results. More precsely, the return loss measurements are presented n four groups of prnted dpole antennas. Each of them corresponds to four dpoles that have the same value of w parameter, but also have dfferent value of l parameter, rangng from 0 mm to 3 mm. The correspondng smulated results are presented n Fgures 3, 4, 5 and 6 respectvely. These curves provde that the return loss characterstcs of the prnted dpole are affected only by the value of l parameter. In fact, the resonance of them s ndependent of the value of l and w parameters. Moreover, n the frequency range of the resonance bandwdth the form of the curve becomes qute more flat as the value of l parameter ncreases. On the other hand, the length of w geometrcal parameter does not affect the form of these curves. These observatons are also ndcated by the correspondng expermental results.

16 436 C. VOTIS ET AL. Fgure 3. Smulated return loss of the prnted dpole antenna for 4 l parameter values and for w parameter equals to 0 mm. These are provded by a Network Analyzer and are also presented n Fgures 7, 8, 9 and 10, respectvely. From these curves a relatvely good agreement between the smulated and measured return loss was observed. For each value of l and w parameters the measured frequency pont of mnmum return loss s approxmately.45ghz and the correspondng resonance bandwdth ranges from.0 GHz to.75 GHz. Another ssue s that the expermental results of return loss presents a resonance pont on frequency range of 4 GHz. Ths result s not n good agreement wth the smulaton results. Ths dfference may be due to effects of coaxal to mcrostrp transton, ncluded n the measurements but not taken nto account n the smulated results. In addton, the mpact of l and w parameters seems not to have mportant nterest n the frequency range of 4 GHz. Fgure 4. Smulated return loss of the prnted dpole antenna for 4 l parameter values and for w parameter equals to 1 mm. Fgure 6. Smulated return loss of the prnted dpole antenna for 4 l parameter values and for w parameter equals to 3 mm. Fgure 5. Smulated return loss of the prnted dpole antenna for 4 l parameter values and for w parameter equals to mm. Fgure 7. Measured return loss of the prnted dpole antenna for 4 l parameter values and for w parameter equals to 0 mm.

17 C. VOTIS ET AL. 437 Fgure 8. Measured return loss of thfde prnted dpole antenna for 4 l parameter values and for w parameter equals to 1 mm. On the other hand, both smulated and expermental return loss results ndcate that as the value of l parameter ncreases, the shape of the correspondng curve becomes flat for wder frequency range but the frequency pont of mnmum remans qute stable, smultaneously. Ths observatons shows that return loss may be demonstrated for a specfc frequency range as parameter l s adjusted. Therefore, for better operaton of prnted dpole antenna at frequency pont of nterest, some of ts geometrc characterstcs have to be modfed. Both smulated and measured results ndcate that a value of l parameter may be consdered to be mportant to acheve effcent return loss measurements for the prnted dpole antenna desgn. Instead, ths observaton for the w parameter does not exst. Return loss and resonance bandwdth seem to be ndependent of w parameter varaton. Fgures 11 and 1 present the smulated and expermental results of return loss for l = 0 mm and w varaton from 0 mm to 3 mm wth step 1 mm. These curves confrm the correspondng ndependence. Fgure 9. Measured return loss of the prnted dpole antenna for 4 l parameter values and for w parameter equals to mm. Fgure 11. Smulated return loss of the prnted dpole antenna for 4 w parameter values and for l parameter equals to 0 mm. Fgure 10. Measured return loss of the prnted dpole antenna for 4 l parameter values and for w parameter equals to 3 mm. Fgure 1. Measured return loss of the prnted dpole antenna for 4 w parameter values and for l parameter equals to 0 mm.

18 438 C. VOTIS ET AL. Based on these crtera, the proposed nvestgaton of l and w parameter offers a way to mplement prnted dpole antenna elements wth the same geometrc characterstcs and wthout declnatons between them on the return loss characterstcs. Ths observaton provdes an attractve soluton on antenna array mplementatons at modern wreless applcatons whch are consttuted by a specfc number of usually same antenna elements. In fact, t s observed that as the resonance bandwdth becomes more flat and unform, the agreement between the correspondng return loss fgures of geometrcally dentcal dpoles become more practcable and realzed. From ths, t s also provded that the l parameter s proposed to approxmate the value of mm, as an optmum value for better performance. These observatons are also renforced by the unchanged frequency range of the resonance bandwdth n each case. Besdes, the radaton patterns of prnted dpole antennas as l and w parameters vary are also presented. Fgures 13, 14, 15 and 16 show these radaton dagrams (a) (a) (b) Fgure 14. Smulaton: Prnted dpole antenna radaton patterns for 4 l parameter values and for w parameter equals to 1 mm, (a) E plane, (b) H - plane. (b) Fgure 13. Smulaton: Prnted dpole antenna radaton patterns for 4 l parameter values and for w parameter equals to 0 mm, (a) E plane, (b) H - plane. (a)

19 C. VOTIS ET AL. 439 at.45 GHz. Each of them corresponds to a defned value of w parameter and four dfferent values of l parameter, too. From these fgures, the correspondng results do not ndcate dfferences as the radaton element remans the same (w = constant), so the values of l and w parameters do not affect the radaton characterstcs of prnted dpole antenna. 4. Conclusons (b) Fgure 15. Smulaton: Prnted dpole antenna radaton patterns for 4 l parameter values and for w parameter equals to mm, (a) E plane, (b) H - plane. A prnted dpole antenna wth ntegrated balun s studed and nvestgated. The complete structure has mplemented, smulated and expermentally measured for several values of l and w parameters. Good agreement between smulated and measured results on return loss and resonance bandwdth has been acheved. Smulated radaton pattern has also been specfed for each value of l and w parameters. Return loss seems to be affected by the varaton of l parameter, but resonance bandwdth and radaton dagram do not depend on t. Moreover, the w parameter varaton on dpole s geometry does not provde changes on reflecton coeffcent values and polar curves of radaton pattern. A proposed value of l parameter s also specfed for dentcal return loss characterstcs among prnted dpole antennas wth smlar geometry characterstcs. In general, as the value of l ncreases, the return loss of the antenna becomes more flat for wder frequency range. Ths effect s crucal for nnovator wreless communcaton engneerng and especally antenna array desgn. 5. Acknowledgements (a) Ths research project (PENED) s co-fnanced by E. U.- European Socal Fund (80%) and the Greek Mnstry of Development-GSRT (0%). 6. References (b) Fgure 16. Smulaton: Prnted dpole antenna radaton patterns for 4 l parameter values and for w parameter equals to 3 mm, (a) E plane, (b) H - plane. [1] D. Edward and D. Rees, A Broadband Prnted Dpole wth Integrated Balun, Mcrowave Journal, Vol. 30, No. 5, 1987, pp [] N. Mchshta, H. Ara, M. Nakano, T. Satoh and T. Matsuoka, FDTD Analyss for Prnted Dpole Antenna wth Balun, Asa Pacfc Mcrowave Conference, Sydney, 000, pp [3] G. S. Hlton, C. J. Ralton, G. J. Ball, A. L. Hume and M. Dean, Fnte Dfference Tme Doman Analyss of a Prnted Dpole Antenna, 19th Internatonal IEEE Antennas and Propagaton Conference, Endhoven, 1995, pp [4] H.-R. Chuang and L.-C. Kuo, 3-D FDTD Desgn Analyss of a.4 GHz Polarzaton Dversty Prnted Dpole Antenna wth Integrated Balun and Polarzaton

20 440 C. VOTIS ET AL. Swtchng Crcut for Wlan and Wreless Communcaton Applcaton, IEEE Transactons on Mcrowave Theory and Technques,Vol. 51, No., 003, pp [5] C. Vots, V. Chrstoflaks and P. Kostaraks, Geometry Aspects and Expermental Results of a Prnted Dpole Antenna, Internatonal Journal Communcatons, Network and System Scences, Vol. 3, No., 010, pp [6] C. A. Balans, Antenna Theory Analyss and Desgn, Wley Interscence, New York, 1997 [7] R. Garg, P. Bharta, I. Bahl and A. Ittpboon, Mcrostrp Antenna Desgn Handbook, Artec House, Canton, 001. [8] D. M. Pozar, Mcrowave Engneerng, Wley, New York, 1998.

21 Int. J. Communcatons, Network and System Scences, 010, 3, do:10.436/jcns Publshed Onlne May 010 ( The Performance Improvement of BASK System for Gga-Bt MODEM Usng the Fuzzy System Abstract K-Hwan Eom 1, Kyo-Hwan Hyun 1, Kyung-Kwon Jung 1 Department of Electronc Engneerng, Dongguk Unversty, Seoul, Korea Department of Electronc Engneerng, Hanlm Unversty, Chuncheun, Korea E-mal: khwanum@dongguk.edu Receved August 1, 009; revsed December 15, 009; accepted March 17, 010 In ths paper we propose an automatc bandwdth control method for the performance mprovement of Bnary Ampltude Shft Keyng (BASK) system for Gga-bt Modem n mllmeter band. To mprove the performance of the BASK system wth a fxed bandwdth, the proposed method s to adjust a bandwdth of low pass flter n recever usng the fuzzy system. The BASK system conssts of a hgh speed shutter of the transmtter and a counter and a repeater of recever. The repeater conssts of four stage converters, and a converter s constructed wth a low pass flter and a lmter. The nputs to the fuzzy system are the remnder and ntegral of remander of counter, and output s a bandwdth. We used a Vterb algorthm to fnd the optmum detecton from output of the counter. Smulaton results show that the proposed system mproves the performance compared to the fxed bandwdth. Keywords: BASK, Gga-Bt MODEM, Bandwdth, Low Pass Flter, Fuzzy System 1. Introducton The 60 GHz band stll beng free and unlcensed, a large bandwdth, for example of the order of 1 GHz, and easly be used. In dgtal modulaton of the 60 GHz band, a problem s ISI (Inter Symbol Interference) [1]. Dgtal base band sgnals often are rectangular pulse tran. When rectangular pulses are passed through a band lmted channel, the pulses wll spread n tme, and the pulse for each symbol wll smear nto the tme ntervals of succeedng symbols. Ths causes ISI and leads to an ncreased probablty of the recever makng an error n detectng a symbol. There are many methods to mnmze ISI as lkelhood sequence estmaton, whtened matched flters and decson-feedback equalzaton []. One approach to mnmzng ISI s to use pulse shapng technques. The most popular pulse shapng flter used n moble communcatons s the rased cosne flter. However, the maxmum value of the RF waveform and rased cosne fltered pulses do not always match [-4]. Also, n Heterodyne method, IF process s gven gan of recever, but an ncrease n analog converson steps, the more the prce wll also ncrease. The BASK system conssts of a hgh speed shutter and a mxer of the transmtter, and a counter and a repeater of the recever for solves these problems. The hgh speed shutter of the transmtter s ntroduced for pulse shapng, whch can mnmze ISI. Usng repeater for mprove SNR and make rectangular pulse tran. The repeater conssts of few stage converters. A converter s constructed wth a low pass flter and a lmter. In ths paper propose an automatc bandwdth control method for performance mprovement of BASK system. Propose method s that adjust a bandwdth of low pass flter n recever usng the fuzzy logc system. The fuzzy logc system s normally used to formulate human knowledge, but here we create the membershp functons and the fuzzy rule base by means of the smulaton results. The nputs of the fuzzy logc system are the remnder and ntegral of remander of counter, and output s bandwdth. We use 8 bt counter and Vterb algorthm wth soft decson. Rule base nference was accomplshed usng the max-mn nference procedure. Defuzzfcaton of the bandwdth output was acheved the center of gravty computaton. In order to verfy the effectveness of the proposed method, smulatons were performed by fxed bandwdth and BER.. BASK System wth a Fxed Bandwdth In mllmeter wave band, BASK system wth a fxed

22 44 K.-H. EOM ET AL. bandwdth of Gga-bt MODEM wthout IF process usng hgh speed shutter for pulse shapng of nput sgnal and mnmze ISI n the transmtter, and usng repeater for mprove SNR and make rectangular pulse tran n recever. Fgure 1 shows the block dagram of BASK system wth a fxed bandwdth. In the transmtter, RCS s rased cosne sgnal generator. The transmtter uses a hgh speed shutter that can truncate the sde lobe of the rased cosne flter. A shutter performs swtchng wndow. The output of a shutter s gven by sn( t/ Ts ) () t, n 1 hs () t t (1) 0, n 0 Where () t s a gan for the symbol perod, s the roll off factor, t s the tme, T S s the symbol perod, and n s the state of the symbol. A shutter functon s to make a constant envelope. The recever uses a repeater wthout IF (Intermedate Frequency) that conssts of two stage converters. A converter s constructed wth the LPF and the lmter. Desgn parameters of converters are bandwdth of the LPF (BLPF) and stffness of the lmter (SL: Stffest Lmter). The theoretcal soluton s gven by y () t SL( G x ()) t () Where G x () t s the nput of the lmter, y () t s the output of the converter, SL s a transfer functon of the lmter. The block dagram of a converter s shown n Fgure. Pulse tran Vterb Encoder Shutter Mxer LPF Antenna Amp The repeater can mprove sgnal-to-nose rato (SNR), and make rectangular pulse tran. 3. Proposed Method The block dagram of proposed automatc bandwdth control s shown n Fgure 3. The proposed method s that adjust the bandwdth of low pass flter n recever usng a fuzzy logc system. The output of counter n recever depends on the pattern sequence deeply, so we need the controls for the ranges of bandwdth to mprove the performance of the system. The nputs to the fuzzy logc system are the remander and ntegral of remander of counter, and output s a bandwdth. In order to create the membershp functons and fuzzy rule base, we smulated on remnder and ntegral of remnder of counter. The smulaton results of the remnder and ntegral remnder of 8 bt counter s shown n Fgure 4. In Fgure 4, we can study that the sum of remnder jumps f a bg remnder happens n negatve or postve. Therefore we apply the bandwdth control usng the fuzzy logc system due to such stuatons. The nputs are fuzzfed accordng to the nput membershp functons and output membershp functons n Fgures 5 and 6. The fuzzy rule-base conssts of a total of 15 rules. The LPF x (t) Gan Gx (t) output σ 1 nput y (t) RCS A c cos( f t) Fgure. The block dagram of a converter. (a) The transmtter Antenna Antenna BDF Amp Rectfler LPF Amp DC cut flter BDF Amp Rectfler LPF Amp DC cut flter Pulse tran Vterb Decoder Counter nth Converter 1th Converter Pulse tran Vterb Decoder F y look Counter nth Converter 1th Converter (b) The recever U n U 1 Repeater U n U 1 Repeater Fgure 1. The block dagram of BASK system wth a fxed bandwdth. Fgure 3. The block dagram of proposed bandwdth control system.

23 K.-H. EOM ET AL Remander (x-8 * Round[x/8]) 40 0 Integral of remander Integral of remander (b) Broad bandwdth Fgure 4. Smulaton of counter for bandwdth control (a) Narrow bandwdth Remander (x-8 * Round[x/8]) (a) Remander Integral of remander (b) Integral of remander Fgure 5. The membershp functon of fuzzy nput (b) Optmal bandwdth Remander (x-8 * Round[x/8]) Fgure 6. The membershp functon of fuzzy output nput/output fuzzy relaton s chosen on the bass of the smulaton results as shown n Table 1. In Table 1, R and IR are remander and ntegral of remander. Lngustc Varables are NB (Negatve Bg), NM (Negatve Medum), NS (Negatve Small), N (Negatve), Z (Zero), P (Postve), PS (Postve Small), PM (Postve Medum) and PB (Postve Bg).

24 444 K.-H. EOM ET AL. Table 1. Fuzzy rules. IR R NB NS Z PS PB N PB PM PS Z NS Z PM PS Z NS NM P PS Z NS NM NB Average bt error rate BW control BW fx Rule base nference was accomplshed usng the maxmn nference procedure. Defuzzfcaton of the bandwdth output was acheved the center of gravty computaton [5]. 4. Smulaton In order to verfy the effectveness of the proposed method, Smulatons were performed usng MATLAB. The carrer frequency was 60 GHz and message data rate was 1 Gbps. In order to mprove SNR, t s better to change angle of lmter as θ 1 < θ < θ 3 < θ 4, and these parameters are not requred exact value. Vterb algorthm parameters are constran length k = 7, codng rate = 1/, and generator polynomnals for octal codes are 171, 133 [6]. Fgure 7 shows the average BER for the sgnal pror to repeater and the sgnal posteror to repeater usng Vterb algorthm. In Fgure 7, SNR of non-shutter, the non-repeater, and the repeater s 31 db, 30 db, and db respectvely when the BER s Fgure 8 shows the average BER for the fxed bandwdth and automatcally controlled bandwdth by fuzzy logc system. In Fgure 8, the proposed automatc bandwdth control method by fuzzy logc system s mproved the SNR Average bt error rate BER performance repeater non-repeater non-shutter Carrer to nose rato E s /N 0 /db Fgure 7. BER performance for the repeater Carrer to nose rato E s /N 0 /db Fgure 8. Smulaton of BER performance. about 8 db at BER of 10-3 aganst the case of fxed bandwdth. 5. Conclusons In ths paper proposed a method for mprovng the performance of the BASK system for automatcally tunng the bandwdth of LPF. The BASK system was constructed a hgh speed shutter of transmtter and a repeater of recever. The shutter was ntroduced for pulse shapng to mprove the ntersymbol nterference and the repeater conssts of few stage converters, and a converter was constructed wth a low pass flter and a lmter. Proposed method was usng fuzzy logc system. Fuzzy nputs were remander and ntegral of remander of counter. Output was bandwdth. In order to verfy the effectveness of the proposed method, smulatons were performed by fxed bandwdth and BER. The smulaton results are summarzed as follows: Fuzzy System has nputs, 1 output, 15 the number of fuzzy rules. So that can be confgured smply. SNR of non-shutter, the non-repeater, and the repeater s 31 db, 30 db, db, respectvely at BER of The proposed method s mproved the SNR about 8 db at BER of 10-3 aganst the case of fxed bandwdth. 6. References [1] V. R. M. Thyagarajan, R. H. M. Hafez and D. D. Falconer, Broadband Indoor Wreless Communcaton n (0 ~ 60) GHz Band: Sgnal Strength Consderatons, Unversal Personal Communcaton, Vol., October 1993, pp [] T. S. Rappaport, Wreless Communcatons, nd Ed-

25 K.-H. EOM ET AL. 445 ton, Prentce Hall, New Jersey, 00. [3] E. Lndskog and A. Paulraj, A Transmt Dversty Scheme for Channels wth Intersymbol Inference, Proceedngs of IEEE Internatonal Conference on Communcatons, New Orleans, Vol. 1, June 000, pp [4] S. Haykn, Communcaton Systems, 4th Edton, John Wley Inc., Canada, 000. [5] R. Johnston, Fuzzy Logc Control, GEC Journal of Research, Vol. 11, No., 1994, pp [6] M. Hosemann, R. Habendorf and G. P. Fettwes, Hardware-Software Codesgn of A 14.4 Mbt - 64 State - Vterb Decoder for An Applcaton-Specfc Dgtal Sgnal Processor, Proceedngs of IEEE Workshop on Sgnal Processng Systems 003, Seoul, 7-9 August 003, pp

26 Int. J. Communcatons, Network and System Scences, 010, 3, do:10.436/jcns Publshed Onlne May 010 ( Analyss and Comparson of Tme Replca and Tme Lnear Interpolaton for Plot Aded Channel Estmaton n OFDM Systems Abstract Dongln Wang Department of Electrcal and Computer Engneerng, Unversty of Calgary, Calgary, Canada Emal: dowang@ucalgary.ca Receved March 10, 010; revsed Aprl 11, 010; accepted May 1, 010 Ths paper analyzes and compares two tme nterpolators,.e., tme replca and tme lnear nterpolator, for plot aded channel estmaton n orthogonal frequency dvson multplexng (OFDM) systems. The mean square error (MSE) of two nterpolators s theoretcally derved for the general case. The equally spaced plot arrangement s proposed as a specal platform for these two tme nterpolators. Based on ths proposed platform, the MSE of two tme nterpolators at the vrtual plot tones s derved analytcally; moreover, the MSE of per channel estmator at the entre OFDM symbol based on per tme nterpolator s also derved. The effectveness of the theoretcal analyss s demonstrated by numercal smulaton n both the tme-nvarant frequency-selectve channel and the tme varyng frequency-selectve channel. Keywords: OFDM, Channel Estmaton, Tme Replca, Tme Lnear Interpolaton, Vrtual Plots 1. Introducton Orthogonal frequency dvson multplexng (OFDM) [1-3] has been wdely used n hgh-speed wreless communcaton systems, such as broadband wreless local area networks (WLANs) [4], wreless metropoltan area networks (WMANs) [5] and worldwde nteroperablty for mcrowave access (WIMAX) [6], due to ts advantages of transformng frequency-selectve fadng channels nto a set of parallel flat fadng sub-channels and elmnatng nter-symbol nterference [7]. Channel estmaton s one of the most essental tasks n compensatng dstorton from channels and performng coherent detecton n OFDM systems. Estmaton s usually performed by usng plot tones [8, 9] and s based on nsertng known plot tones n each OFDM symbol, where nterpolaton n tme-frequency grd [10] plays an mportant role n the estmaton process. The usage of vrtual plot tones [11-13] and tme nterpolaton can reduce the redundancy and guarantee a hgher transmsson bt rate. Among tme nterpolaton methods, tme replca [14, 15] s wdely used n tme-nvarant or slow tme-varyng channel, whch s smple to mplement and also effcent for subcarrer usage; tme lnear nterpolaton [16-18] s wdely used n slow or fast tme-varyng channel, because t s smple to realze and usually can gve a satsfactory performance. However, some nterestng questons are rased as follows: 1) what knd of tme-varyng channel s slow enough to utlze tme replca? ) Conversely, what knd of tme-varyng channel s so fast that we have to employ tme lnear nterpolaton nstead of tme replca? And 3) how much does tme lnear nterpolaton perform better than tme replca by for a tme-nvarant channel? To answer these questons above, ths paper analyzes and compares the performances of tme replca and tme lnear nterpolator n both the tme-nvarant frequencyselectve channel and the tme varyng frequency-selectve channel. The MSE of both tme nterpolators s theoretcally derved for the general cases. The equal spaced plot arrangement s employed as a specal platform for both tme nterpolators, where the postons of vrtual plot tones n one OFDM symbol correspond to those of plot tones of ts last and next OFDM symbols. Channel state nformaton (CSI) [19] at plot tones s estmated by least square (LS) estmator. CSI at vrtual plot tones n one OFDM symbol s obtaned by ether of tme nterpolators, where tme replca s to completely replcate the CSI at plot tones of ts last OFDM symbol whle tme lnear nterpolator s to lnearly nterpolate values by usng the estmated CSI at the correspondng plot tones of both ts last and next OFDM symbols. CSI at data

27 D. L. WANG 447 tones s fnally obtaned by frequency lnear nterpolaton [0]. Ths paper s organzed as follows. In Secton, the MSEs of two nterpolators,.e., tme replca and tme lnear nterpolaton, are theoretcally derved for the general case. In Secton 3, the equally spaced plot arrangement s proposed as a specal platform for analyzng these two tme nterpolators. In Secton 4, based on the proposed platform, the MSE of two tme nterpolators at the vrtual plot tones s derved analytcally; moreover, the MSE of channel estmators at the entre OFDM symbol based on these two tme nterpolators s also derved, respectvely. Numercal results are reports n Secton 5, followed by concluson n Secton 6. Notaton: g denotes the modulus. g s the E g s the expectaton operaton -norm operaton. k { } on k. { } Ekl, g means the expectaton on both k and l. Vark { g } means the varance on k. δ m m, + j( k) denotes the varaton of the CSI of the k th tone from the ( m ) th OFDM symbol to the ( m+ j) th OFDM symbol. δ m ( k ) denotes the varaton of the CSI of the k th tone from the m th OFDM symbol to the R L ( m + 1th ) OFDM symbol. em ( k ) and em ( k) are the channel estmaton errors of the m th OFDM symbol at the k th tone where tme replca or tme lnear nterpolaton are employed for CSI estmaton at the vrtual plot tones, respectvely.. MSE of Two Tme Interpolators Assume that each OFDM symbol has N subcarrers where plots occupy P subcarrers. Denote the set of plot tones by I P. By LS estmaton, the CSI at plot tones n th the m OFDM symbol can be obtaned as ˆ Ym ( k) Hm ( k) = X ( k) (1) m where Xm ( k ) and Ym ( k ) are the transmtted and receved plots of the m th OFDM symbol, respectvely. Assumng the plot tones Xm ( k ) = 1 for convenence of analyss, we have Hˆ ( k) = H ( k) + W ( k) () m m m where Hm ( k ) represents the true value and Wm ( k ) s a complex-valued sample of addtve whte Gaussan nose (AWGN) process at the m th OFDM symbol, Wm ( k)~ CN ( 0, σ ). Assumng that along the tme axs n Fgure 1, the data tones n the m OFDM symbol correspond to the th plot tones n both the ( m p) th and the ( m+ q) th OFDM symbol, the CSI at the data tones n the m th OFDM symbol can be obtaned by tme nterpolaton by usng the estmated CSI at the plot tones of both the ( m p) th and the ( m q) th + OFDM symbol, whch s thus called the vrtual plot tones. Denote the set of vrtual tones by I PP. In ths secton, we wll analyze and compare the MSE performance of two tme nterpolators: tme replca and tme lnear nterpolator..1. Tme Replca Tme replca at the vrtual plot tones n the m th symbol s to replcate the CSI at the plot tones n the ( m p) th symbol, Hˆ ( k) = Hˆ ( k), k I. (3) m m p PP By () and (3), the estmaton error of tme replca at the k th tone can be expressed as ξ e ( k ) = Hˆ ( k )- H ( k) R m m- p m = H ( k )- H ( k ) + W ( k). m-p m m-p The MSE usng tme replca can thus be obtaned as R { ( ) } { - ( )- ( ) } { δ, ( k) } σ. = E e k = E H k H k + σ R k m k m p m = E + k m pm.. Tme Lnear Interpolaton (4) (5) However, f usng tme lnear nterpolaton, the estmated CSI can be obtaned as follows, ˆ p ˆ q H ( ) ˆ m k = Hm- p( k) + Hm q( k) (6) + p + q p + q for k I PP. By () and (6), the estmaton error of tme th lnear nterpolaton at the k tone can be expressed as L p q em( k) = ( Hm-p( k )- Hm( k) ) + Wm-p( k) p + q p + q (7) p q = ( Hm+ q( k )- Hm( k) ) + Wm+ q( k). p + q p + q Fgure 1. The vrtual plot tones n the m th OFDM symbol are tme-nterpolated by usng the plot tones at both the ( m p ) th and the ( m+ q ) th OFDM symbol.

28 448 D. L. WANG Based on (7), the MSE of tme lnear nterpolaton can thus be obtaned as ξ L k m L { ( ) } = E e k pδ k q k p + q = Ek + p + q p + q p + q m pm, ( ) δmm, + q( ) σ ( ) (8)..3. Comparson Subtractng (8) from (6), the dfference between ξ can be expressed as L pq { ( k) } ( p + q ) ξ ξ = E δ + σ R L k m pm, E k pδm pm, ( k) qδmm, + q( k) p + q p + q. ξ R and (9) From (9), one can conclude that 1) In a tme-nvarant frequency-selectve channel, ξ L s always lower than ξ R by 10log ( p + q) db; whle pq n a tme-varant frequency-selectve channel, the performance comparson depends on the specfc channel varaton; ) Consderng a real-valued channel varaton, n low nose envronment, when δm pm, ( k) δ mm, + q( k) < 0 and δmm, + q( k) > δ m pm, ( k), ξr < ξ L ; 3) Consderng a real-valued channel varaton, n nosy envronment, when δm pm, ( k) δ mm, + q( k) 0 or δm pm, ( k) δ mm, + q( k) < 0 but δmm, + q( k) < δ m pm, ( k), ξ > ξ. R L 3. Specal Case: Plot Arrangement and Channel Estmators Assume that each OFDM symbol has N subcarrers where plots occupy P subcarrers and vrtual plots supermposed wth data samples also occupy P subcarrers. Fgure shows the proposed plot arrangement as a platform, whch s a specal case but not loss of generalty, where along frequency axs, the plot spacng s L and the spacng between plot and adjacent vrtual plot s L. From Fgure, one can see that along tme axs, the plot spacng s and the spacng between plot and adjacent vrtual plot s 1. Also, by LS estmaton, the CSI at plot tones can be obtaned by (1) Tme Interpolaton at Vrtual Plot Tones Denote the set of vrtual tones by I PP. The CSI at vr- Fgure. The proposed plot arrangement as a specal platform, where the plot tones n one OFDM symbol correspond to the vrtual plot tones n ts adjacent OFDM symbol. tual plot tones s obtaned by tme nterpolaton. In ths specal plot arrangement, snce the vrtual plot tones at the m th symbol corresponds to the plot tones at the ( m 1th ) symbol, tme replca at the vrtual plot tones n one symbol s to replcate the CSI at the plot tones of ts last symbol, Hˆ ( k) = Hˆ ( k), k I. (10) m m-1 PP On the other hand, f usng tme lnear nterpolaton, we can get ˆ ˆ ˆ Hm-1( k) + Hm+ 1( k) Hm( k) =, k IPP. (11) 3.. Frequency Interpolaton at Data Tones Denote the set of data tones as I D. Usng frequency lnear nterpolaton [0], the CSI at the whole OFDM symbol can be expressed as L l ˆ l ( ) ˆ Hm k + Hm( k+ L) L L ˆ H ( ) 1 1 m k+ l when k + L P = Hˆ m( k) when k = 1+ L( P 1. ) ( ) (1) where ( k+ l) ID, k IP IPP, 1 l L 1. Note that the CSI for data tones located on the rght sde beyond the ( 1+ PL) th plot/vrtual plot tone s decded by the edge nterpolaton. 4. Performance Analyss for the Specal Case Ths secton analyzes the performance of ths specal case n terms of the MSEs of tme nterpolators and the MSEs of the correspondng channel estmators.

29 D. L. WANG MSE of Tme Interpolators For ths specal case, the MSE of tme replca n (5) becomes { k } ξ ( ). R = Ek δm 1 + σ (13) On the other hand, the MSE of tme lnear nterpolaton n (8) becomes So, based on (13) and (14), the dfference between and ξ can be obtaned as follows, L δm 1( k) δm( k) ξl = Ek + σ. (14) { 1 } ξr ξl = Ek δm 1 δm 1( k) δm( k) ( k) + σ Ek. (15) And, from (15), one can conclude that 1) n a tmenvarant frequency-selectve channel, ξ L s always lower than ξ R by 3 db; n a tme-varant frequencyselectve channel, the performance dfference depends on the specfc channel varaton; ) n most stuatons, as the general case n (9), ξr > ξ L,.e., tme lnear nterpolaton s better than tme replca. 4.. MSE of Channel Estmaton Tme Replca By LS estmaton on plot tones, tme replca on vrtual plot tones and frequency nterpolaton on data tones, the correspondng MSE of channel estmaton can be expressed as ξ R P P N P ξlrl = ξp + ξr + ξ RF, (16) N N N where ξ P s the MSE of LS estmaton and ξ RF s the MSE of frequency nterpolaton when usng tme replca at vrtual plot tones. As an average of both odd and even OFDM symbols, except for the rght sde ( L 1) tones usng the edge nterpolaton, a half of other data tones wth the ndex ( k+ l) ID have k IP whle ( k+ L) IPP for frequency lnear nterpolaton; for the remanng data tones, k I PP whle ( k+ l) IP for frequency lnear nterpolaton. Hence, usng (1), we can get ξ RF n (17), L l l where ef( k+ l) = Hm( k) + Hm( k+ L) Hm( k + l), L L k IP IPP, 1 k 1+ L( P ), and e F (1 + L(P -1) + l) = H m(1 + L(P -1))- H m(1 + L(P -1) + l), are the nherent errors by frequency nterpolaton, ξ F s the nherent MSE of frequency nterpolaton. By substtutng (17) nto (16), as the followng (18), ξ ˆ { ( )- ( ) } = E H k+ l H k+ l = RF kl, m m ξ LRL can be expressed 1 L 1 L l l 1 Ekl, ef( k l ) Wm( k ) m 1( k) N P δ L L 1 L 1 l L l + 1 Ekl, ef( k+ l ) + Wm( k ) + δm 1( k) N P L L L 1 + El{ ef(1 + L(P 1) + l ) } = ξf N P L 1 L L N P σ L 1 L Ek{ δm 1( k) }, 6L N P P P N P ξ = ξ + ξ + ξ N N N LRL P R F ( L 1)( N P L 1) { δ m 1 k } + σ E k + + 6NL ( ). (17) (18) 4... Tme Lnear Interpolaton By LS estmaton on plot tones, tme lnear nterpolaton on vrtual plot tones and frequency nterpolaton on data tones, the MSE of channel estmaton can be expressed as P P N P ξ, LLL = ξp + ξr + ξ LF (19) N N N where ξ LF s the MSE of frequency nterpolaton when usng tme lnear nterpolaton at vrtual plot tones. Usng (1), ξ LF can be obtaned as shown n (0). Substtutng (0) nto (19), ξ LLL can be expressed as the followng (1), ξ ˆ { ( )- ( ) } = E H k+ l H k+ l = LF kl, m m 1 L 1 1 N P L l l δ Ekl, ef( k+ l ) + Wm( k ) + L L 1 L N P m 1 ( k) δm( k) l L l δm 1( k) δm( k) Ekl, ef( k+ l ) + Wm( k ) + L L L 1 + El{ ef(1 + L(P 1) + l ) } = N P

30 450 D. L. WANG ξ F L 1 L σ 6L N P L L N P L δm 1( k) δm( k) 1 Ek. P P N P ξ = ξ + ξ + ξ N N N LLL P R F + σ ( L 1)( N P L+ 1) 6NL δ ( k) δm( k) + E. m 1 k Comparson Subtractng (1) from (18), the dfference between and ξ LLL can be obtaned as ( L 1)( N P L+ 1) P P ξlrl ξlll = σ + + N N 6NL (0) (1) ξ LRL δm 1( k) δm( k) Ek{ δm 1( k) } Ek. () From (), one can notce that P 1) Snce N >> P, σ s neglgble and the dfferental MSE usng () s approxmately ndependent N wth nose; ) In a tme-nvarant frequency-selectve channel, ξ LRL s approxmately equal to ξ LLL ; whle n a tme-varant frequency-selectve channel, the performance comparson depends on the specfc channel varaton; 3) Consderng a real-valued channel varaton, n low nose envronment, when δm( k) δ m 1() k < 0 and δ m() k > δ ( ) m 1 k, ξlrl < ξ LLL ; 4) ξlrl ξlll < ξr ξ L. whle n the even OFDM symbols, the plot s nserted at the ( 5+ 8j) th tone. The sx-ray multpath Raylegh fadng channel s consdered. The average power delay profle s selected as λ = exp( l) l 5 l= 0 λ l, 0 l 5. (3) Fgure 3 shows the MSE performance of tme nterpolator and channel estmaton n the tme-nvarant frequency-selectve channel, where one can see that tme lnear nterpolator generatng less nose has a 3 db lower MSE than tme replca at the vrtual plot tones. However, for the correspondng channel estmaton at the whole OFDM tones, tme lnear nterpolator performs smlarly to tme replca due to a neglgble nose. Fgure 4 shows the MSE performance n a tme varyng channel where the parameters are E { ( ) k δ m 1 k } = 0.001, { ( ) 6 1 } 10 Vark m k = Var 6 { δ () k } = 10 k m δ, { δ } E ( ) 0.00 k m k =, and, respectvely. For nterpolaton at vrtual plot tones, when SNR 5 db, tme lnear nterpolator performs better than tme replca due to better nose reducton; when SNR > 5 db, tme replca, whch guarantees a more accurate nterpolaton n a low nose envronment, performs better than lnear nterpolator. Whle for the correspondng channel estmaton, when SNR 5 db, tme lnear nterpolator performs very smlarly to tme replca due to better nose reducton; when SNR > 5 db, tme replca also performs better than tme lnear nterpolator. Fgure 5 shows the MSE performance n the tme varyng channel where the parameters are E { ( ) k δ m 1 k } MSE[dB] MSE of tme nterpolator Tme replca Tme lnear nterp MSE[dB] MSE of channel estmaton Tme replca Tme lnear nterp 5. Numercal Results The OFDM system under consderaton s wth N = 51 subcarrers, and L = 8 equspaced plot tones n each symbol. The length of cyclc prefx s 3. The nterpolaton dstances p = q = 1. The modulaton s QPSK. The plot tones are all 1. For 0 j 63, n the odd OFDM 1+ 8j th tone; symbols, the plot s nserted at the ( ) SNR[dB] SNR[dB] Fgure 3. MSE of tme nterpolator and channel estmaton n tme-nvarant frequency-selectve channel.

31 D. L. WANG 451 MSE[dB] MSE of tme nterpolator Tme replca Tme lnear nterp SNR[dB] MSE[dB] MSE of channel estmaton Tme replca Tme lnear nterp SNR[dB] Fgure 4. MSE of tme nterpolator and channel estmaton n tme-varant frequency-selectve channel, where the expectaton s equal to E { ( ) 1 } k dm - k =, the varance s equal to { ( ) 6 1 } 10 - Vark dm- k =, the expectaton Ek{ d ( ) m k } 6 =- 0.00, and varance Vark{ d ( ) m k } 10 - =, respectvely, 0 k Conclusons Tme replca and tme lnear nterpolaton were analyzed and compared, especally under our proposed plot arrangement. The MSEs of both tme nterpolators were derved analytcally for both nterpolatons at the vrtual plot tones and ther correspondng channel estmaton at the entre OFDM symbol. Numercal smulaton results were demonstrated to reach an agreement wth theoretcal analyss. From the gven results, one can see that, n a tme-nvarant frequency-selectve channel, when the nterpolaton dstances p = q =1, tme lnear nterpolator has a 3 db lower MSE than replca at the vrtual plot tones whle they provde a smlar performance at the entre OFDM symbol. Moreover, one can also see that, n a tme varyng frequency-selectve channel, tme lnear nterpolator outperforms tme replca except the case, n a low nose envronment, the CSI varaton from the last OFDM symbol to the present symbol s negatve to and has a smaller absolute value than that from the present symbol to the followng symbol. 7. Acknowledgements MSE of tme nterpolator Tme replca Tme lnear nterp MSE of channel estmaton Tme replca Tme lnear nterp The author would lke to thank all the anonymous revewers of the paper. The crtcal comments by all the revewers have helped us to mprove the qualty of our paper. MSE[dB] SNR[dB] MSE[dB] SNR[dB] Fgure 5. MSE of tme nterpolator and channel estmaton n tme-varant frequency-selectve channel, where the expectaton s equal to E { ( ) 1 } k dm - k =, the varance s equal to { ( ) 6 1 } 10 - Vark dm- k =, the expectaton Ek{ d ( ) m k } 6 = 0.00, and varance Vark{ d ( ) m k } 10 - =, respectvely, 0 k 51. = 0.001, { ( ) 6 1 } 10 Vark m k = E ( ) 0.00 k δ m k =, and 6 Vark{ ( ) m k } 10 δ =, respectvely. Tme lnear nterpolator always performs better than tme replca for both nterpolaton at the vrtual plot tones and the correspondng channel estmaton at the entre tones. δ, { } 8. References [1] M. Engels, Wreless OFDM Systems, Kluwer Academc Publshers, New York, 00. [] L. Hanzo, OFDM and MC-CDMA: a Prmer, John Wley & Sons, Inc., Hoboken, 006. [3] H. Schulze, Theory and Applcatons of OFDM and CDMA: Wdeband Wreless Communcatons, John Wley & Sons, Inc., Hoboken, 005. [4] B. Bng, Wreless Local Area Networks: The New Wre- Less Revoluton, Wley-Interscence, New York, 00. [5] S. Methley, Essentals of Wreless Mesh Networkng, Cambrdge Unversty Press, Cambrdge, 009. [6] M. Ma, Current Technology Developments of Wmax Systems, Sprnger Verlag, New York, 009. [7] C. Pandana, Y. Sun and K. J. R. Lu, Channel-Aware Prorty Transmsson Scheme Usng Jont Channel Estmaton and Data Loadng for OFDM Systems, IEEE Transactons on Sgnal Processng, Vol. 53, No. 8, August 005, pp [8] R. Neg and J. Coff, Plot Tone Selecton for Channel Estmaton n a Moble OFDM System, IEEE Transactons on Consumer Electrononcs, Vol. 44, No. 3, August 1998, pp

32 45 D. L. WANG [9] W. Zhang, X.-G. Xa and P. C. Chng, Clustered Plot Tones for Carrer Frequency Offset Estmaton n OF- DM Systems, IEEE Transactons on Wreless Communcatons, Vol. 6, No. 1, 007, pp [10] X. D. Dong, W.-S. Lu and A. C. K. Soong, Lnear Interpolaton n Plot Symbol Asssted Channel Estmaton for OFDM, IEEE Transactons on Wreless Communcatons, Vol. 6, No. 5, May 007, pp [11] I. Budarjo, I. Rashad and H. Nkookar, On the Use of Vrtual Plots wth Decson Drected Method n OFDM Based Cogntve Rado Channel Estmaton Usng x1-d Wener Flter, Proceedngs of IEEE Internatonal Conference on Communcatons, Bejng, May 008, pp [1] Q. F. Huang, M. Ghogho and S. Freear, Plot Desgn for MIMO OFDM Systems wth Vrtual Carrers, IEEE Transactons on Sgnal Processng, Vol. 57, No 5, May 009, pp [13] J. H. Zhang, W. Zhou, H. Sun and G.Y. Lu, A Novel Plot Sequences Desgn for MIMO OFDM Systems wth Vrtual Subcarrers, Proceedngs of Asa-Pacfc Conference on Communcatons, Perth, 007, pp [14] R. Prasad, OFDM for Wreless Communcatons Systems, Artech House, Boston, 004. [15] A. R. S. Baha, Mult-Carrer Dgtal Communcatons, Sprnger Verlag, New York, 004. [16] K. Jhyung, P. Jeongho and H. Daesk, Performance Analyss of Channel Estmaton n OFDM Systems, IEEE Sgnal Processng Letters, Vol. 1, No. 1, January 005, pp [17] P. Jeongho, K. Jhyung, P. Myonghee, K. Kyunbyoung, K. Changeon and H. Daesk, Performance Analyss of Channel Estmaton for OFDM Systems wth Resdual Tmng Offset, IEEE Transactons on Wreless Communcatons, Vol. 5, No. 7, July 006, pp [18] H. Myeongsu, Y. Takk, K. Jhyung and K. Kyungchul, OFDM Channel Estmaton Wth Jammed Plot Detector Under Narrow-Band Jammng, IEEE Transactons on Vehcular Technology, Vol. 57, No. 3, May 008, pp [19] A. Rosenzweg, Y. Stenberg and S. Shama, On Chann- Els wth Partal Channel State Informaton at the Transmtter, IEEE Transactons on Informaton Theory, Vol. 51, No. 5, May 005, pp [0] S. Coler, M. Ergen, A. Pur and A. Baha, Channel Est- Maton Technques Based on Plot Arrangement n OF- DM Systems, IEEE Transactons on Broadcastng, Vol. 48, No. 3, September 00, pp.3-9.

33 Int. J. Communcatons, Network and System Scences, 010, 3, do:10.436/jcns Publshed Onlne May 010 ( ASIP Soluton for Implementaton of H.64 Mult Resoluton Moton Estmaton Abstract Feth Tll, Akram Ghorbel CITRA COM Research Laboratory, Engneerng School of Communcatons(SUP COM), Tuns, Tunsa E-mal: Receved March 19, 010; revsed Aprl 0, 010; accepted May 15, 010 Moton estmaton s the most mportant module n H.64 vdeo encodng algorthm snce t offer the best compresson rato compared to ntra predcton and entropy encodng. However, usng the allowed features for nter predcton such as varable block sze matchng, mult-reference frames and fractonal pel search needs a lot of computaton cycles. For ths purpose, we propose n ths paper an Applcaton Specfc Instructon-set Processor (ASIP) soluton for mplementng nter predcton. An exhaustve full and fractonal pel combned wth varable block sze matchng search are used. The soluton, mplemented n FPGA, offers both performance and flexblty to the user to reconfgure the search algorthm. Keywords: Moton Estmaton, Half Pel, Quarter Pel, ASIP 1. Introducton The fast growth of dgtal transmsson servces has created a great nterest n dgtal transmsson of mage and vdeo sgnals. These sgnals requre very hgh bt rates n order to guarantee good vdeo qualty. Therefore, compresson s used to reduce the amount of data needed for representng such sgnals. Compresson s acheved by explotng spatal and temporal redundances n sgnals [1]. H.64 vdeo codng standard currently allows an approxmately :1 advantage n terms of bandwdth savngs over MPEG-, and t has the potental to allow further bandwdth savngs of 3:1 and beyond. In other words, an H.64 coded stream needs roughly half of bt-rates to provde the same qualty got by an MPEG- encoder. It also ncludes a vdeo codng layer, whch effcently represents the vdeo content ndependently of the targeted applcaton. A network adaptaton layer whch formats the vdeo data and provdes header nformaton n a manner approprate to a partcular transport layer s used. Fnally, n order to decrease the decoder complexty, several applcaton-targeted profles and levels are defned whch enable ts successful use n dfferent vdeo applcatons and markets []. Despte the fact that t has kept the same codng aspect as prevous standards based manly on predcton, transform and entropy encodng, H.64 has ntroduced some key feature modules that have ncreased consderably the codng effcency as well as more flexblty n most of the codng process. However, H.64 s also a substantally more complex standard than MPEG-; and both the H.64 encoders and decoders are much more demandng n terms of computatons and memory than ther MPEG- counterparts [3]. Ths, coupled wth the substantal amount of research needed to properly mplement and optmze the entre relevant H.64 features, makes the development of hgh-qualty H.64 encoders a dauntng task. In addton to the complexty added by H.64 standard, low power consumpton, hgh performance and scalablty are the major constrants mposed to desgners n the development of vdeo encoders and decoders [4]. In fact, wth the dversty of confguratons supported by ths standard n terms of resolutons and applcatons, scalable archtectures for vdeo encoders are much apprecated by servce provders. In ths context, nether hardware mplementaton solutons are effcent snce they lack flexblty, nor software solutons present good performance snce processors are no longer satsfyng the hgh computatonal processng tasks [5]. To meet all these constrants, processor characterstcs can be customzed to match the applcaton profle. Customzaton of a processor for a specfc applcaton holds the system cost down, whch s partcularly mportant for embedded consumer products manufactured n hgh volume. Applcaton Specfc Instructon set Processors (ASIPs) are n between custom hardware archtectures

34 454 F. TLILI ET AL. offerng good processng performance and commercal programmable DSP processors wth hgh programmablty possbltes. They offer good programmablty and performance level but are targeted to a certan class of applcatons as to lmt the amount of hardware area and power needed [6]. Ths paper s organzed as follows: Secton presents a complexty analyss of the dfferent encoder s modules followed by the descrpton of moton estmaton standardzed by H.64. In Secton 3, we wll present the proposed algorthm for mult resoluton moton estmaton. Secton 4 presents the proposed ASIP soluton. In Secton 5 we wll present mplementaton results. Fnally, we enclose the paper by Secton 6 n whch we wll conclude ths work.. H.64 Vdeo Encoder Study.1. Man Innovatons of H.64 To acheve the requred performance, H.64 allows some key features that ensure good codng effcency. The man nnovatons of ths standard are: - Intra predcton process. - Tree structured moton estmaton, weghted predcton, multple resoluton search. - Spatal n loop deblockng flter. - Integer DCT lke Transform. - Effcent Macro Block Feld Frame codng - CABAC whch provdes a reducton n bt-rate from 5% to 15% over CAVLC... Complexty Analyss of H.64 Vdeo Encoder In order to analyze the complexty of the H.64 encodng procedure, some proflng tasks were done on the several modules of the encoder mentoned above. For ths reason, some mplementatons were performed on sngle chp DSP usng CIF resoluton n baselne profle to get the most accurate results snce we have to avod nter-chp communcaton that can bother the proflng results. Fgure 1 presents the proflng results of UBVdeo encoder mplemented on DM64 DSP of Texas Instruments [7]. We can see that the most consumng vdeo tasks are moton search whch s usng about 30% of the processng tme whle the ntra predcton, moton compensaton and encodng (ncludng transform, quantzaton and entropy encodng) are usng only 3% of the system resources. Moton search ncludes only the best matchng search whle all load and store tasks are ncluded n data transfer task whch s usng about 3% of system resources. The remanng 15% of the resources are used by other tasks such as rate control, vdeo effect detecton and btstream formattng. Hence, we can see that moton estmaton s a bottle-neck for vdeo encodng algorthms whch s takng most of system resources. However, moton estmaton s the most mportant module n the compresson procedure due to ts effcency. In ths context, some vdeo encoders are usng FPGA solutons for mplementng moton estmators as hardware accelerators snce DSPs cannot handle the processng requred by such tasks. 3. Proposed Moton Estmaton Implementaton 3.1. H.64 Moton Estmaton Lumnance component of each macro-block (16 16 samples) may be splt up n 4 ways: 16 16, 16 8, 8 16 or 8 8 as shown n Fgure. Each of the sub-dvded regons corresponds to a macro-block partton. If the 8 8 mode s chosen, each of the four 8 8 macro-block parttons wthn the macro-block may be splt n a further 4 ways: 8 8, 8 4, 4 8 or 4 4 as presented n Fgure 3. Parttons and sub-parttons gve rse to a large number of possble combnatons wthn each macroblock. Ths method of parttonng macro-blocks nto moton compensated sub-blocks of varyng sze s known as tree structured moton compensaton. In addton to the varable block sze matchng, H.64 defnes mult resoluton search process n order to provde better qualty especally for non translatonal moton and alasng caused by camera nose. Expermental analyss shows that the half and quarter-sample-accuracy 30% 3% 3% 15% Fgure 1. UbVdeo encoder profle. Intra Predcton/Moton Compensaton/Encode Others Data Transfers Moton Search Fgure. Macro-block partton Fgure 3. Macro-block sub partton

35 F. TLILI ET AL. 455 moton search adopted by H.64/AVC provde a codng gan of db compared wth MPEG- and H.63, whch corresponds to a bt-rate savngs of up to 30% [8]. Half pel search s performed on pxels nterpolated usng a 6 tap low pass flter. Furthermore, a quarter pel resoluton search s establshed usng a b-lner flter appled on half pel nterpolated pxels. 3.. Proposed Moton Estmaton Algorthm The frst step of the proposed ME algorthm conssts n full pel resoluton search. Current MB s searched n a predefned search area n the reference frame. In order to avod unused computatons and data load, the search s performed on 4 4 parttons base of the MB. For each 4 4 block, we search for the best matchng poston n the reference area. Every 4 4 block s ndependently parsed n all reference area. After that, a mergng process s started n order to determne the best partton to be used for the current MB based on the best poston whch s stored relatve to the top left pxel of the 4 4 block. The mergng process s frst used to determne f the current MB can be coded n parttons above than 4 4. So, we compare the best postons of adjacent blocks for all 8 8 parttons: f all blocks have the same best poston, current sub partton s 8 8, otherwse, t could be 8 4, 4 8 or 4 4. If 8 8 mode s selected, a best poston of the top left pxel s stored. After that, we determne the MB predcton type that can be 16 16, 16 8, 8 16 or 8 8. A mergng process smlar to the prevous one s also used: f all 8 8 sub parttons have the same type and the same best poston, MB predcton type s 16 16; otherwse t could be 16 8, 8 16 or 8 8. After fxng the MB predcton type, a moton vector s stored for each partton. Obvously, the more we use sub parttons, more data to be transferred ncreases. We note that at least 40% of nter predcton data s used to code moton vectors. For ths reason, t s better to use bgger parttons when possble. So, a predcton cost can be added by makng condtons for the merge process based on tolerance of one or two pxels n the best postons: for example, f two 8 8 blocks have the best postons dsplaced of 1 pxel, we can decde to merge them nto one 16 8 partton. After searchng for the best matchng and the best partton, we start fractonal pel search. Accordng to the best poston, for each MB partton we nterpolate the possble 8 half pxels postons around the selected partton as shown n Fgure 4. The nterpolaton s equvalent to an up-samplng of the frame pxels usng 6 tap low pass flter. After that, a further search s performed n quarter pel accuracy usng another nterpolaton process. Based on the best poston obtaned n half pel search, we generate pxels of all the 8 possble postons around the best locaton. We note that moton vectors are multpled by 4 n order to menton to the decoder f t has to nterpolate pxels for moton compensaton or not. 4. Proposed ASIP Soluton 4.1. Analyss of the Proposed Moton Estmaton Algorthm In our work, we wll adopt nstructon selecton methodology based on hardware archtecture: frst the hardware archtecture s fxed contanng selected functonal unts (FU) and then, nstructon set archtecture s determned accordng to the FUs. For ths purpose, proposed algorthm s analyzed n order to pck up the most complex modules. These modules wll be mplemented n ndependent hardware blocks (dedcated FUs). Proposed algorthm s composed manly of 3 parts: full pel search, half pel nterpolaton and ts assocated search and fnally quarter pel search wth ts fnal search. In full pel search, the MB parses the whole reference area and 4 4 SADs are computed. In ths step, the most complex process s the SAD computaton snce t ncludes dfference computaton, absolute value determnaton and accumulaton. In [9], an analyss was performed on a moton estmaton algorthm usng SAD as a dstorton measure; we found that SAD computaton s usng more than 97% of system resources. In addton, sub pel moton estmaton s also complex. In fact, the nterpolaton process for half pel s usng 6-tap flter. Half samples are calculated through a 6-tap Wener flter n both horzontal and vertcal dmensons. The nterpolaton s processed as represented n Fgure 5: dashed pxels correspond to full pxels n an 8x8 bloc. Non dashed pxels are half pxels that are calculated. For example, to nterpolate half pxel b, we use E, F, G, H, I and J as full pxels. Calculaton process s done as follows: b = Clp1 (((E 5 F + 0 G + 0 H 5 I + J) + 16) >> 5); clp functon s used to provde result n the nterval [0, 55]: f result s less than 0 we affect 0 to b and f t s more than 55 we affect 55 to b. The same calculaton process s done for vertcal rows as h. F Full pel moton vector H H H F H H0 H1 H H3 H5 F H6 H7 Half pel search Fgure 4. Fractonal accuracy pxel search. H4 Q0 Q3 Q5 Q1 Q H4 Q4 Q6 Q7 Quarter pel search

36 456 F. TLILI ET AL Instructon Set Selecton Fgure 5. Half pel nterpolaton process. Hence, half pel nterpolaton, as any flterng process s a very tme consumng task and needs a lot of data load and store. Smlarly, quarter pel nterpolaton s usng blnear flter to generate quarter pxels. Although the smplcty of the flter, ths process also needs a lot of tmng snce t s appled to a large number of data. In concluson, the man complex modules n our proposed algorthm are the moton search, half pel nterpolaton and quarter pel nterpolaton. In our archtecture, we wll use hardware accelerators for these modules for better performance for our ASIP. 4.. Functonal Unt Selecton In our proposal, 3 hardware accelerators are used: SAD calculator, half pel nterpolator and quarter pel nterpolator. The SAD calculator wll be used to handle all SAD computaton process ncludng data load from nternal memory and SAD calculaton. The result s stored n a general purpose regster. Half pel nterpolator module s used to nterpolate half pxels accordng to the standardzed flter. Ths module loads data from nternal memory and nterpolates pxels. Due to the complexty of nterpolaton, half pxels are stored n an nternal memory to be used n further possessng tasks such as quarter pel nterpolaton or even half pxels. Fnally, quarter pel nterpolator loads data from nternal memory and apples blnear flter to generate quarter pxels. In order to avod storng quarter pxels n memory, a SAD calculator s ntegrated n ths module: reference pxels are loaded and quarter pel resoluton SAD s computed. In moton compensaton process, these pxels are re-computed snce ther computaton s not as complex as half pxels. In addton to the hardware accelerators for vdeo processng, an Arthmetc and Logc Unt s used n the soluton n order to accumulate SADs, generate pxel locatons and memory addresses Vdeo Instructons SAD4Px(DestReg,Curr_Px_Addr,Ref_Px_Add r,ptch): ths nstructon s used to compute SAD of 4 pxels based on current and reference pxel locaton and Ptch value. The choce of the 4 pxels sze s based on the fact that the smallest partton allowed s 4 4; so to avod usng SAD nstructons for all parttons, we call ths nstructon as much as the current partton contans 4 pxel lnes. Snce we adopt RISC (Reduced nstructon Set Computer) archtecture, current and reference pxel locatons as well as Ptch value are stored n Specal Purpose Regsters (SPR). These regsters are used only for vdeo nstructons snce they need more than nput operands. Output of ths nstructon s stored n a General Purpose Regster (GPR), DestReg n order to be accumulated to consttute the requred SAD. The choce of the SAD computaton sze offers the flexblty to the user to choose block lnes to be compared. In fact, we can compute only some specfc lnes n order to mnmze the processng (for example odd lnes or even lnes). Interp4HafPx(RefPxAddr,Ptch): nterpolates 4 half pxels and stores the result n nternal memory. Input operands nclude the reference pxel address whch refers to the frst full pxel from whch we start nterpolaton and a ptch value that s used for data load n case of vertcal nterpolaton. Ths value s used to gve the programmer the flexblty of modfyng the search wndow sze. These operands are loaded from SPRs whle output nterpolated pxels are stored n half pel memory snce there s no need to store them n regsters. In our moton estmaton algorthm, after callng ths nstructon to nterpolate half pxels of 1 MB, SAD4Px nstructon can be called n order to compute SAD n half pel resoluton. For ths reason, the ptch value s used n ths nstructon snce the loadng step n half pel memory s equal to. Hence, we avod the use of SAD nstructons (one for full pel SAD and the other for half pel SAD). Interp4QpxSAD(DestReg,Ref_px,Curr_px,Ptch): used to nterpolate 4 quarter pxels and compute quarter pel resoluton SAD. We have chosen to separate half pel nterpolaton from quarter pel nterpolaton n order to gve the user the flexblty to stop the search at any resoluton accordng to the complexty of the algorthm. However, quarter pels are not stored and the correspondng SAD s mmedately computed. In fact, quarter pels are no longer used by the system except the best match that s used for moton compensaton where the best matchng pxels are used. So, to avod usng huge memory sze correspondng to store all nterpolated pxels, we made the choce not to store them and to recompute the best matchng pxels when requred n moton compensaton snce ther re-computaton s easy as op-

37 F. TLILI ET AL. 457 posed to half pels. Ths nstructon returns the SAD of the current poston and the ALU decdes for the best one to be used n moton compensaton. Input operands to ths nstructon, reference and current pxels postons as well as ptch value are stored n SPRs. The output s stored n GPR, DestReg to be processed by the ALU for further decsons Memory Instructons Memory nstructons are used to transfer data between memory and regsters or nter regster transfer. Four nstructons are used for ths purpose: MOVSG(Src,Dest) s used to move data from specfc to general purpose regster. The operands of ths nstructon are formed by the addresses of regsters to be manpulated. MOVGS(Src,Dest) s used to perform the nverse operaton performed by MOVSG. LOAD(SrcAddr,DestReg) s used to load data from data memory to general purpose regster. SrcAddr s the source address of data to be loaded whle DestReg n the destnaton regster ID. STORE(SrcReg,DestAddr) s used to store the content of a general purpose regster n memory. The operands are SrcReg correspondng to the source regster ID and DestAddr s the destnaton memory address Arthmetc and Logc Instructons The man goal of these nstructons s the accumulaton of SAD values computed for each 4 pxels, computng pxel addresses, compare MB SADs and provde data for condtonal jump. ALU nstructons are processng only data from general purpose regsters. We defned 3 arthmetc nstructons: ADD, SUB and MUL are used respectvely for addton, subtracton and multplcaton operatons. These nstructons have 3 operands: the frst one s the destnaton regster ID contanng the operaton result whle the remanng operands are the IDs of regsters contanng source data to be processed. SHIFT(SrcReg1,SrcReg,SrcReg3) s used for shftng data contaned n SrcReg1 by the number of bts contaned n SrcReg. The shft drecton s ndcated by SrcReg Control Instructon The nstructon JUMP ntroduces a change n the control flow of a program by updatng the program counter wth an mmedate value that corresponds to an effectve address. The nstructon has bts condton feld (cc) that specfes the condton that must be verfed for the jump: n f case the outcome of the last executed arthmetc s negatve, postve or zero. Not only ths nstructon s mportant for algorthmc purposes, but also for mprovng code densty, snce t allows a mnmzaton of the number of nstructons requred to mplement a ME algorthm and therefore a reducton of the requred capacty of the program memory Archtecture of the Proposed ASIP Data Word Length Data word length s a tradeoff between performance and complexty. In fact, the data word length corresponds to the nstructon word length whch s stored and manpulated by the processor. Hence, n case of longer nstructon word length, we have the possblty of usng more nstructons and more regsters whch wll accelerate the processng snce memory access wll be reduced. However, the nstructon decoder wll be more complex as well as the nterconnecton between components; therefore, the processor area wll be larger. In our proposal, we have only 1 nstructons whch can be coded on 4 bts. In order to smplfy the hardware archtecture, we have chosen to use 16 bts to code all nstructons. So, 1 bts can be used to address the regster fle Regster Fle Sze Snce the nstructon length s 16 bts and 4 bts are used to code nstructons, the 1 remanng are used to code the dfferent regsters used. Snce arthmetc nstructons are usng 3 GPPs, we wll code each regster on 4 bts, so 16 GPPs can be used n our archtecture. On the other sde, vdeo nstructons are usng both GPPs and SPPs. So, 8 bts only can be used to code 3 regsters n the nstructon call: each regster s addressed on bts. So, 4 SPPs are used. At ths stage, we can see the mportance of the use of GPPs and SPPs: f we use only one regster type, when callng vdeo nstructon, 1 bts are used to code 4 regsters: 3 bts are used per regster as a consequence. Therefore, only 8 regsters are used n ths case whle n our desgn we are usng 0 regsters wth the same nstructon length. Table 1 presents the dfferent Table 1. Instructon set archtecture of the proposed ASIP. Instruton SAD4Px 0000 RestReg R1 R R3 - Interp4HafPx R1 R - Interp4QpxSAD 0010 DestReg R1 R R3 - MOVSG Src DestReg MOVGS 0011 SreReg Dest LOAD 0100 #addr DestReg STOR 0101 SreReg #addr ADD 0110 DestReg SreReg1 SreReg SUB 0111 DestReg SreReg1 SreReg MUL 1000 DestReg SreReg1 SreReg SHIFT 1001 SreReg1 SreReg SreReg3 JUMP 1010 CC #addr

38 458 F. TLILI ET AL. nstructons wth the correspondng codes, operands wth ther correspondng sze Mcro Archtecture Fgure 6 presents the mcro archtecture of the proposed ASIP. The soluton s composed of an nstructon fetch module to load nstructons from program memory, nstructon decoder to enable the several functonal unts and a regster fle to store processed data. Vdeo functonal unts are connected to the nternal data memory and the ALU. Data load from external memory to nternal memory s handled by a drect memory access controller. 5. Implementaton Soluton and Results The proposed ASIP was mplemented and syntheszed on Vrtex II Pro FPGA Memory Management In our moton estmaton algorthm, the search regon area s fxed to 31 3 pxels. We note that we need to extend ths search regon by 16 pxels n both sdes (rght and bottom) snce the last rght-bottom poston must be dsplaced of a (15, 1) vector from the centre. Furthermore, to nterpolate boundary pxels, an extenson of three pxels s needed for each sde. Fgure 7 descrbes the search area wth the several extensons. Hence, the total search area has to be 53 45; so 385 pxels have to be loaded from external to nternal memory. Internal memory s desgned to be 18 Kb block RAM ntegrated n Vrtex II FPGA. We note also that a further 1 18 Kb block RAM s also needed to store the current MB. Internal memory s 8 bts wdth for mplementaton constrants: snce we adopt exhaustve search, the whole reference area s parsed n order to search for the best matchng MB; so, f we load more than one pxel from reference area, we wll be faced to an algnment problem. To avod such problems, we have chosen to load one pxel n each cycle assumng that ths procedure s more consumng n tme. Data load to nternal memory s ensured by Drect Memory Access controllers whch handles the transfer process whle the CPU s runnng. When transfer s fnshed, an nterrupt sgnal s mentoned. Synthess results of the DMA controller shown n Table presents that ths module usng roughly 10% of the avalable FPGA resources and can be run at 05 Mhz clock frequency. 5.. SAD Engne Ths engne s used to compute the SAD of 4 pxels. Ths module loads reference and current pxels from the nternal memory and performs the SAD of 4 pxels n one call. The SAD module can be used n the SAD computaton of the full pel or half pel search. As descrbed n Fgure 8, the SAD engne s provdng the output after 9 cycles from the start sgnal. The output s fnally returned to the regster fle. We note that TMS30C64 DSP s provdng SAD of blocks Fgure 6. Archtecture of the proposed ASIP.

39 F. TLILI ET AL. 459 Table. Synthess results of DMA controller. Fgure 7. Search area organzaton. Devce utlzaton summary Number of Slces 190 out of % Number of Slces Flp Flops 178 out of 816 6% Number of 4 nput LUTs: 300 out of % Number of GCLKs 1 out of 16 6% Tmng Summary: Mnmum perod/maxmum Frequency ns/05.05 MHz Mnmum nput arrval tme before clock 5.94 ns Maxmum output requred tme after clock ns Maxmum combnatonal path delay No path found Fgure 8. Tmng dagram of SAD engne. (splt_sad8 8) n 00 cycles n the best case: when all data paths are fully used [10] whle our system can provde the same result after 144 cycles wthout usng ppelne Half Pel Interpolator In our mplementaton, the proposed algorthm s derved by mnmzng the number of memory access. The formulas to compute half-pxel nterpolatons are proposed by usng the symmetry of the 6-tap FIR flter coeffcents, resultng n sgnfcant reducton of the multplcatons [11]. Ths engne s provdng 4 nterpolated pxels n each call. Input pxels are stored n 6 regsters; the sze of each one s 3 bts as descrbed n Fgure 9: We note that pxels P3 to P6 form a lne of a selected 4 4 block to be nterpolated. The output pxels are H0 to H3. A Sngle Instructon Multple Data scheme s adopted n our mplementaton. In ths mode, adders and multplers are appled smultaneously to the pxels of regsters n order to get all nterpolated pxels at the same tme. All control sgnals are provded by an FSM. We note that the nterpolaton takes 15 cycles ncludng the load process from nternal memory. Synthess results are shown n Table 3. Fgure 9. Input regsters for halfpel nterpolaton. Table 3. Synthess results of half pel nterpolator. Devce utlzaton summary Number of Slces 354 out of % Number of Slces Flp Flops 460 out of % Number of 4 nput LUTs: 343 out of 816 1% Number of MULT18X18s 4 out of 1 33% Number of GCLKs 1 out of 16 6% Tmng Summary: Mnmum perod/maxmum Frequency ns/ MHz Mnmum nput arrval tme before clock ns Maxmum output requred tme after clock ns Maxmum combnatonal path delay 3.80 ns

40 460 F. TLILI ET AL. Fgure 10. Tmng dagram of Quarte pel nterpolator Quarter Pel Interpolator When recevng Interp4QpxSAD(Ref_px,Curr_px,Ptch) nstructon, quarter pel nterpolaton and SAD computaton are started. Frst, pxels loaded from half pel memory are fed nto the nterpolator module, then, the resultng quarter pxels are transmtted to the SAD module to be compared to the current pxels. We note that QP nterpolator nterpolates and generates the SAD of 4 pxels n each call. Quarter pel SADs are returned after 14 cycles as shown n the tmng dagram shown n Fgure Conclusons Ths paper has presented effcent nstructons for mplementng moton estmaton process usng most of the key features standardzed n H.64. Frst, we analyzed the complexty of typcal H.64 encoder. From ths step, we concluded that ME s a bottle neck for the mplementaton. Then, we presented and analyzed an algorthm for ME. Based on the analyss, we proposed effcent accelerators for some modules whch need most of the processng tme. Based on the suggested hardware archtecture, we fxed the nstructon set archtecture provdng to users large codng flexblty ensurng scalablty and mult-standard support. Proposed ASIP was mplemented on Vrtex II pro FPGA wth a total area use about 61% of the FPGA Slces and 43% of the total LUTs. The mplemented modules can be run on 17 MHz clock. 7. References [1] Q. Y. Sh and H. F. Sun, Image and Vdeo Compresson for Multmeda Engneerng: Fundamentals, Algorthms, and Standards, nd Édton, CRC Press, Boca Raton, 008. [] Draft 3rd Edton of ISO/IEC (E), Redmond, WA, USA, July 004. [3] F. Kossentn and A. Jerb, Explorng the Full Potental of H.64, NAB, 007. [4] S. D. Km, J. H. Lee, C. J. Hyun and M. H. Sunwoo, ASIP Approach for Implementaton of H.64/AVC, Journal of Sgnal Processng Systems, Vol. 50, No. 1, 008, pp [5] P. Harm, et al., Applcaton Specfc Instructon-Set Processor Template for Moton Estmaton n Vdeo Applcatons, IEEE Transactons on Crcuts and Systems for Vdeo Technology, Vol. 15, No. 4, Aprl 005, pp [6] M. Kumar, M. Balakrshnan and A. Kumar, ASIP Desgn Methodologes: Survey and Issues, 14th Internatonal Conference on VLSI Desgn, Bangalore, 001. [7] I. Werda and F. Kossentn, Analyss and Optmzaton of UB Vdeo s H.64 Baselne Encoder f Texas Instru-

41 F. TLILI ET AL. 461 ment s TMS30DM64 DSP, IEEE Internatonal Conference on Image Processng, Atlanta, October 006. [8] S. Yang, et al., A VLSI Archtecture for Moton Compensaton Interpolaton n H.64/AVC, 6th Internatonal Conference on ASIC, shangha, 005. [9] W. Geurts, et al., Desgn of Applcaton-Specfc Instructon-Set Processors for Mult-Meda, Usng a Retargetable Complaton Flow, Proceedngs of Global Sgnal Processng (GSPx) Conference, Target Compler Technologes, Santa Clara, 005. [10] M. A. Benayed, A. Samet and N. Masmoud, SAD Implementaton and Optmzaton for H.64/AVC Encoder on TMS30C64 DSP, 4th Internatonal Conference on Scences of Electronc, Technologes of Informaton and Telecommuncatons (SETIT 007), Tunsa, 5-9 March 007. [11] C.-B. Sohn and H.-J. Cho, An Effcent SIMD-based Quarter-Pxel Interpolaton Method for H.64/AVC, Internatonal Journal of Computer Scence and Securty, Vol. 6, No. 11, November 006, pp

42 Int. J. Communcatons, Network and System Scences, 010, 3, do:10.436/jcns Publshed Onlne May 010 ( Mcrostrp Low-Pass Ellptc Flter Desgn Based on Implct Space Mappng Optmzaton Abstract Saeed Tavakol, Mahdeh Zenadn, Shahram Mohanna Faculty of Electrcal and Computer Engneerng, the Unversty of Sstan and Baluchestan, Zahedan, Iran E-mal: Receved February 1, 010; revsed March 7, 010; accepted Aprl 8, 010 It s a tme-consumng and often teratve procedure to determne desgn parameters based on fne, accurate but expensve, models. To decrease the number of fne model evaluatons, space mappng technques may be employed. In ths approach, t s assumed both fne model and coarse, fast but naccurate, one are avalable. Frst, the coarse model s optmzed to obtan desgn parameters satsfyng desgn objectves. Next, auxlary parameters are calbrated to match coarse and fne models responses. Then, the mproved coarse model s re-optmzed to obtan new desgn parameters. The desgn procedure s stopped when a satsfactory soluton s reached. In ths paper, an mplct space mappng method s used to desgn a mcrostrp low-pass ellptc flter. Smulaton results show that only two fne model evaluatons are suffcent to get satsfactory results. Keywords: Implct Space Mappng Optmzaton, Mcrostrp Low-Pass Ellptc Flter, Surrogate Model 1. Introducton Consderng the development of computer-aded desgn methods, optmzaton has become a wdely used technque n desgn of mcrowave crcuts. A typcal desgn problem s to choose the desgn parameters to get the desred response. The space mappng (SM), ntroduced n [1], s a powerful technque to optmze complex models. The am of ths technque s to make a shortcut usng a cheaper but less accurate model, coarse model, to gan nformaton about the optmal parameter settng of the expensve but accurate model, fne model. To obtan the optmal desgn for the fne model, the SM establshes a mappng between the parameters of the two models teratvely [1,]. In some cases, ths mappng s not explct and t s hdden n the coarse model. The mplct space mappng (ISM) [3], descrbed below, addresses ths ssue. Frst, the coarse model s optmzed to obtan desgn parameters satsfyng the desgn objectves. Second, an auxlary set of parameters n the coarse model, whch always reman fxed n the fne model, s calbrated to match coarse and fne models responses. Ths step s known as the parameter extracton step. Examples of the auxlary parameters are physcal parameters, such as relatve delectrc constant, and geometrcal parameters, such as substrate heght. The coarse model wth updated values of auxlary parameters s known as the surrogate, calbrated coarse, model. Consderng the re-calbrated auxlary parameters fxed, then, the calbrated coarse model s re-optmzed to obtan a new set of desgn parameters. These desgn parameters are gven to the fne model to evaluate ts performance [4]. The desgn procedure s stopped when a satsfactory soluton s reached. In ths paper, an optmzaton procedure based on ISM technque s appled to a mcrostrp low-pass ellptc flter. Aglent ADS and ADS Momentum [5] are employed to smulate coarse and fne models, respectvely.. Implct Space Mappng Approach The desgn objectve s to calculate an optmal soluton for the fne model, as follows x arg mn R x (1) f f f x f x f where s a sutable objectve functon. The fne model s response, R f, s, for example, S 11 at selected frequency ponts. s the optmal fne model parameters to be determned. It can be found usng the followng teratve procedure where k 1 f s f x f K x arg mn R x, p () R s refers to the surrogate model s response. To

43 S. TAVAKOLI ET AL. 463 solve Equaton (1), a two-step procedure s employed. In the frst step, the auxlary parameters are calbrated so that the surrogate and fne models responses become smlar enough. The auxlary parameters are calculated usng the followng equaton p k arg mn R K K f xf Rs xf, p (3) P 0 where p refers to the ntal auxlary parameters. Consderng the re-calbrated auxlary parameters fxed, then, the new surrogate model s re-optmzed to obtan a new new set of desgn parameters, x f, n the second step. If the fne model s response for ths new set of desgn parameters satsfes the desgn specfcatons, the algorthm s stopped. Otherwse, t re-calculates the auxlary parameters for the current desgn parameters [4,6]. a low-pass ellptc flter wth a cut-off frequency of 7 GHz. The structure of ths flter s llustrated n Fgure 1. The coarse model s composed of emprcal models of smple mcrostrp elements, as shown n Fgure. The desgn specfcatons are as follows: L4 L1 L1 L7 L1 L1 L L1 L1 L6 L1 L1 L L1 L3 L1 L3 L5 L1 W3 3. Mcrostrp Low-Pass Ellptc Flter W1 L1 L1 W1 Low-pass flters are components, whch are used to elmnate unwanted harmoncs. Low-pass ellptc flters can provde a farly sharp cut-off frequency [7]. In ths paper, ISM technque s appled to the optmzaton problem of L8 L1 L1 L1 W W4 Fgure 1. Mcrostrp low-pass ellptc flter structure. MLOC TL8 Subst="MSub" MLIN TL11 Subst="MSub" MLIN TL16 Subst="MSub" MTEE Tee1 MLIN MLIN MLIN TL1 TL TL17 Subst="MSub3" Subst="MSub" MLIN TL7 Subst="MSub" MLIN MLIN TL MTEE TL3 MTEE Tee MLIN Subst="MSub3" TL4 Subst="MSub3" TeeMLIN Subst="MSub4" Mod=Krschnng TL4 Subst="MSub4" MLIN TL5 Subst="MSub4" MLIN MLOC TL3 TL9 Subst="MSub3" Subst="MSub3" MLIN TL18 Subst="MSub" MLIN TL5 Subst="MSub4" Term Term1 MLIN Num=1 TL1 Z=50 Ohm Subst="MSub1" MLIN TL0 Subst="MSub1" MTEE Tee4 Subst="MSub" MLIN TL19 Subst="MSub" MTEE Tee3 Subst="MSub1" MLIN TL6 Subst="MSub4" MLIN TL7 Subst="MSub1" MLIN TL4 Subst="MSub1" Term Term Num= Z=50 Ohm MLIN TL8 Subst="MSub" MLIN TL6 Subst="MSub4" MLOC TL31 Subst="MSub" MLOC TL30 Subst="MSub4" Fgure. Coarse model smulated by ADS.

44 464 S. TAVAKOLI ET AL. S , 0.001GHz 7 GHz S ,11.65GHz 11.7 GHz The flter structure s made of a perfect conductor on the top of a substrate wth a relatve delectrc constant of 10 and a heght of 635 µm, backed wth a perfect conductor ground plane. When desgnng a coarse model n ADS, ts parameters could be tunable. Ths tunng capablty allows one to graphcally see how the parameters affect the responses. As a result, desgn parameters for the desgn procedure and parameter extracton step can approprately be chosen. We set L m, L5 403 m and L 8 18 m because ADS tunng process shows that these parameters do not have sgnfcant effects on desgn specfcatons. Now, the desgn parameters and auxlary parameters are gven by x f WWWW 1,, 3, 4, L1, L, L3, L6, L7 and p [ 1 hhhh,, 3, 4, r 1, r, r3, r4], respectvely, where h and refer to the heght and relatve delectrc constant for each mcrostrp lne havng a wdth of W. In the parameter extracton step, we use ADS quas-newton optmzaton algorthm to match the fne and surrogate models magntude of scatterng parameters. The optmal coarse model s obtaned usng the ADS gradent optmzaton algorthm. The man advantage of mplct space mappng optmzaton technque s that, n ths example, the desgn algorthm requres only one teraton,.e., two fne model smulatons. The coarse and fne models responses for the ntal and fnal desgn parameters are demonstrated n Fgure 3 and Fgure 4, respectvely. Table 1 shows the ntal and fnal values of desgn parameters. The orgnal and fnal values of auxlary parameters are gven n Table. (4) S1 (magntude) Fnal response of the coarse(-) and fne(--) models Frequency (Hz) x 10 9 Fgure 4. Coarse and fne models responses for fnal solutons. Table 1. Desgn parameters. Desgn parameters Intal values Fnal values W ( m ) W ( m ) W ( m ) W ( m ) L ( m ) L ( m ) L ( m ) L ( m ) L ( m ) S1 (magntude) Intal response of the coarse(-) and fne(--) models Table. Auxlary parameters. Auxlary parameters Orgnal values Fnal values h ( m) h ( m) h ( m) h ( m) r r r r Frequency (Hz) x 10 9 Fgure 3. Coarse and fne models responses for ntal solutons. 4. Conclusons Usng mplct space mappng method, the desgn parameters for a mcrostrp low-pass ellptc flter were determned. It was shown that ths technque led to de-

45 S. TAVAKOLI ET AL. 465 creasng the number of fne model evaluatons. Frst, the coarse model was optmzed to obtan desgn parameters satsfyng the desgn objectve. Second, auxlary parameters were calbrated n the coarse model to match coarse and fne models responses. Thrd, the mproved coarse model was re-optmzed to obtan a new set of desgn parameters. Fnally, the resultng desgn parameters were gven to the fne model to evaluate ts performance. The desgn procedure was repeated by the tme a satsfactory soluton was obtaned. Smulaton results showed that only two evaluatons of the fne model were suffcent to get satsfactory results for the gven case-study applcaton. 5. References [1] J. W. Bandler, R. M. Bernack, S. H. Chen, P. A. Grobelny and R. H. Hemmers, Space Mappng Technque for Electromagnetc Optmzaton, IEEE Transactons on Mcrowave Theory and Technques, Vol. 4, No. 1, 1994, pp ,. [] J. W. Bandler, Q. S. Cheng, S. A. Dakroury, A. S. Mohamed, M. H. Bakr, K. Madsen and J. Søndergaard, Space Mappng: the State of the Art, IEEE Transac- tons on Mcrowave Theory and Technques, Vol. 5, No. 1, 004, pp [3] J. W. Bandler, Q. S. Cheng, N. K. Nkolova and M. A. Ismal, Implct Space Mappng Optmzaton Explotng Preassgned Parameters, IEEE Transactons on Mcrowave Theory and Technques, Vol. 5, No. 1, 004, pp [4] S. Kozel, Q. S. Cheng and J. W. Bandler, Space Mappng, IEEE Mcrowave Magazne, Vol. 9, No. 6, December 008, pp [5] Aglent Advanced Desgn System (ADS), Ver. 008A, Aglent Technologes, Santa Rosa, CA, 008. [6] J. W. Bandler, Q. S. Cheng, D. H. Gebre-Maram, K. Madsen, F. Pedersen and J. Søndergaard, EM-based Sur- Rogate Modelng and Desgn Explotng Implct, Frequency and Output Space Mappngs, IEEE MTT-S Internatonal Mcrowave Symposum Dgest, Phladelpha, 003, pp [7] M. C. V. Ahumada, J. Martel and F. Medna, Desgn of Compact Low-Pass Ellptc Flters Usng Double-Sded MIC Technology, IEEE Transactons on Mcrowave Theory and Technques, Vol. 55, No. 1, January 007, pp

46 Int. J. Communcatons, Network and System Scences, 010, 3, do:10.436/jcns Publshed Onlne May 010 ( Partcle Swarm Optmzaton Based Approach for Resource Allocaton and Schedulng n OFDMA Systems Chlukur Kalyana Chakravarthy 1, Prasad Reddy 1 Department of Computer Scence and Engneerng, Maharaj Vjayaram Gajapath Raj College of Engneerng, Vzanagaram, Inda Department of CS&SE, Andhra Unversty, College of Engneerng, Vsakhapatnam, Inda E-mal: kch.chlukur@gmal.com, prof.prasadreddy@gmal.com Receved March 9, 010; revsed Aprl 10, 010; accepted May 11, 010 Abstract Orthogonal Frequency-Dvson Multple Access (OFDMA) systems have attracted consderable attenton through technologes such as 3GPP Long Term Evoluton (LTE) and Worldwde Interoperablty for Mcrowave Access (WMAX). OFDMA s a flexble multple-access technque that can accommodate many users wth wdely varyng applcatons, data rates, and Qualty of Servce (QoS) requrements. OFDMA has the advantages of handlng lower data rates and bursty traffc at a reduced power compared to sngle-user OFDM or ts Tme Dvson Multple Access (TDMA) or Carrer Sense Multple Access (CSMA) counterparts. In our work, we propose a Partcle Swarm Optmzaton based resource allocaton and schedulng scheme (PSORAS) wth mproved qualty of servce for OFDMA Systems. Smulaton results ndcate a clear reducton n delay compared to the Frequency Dvson Multple Access (FDMA) scheme for resource allocaton, at almost the same throughput and farness. Ths makes our scheme absolutely sutable for handlng real tme traffc such real tme vdeo-on demand. Keywords: OFDMA, Resource Allocaton, Schedulng, Qualty of Servce, Delay 1. Introducton TDMA and FDMA used for dstrbutng subcarrers n OFDM systems form statc subcarrer management schemes. Whle n OFDM-TDMA, one of the users s assgned all the subcarrers for the entre schedulng nterval, n the OFDM-FDMA, each user s assgned predetermned number of subcarrers. However, nether of these technques s tme or frequency effcent: TDMA s a tme hog and FDMA s a bandwdth hog. OFDMA s a mult-user OFDM that allows multple access on the same channel (a channel beng a group of evenly spaced subcarrers, as dscussed above). WMAX uses OFDMA, extended OFDM, to accommodate many users n the same channel at the same tme. In OFDMA, the OFDMA subcarrers are dvded nto subsets of subcarrers, each subset representng a subchannel (see Fgure 1). Dynamc subcarrer allocaton schemes whch consder the nstantaneous channel condtons have been the man area of research nterest recently. The resource allocaton s usually formulated as a constraned optmzaton problem, to ether 1) mnmze the total transmt power wth a constrant on the user data rate [1,] or ) maxmze the total data rate wth a constrant on total transmt power [3-5]. The frst objectve s approprate for fxed-rate applcatons, such as voce, whereas the second s more approprate for bursty applcatons, such as data and other IP applcatons. In the downlnk, a subchannel may be ntended for dfferent recevers or groups of recevers; n the uplnk, a transmtter may be assgned one or more subchannels. The subcarrers formng one subchannel may be adjacent or not. The standard ndcates that the OFDM symbol s dvded nto logcal subchannels to support scalablty, Fgure 1. OFDMA frame structure.

47 C. K. CHAKRAVARTHY ET AL. 467 multple access and advanced antenna array processng capabltes. The multple access has a new dmenson wth OFDMA where n a downlnk or an uplnk user wll have a tme and a subchannel allocaton for each of ts communcatons. The man motvaton for adaptve subcarrer allocaton n OFDMA systems s to explot multuser Dversty. In a K-user system n whch the subcarrer of nterest experences..d. Raylegh fadng that s, each user s channel gan s ndependent of the others, as the number of users ncreases, the probablty of gettng a large channel gan ncreases. Further, t was observed that majorty of the gan s acheved from only the frst few users. Adaptve modulaton s the means by whch good channels can be exploted to acheve hgher data rates. WMAX systems use adaptve modulaton and codng n order to take advantage of fluctuatons n the channel. The basc dea s qute smple: Transmt as hgh a data rate as possble when the channel s good, and transmt at a lower rate when the channel s poor, n order to avod excessve dropped packets. Whle Lower data rates are acheved by usng a small constellaton, such as QPSK, and low-rate error-correctng codes, such as rate convolutonal or turbo codes, the hgher data rates are acheved wth large constellatons, such as 64 QAM, and less robust error correctng codes; for example, rate convolutonal, turbo, or LDPC codes. However, a key challenge n AMC s to effcently control three quanttes at once: transmt power, transmt rate (constellaton), and the codng rate. In theory, the best power-control polcy from a capacty standpont s the so-called waterfllng strategy, n whch more power s allocated to strong channels and less power allocated to weak channels [6]. In practce, the opposte may be true n some cases. For example, n regons of low gan, the transmtter would be well advsed to lower the transmt power, n order to save power and generate less nterference to neghborng cells [7]. As mentoned earler, OFDMA thus facltates the explotaton of frequency dversty and multuser dversty to sgnfcantly mprove the system capacty. In a multuser System, the optmal soluton s not necessarly to assgn the best subcarrers seen by a sngle chosen user snce the best subcarrer of one user s also the best subcarrer for another user who has no other good subcarrers. Hence, a dfferent approach should be consdered for schedulng the best user. We consder the problem where K users are nvolved n the OFDMA system to share N subcarrers. Each user allocates non overlappng set of subcarrers Sk where the number of subcarrers per user s J(k). The allocaton module of the transmtter assgns subcarrers to each user accordng to some QoS crtera. QoS metrcs n the system are rate and BER. Each user s bt stream s transmtted usng the assgned subcarrers and adaptvely modulated for the number of bts assgned to the subcarrer. The power level of the modulaton s adjusted to meet QoS for gven fadng of the channel (see Fgure ). User 1 User User 3 User N OFDMA TRANSCEIVER OFDMA TRANSCEIVER DATA SUB CARRIER ALLOCATION Channel State Informaton Channel Stae SUB Carrer Informaton SUBCARRIER ALLOCATION ALGORITHM CHANNEL ESTIMATOR SUBCARRIER SELECTOR Subcarrer Informaton for User N Fgure. Downlnk OFDMA System archtecture.

48 468 C. K. CHAKRAVARTHY ET AL. Pkn, fk Ckn, BERk kn, where k k,n If k,n s the ndcator of allocatng the nth subcarrer to the kth user, the transmsson power allocated to the nth subcarrer of kth user can be expressed as = (, )/ f C s the requred receved power wth unty channel gan for relable recepton of c bts per symbol. Therefore, the resource allocaton problem wth an mposed power constrant can be formulated as subject to N kn, kn, k n1 kn, kn, max C, γ R C γ for all k f ( C, BER ) Pr γ P K N k k, n k k1 n1 αk,n kn, max The lmt on the total transmsson power s expressed as P max for all n {1,..., N}, k {1,..., K} and Ckn, {1,..., M}. The proposed method uses the Partcle Swarm Optmzaton for resource allocaton and schedulng n a multuser scenaro, consderng the rate, power and the subcarrer allocaton constrants.. Partcle Swarm Optmzaton Partcle Swarm Optmzaton (PSO) s motvated from the smulaton of socal behavor of anmals. It was ntroduced by Eberhart & Kennedy n In PSO, potental solutons (partcles) move dynamcally n space. PSO s smlar to the other evolutonary algorthms n whch the system s ntalzed wth a populaton of random solutons. A lst of Genetc algorthms s gven n [8-1]. Each potental soluton, call partcles, fles n the D-dmensonal problem space wth a velocty whch s dynamcally adjusted accordng to the flyng experences of ts own and ts colleagues. The locaton of the th partcle s represented as X ( x 1,, xd,, xd ). The best prevous poston (whch gvng the best ftness value) of the th partcle s recorded and represented as P ( p 1,, pd,, pd), whch s also called pbest. The ndex of the best pbest among all the partcles s represented by the symbol g. The locaton Pg s also called gbest. The velocty for the th partcle s represented as V ( v 1,, vd,, vd ). The partcle swarm optmzaton concept conssts of, at each tme step, changng the velocty and locaton of each partcle toward ts pbest and gbest locatons. The partcle swarm optmzaton concept conssts of, at each tme step, changng the velocty and locaton of each partcle toward ts pbest and gbest locatons accordng to the equatons v = w v c1rand() d pd xd c rand() pgd xd and x d xd + vd respectvely.where w s nerta weght, c1 and c are acceleraton constants [13] whch s responsble for keepng the partcle movng n the same drecton, and rand() d s a random functon n the range [0, 1]. For the frst equaton, the frst part represents the nerta of pervous velocty; the second part s the cognton part, whch represents the prvate thnkng by tself whch causes the partcle to move to regons of hgher ftness; the thrd part s the socal part, whch represents the cooperaton among the partcles [14]. Thus the socal component causes the partcle to move to the best regon the swarm has found so far. The PSO algorthm conssts of just three steps, whch are repeated untl some stoppng condton s met [15]: 1) Evaluate the ftness of each partcle ) Update ndvdual and global best ftnesses and postons 3) Update velocty and poston of each partcle Further, velocty clampng s used to prevent the partcle to move too much away from the search space, the lmts beng confned to [ V max, V max ] f the search space spans from [ P max, P max ][16]. 3. The Proposed System In ths work, we propose a Partcle Swarm Optmzaton (PSO) Approach combned wth Credt based schedulng to guarantee QOS n WMAX. Formal defnton of our schedulng model: and Mnmze Z = k 1kK N rx 1 k x k D p 1 P N Subject to tx,1 1 k u k K (1) N p,1,1 ; 1 j xk m j j M k K () where terms 1 and refer to the tme and power constrants durng schedulng respectvely. x k are decson varables, where 1 N and 1 k K, x k s 1 f a subcarrer has been allocated, 0 otherwse. The decson of whether to grant the subchannel to the subcarrer s based on whether the subcarrer les wthn the range of exstng subcarrers for a subchannel. A crtera such as rejecton of the subcarrer f t s drectly adjacent to a prevously allocated subcarrer wthn the same subchannel and acceptance f not so s used. Ths results n an mprovement of SINR. N s the total number of subcarrers per user and K total number of users, r s the rate of each allocated subcarrer, D s the target rate for each user, t s the allocaton tme for the subcarrer, u s the allowable deadlne for a user, p s the power allocaton for each subcarrer, m s the maxmum power allocaton for user. C s the maxmum allowable credts for a user. We defne dfferent penalty factors as follows:

49 C. K. CHAKRAVARTHY ET AL. 469 Penalty factor for volatng tme constrant k max 0, N tx k k1 1 U Penalty factor for volatng power constrant m K M N p x j j k k1 j1 1 K N N max mn xk,0max 0, xk max k1 1 1 In addton, n the thrd equaton, we also defne a constrant on the user usng a large porton of a subcarrer repeatedly because ths wll deny opportuntes to other users over ths subchannel. We refer to such users as selfsh users. We ntally assgn some credts δ k to each user k, whch are ncremented when each user gans addtonal subcarrers and decremented when the user loses them. We defne credt thresholds δ mn and δ max such that δ mn δ k δ max and γ s the penalty factor for volatng the credt usage. The ftness functon of each user can be evaluated as: Mnmze H(x) = Z k + w 1 + w + w 3, where w 1, w, w 3 denote the weghts for the penalty terms. For generaton of the ntal swarm, the partcle gves more preference to tems that have a closer rate to the target rate. The mappng of the veloctes to the probabltes can -v be carred out by the sgmod functon SV ( )=1 (1+e j j ) where postve veloctes drve the bt towards 1 value whle negatve veloctes towards the 0 bt values (see Fgure 3). The partcle generaton s based on the selecton rule r - D s mnmum subject to S(v j ) = 1.Hence t gves more selecton probablty to users that have closer rate to the target rate and are represented by partcles wth postve veloctes (see Fgure 4). 1.0 START Generate ntal swarm (for each user) Evaluate the ftness of the swarm (for each user) usng ftness functon Intalze pbest of each partcle (subcarrers of each user) and gbest of the swarm (for each user) Update veloctes and partcle postons Revaluate the swarm (for each user), arrange users n the order of ther ftness Reduoe number of subcarrers, power, credts for top n users wth best ftness by δ and ncrease the same for the bottom n users wth worst ftness Probablty Has maxmum teraton reached? Velocty YES END NO Fgure 3. Sgmod functon for probablty-velocty mappng. Fgure 4. Subcarrer allocaton and schedulng n PSORAS.

50 470 C. K. CHAKRAVARTHY ET AL. 4. The Smulaton Model and Results We consder the downlnk of an OFDMA system wth N subchannels and K users. The tme axs s dvded nto frames. A frame s further dvded nto S tme slots, each of whch may contan one or several OFDM symbols. The duraton of a frame s set to be 5 ms, thus we can assume that the channel qualty remans constant wthn a frame, but may vary from frame to frame. In our smulaton, there are 104 subcarrers, 1 to 50 users n the IE- EE OFDMA system. Each user transmts 80 bts n an OFDMA symbol. The modulaton type n the OFDMA system s confned to QPSK, 16-QAM, 64-QAM. (see Fgures 5-7) Followng are the smulaton results for the varaton n average throughput, farness ndex and the average delay wth the number of users. The results clearly ndcate a reducton n the delay wth the proposed swarm based approach compared to the Naïve allocaton of subcarrers,.e. allocaton on avalablty bass n FDMA wthout consderng varaton n channel condtons. The results have been evaluated for dfferent sets of target rates and target powers for the subcarrers and the prortes of users are vared after every 5 ms based on the Fgure 7. Number of users Vs. Average Delay. calculated ftness. The throughput farness ndex has been calculated as τ n =( Thmax - Thmn )/ Th mn, where Th mn and Th max are the mnmum and maxmum values of throughput of each user over n frames measured n bts. 5. Conclusons Swarm optmzaton s ncreasngly fndng ts place n multuser downlnk MIMO schedulng, smart Antenna array systems etc. In our work, we have proposed a PSO-based far Resource allocaton and schedulng algorthm for the IEEE System. We have compared our results wth the statc FDMA algorthm and have found t offers better delay characterstcs wth ncreasng number of users whle stll mantanng the farness and throughput utlzaton. Ths makes the proposed scheme absolutely useful for real-tme applcatons. 6. References Fgure 5. Number of users Vs. Average Throughput. Fgure 6. Number of users Vs. Throughput Farness Index. [1] D. Kvanc, G. L and H. Lu, Computatonally Effcent Bandwdth Allocaton and Power Control for OFDMA, IEEE Transactons on Wreless Communcatons, Vol., No. 6, 003, pp [] C. Wong, R. Cheng, K. Letaef and R. Murch, Multuser OFDM wth Adaptve Subcarrer, Bt, and Power Allocaton, IEEE Journal on Selected Areas n Communcatons, Vol. 17, No. 10, 1999, pp [3] J. Jang and K. Lee, Transmt Power Adaptaton for Multuser OFDM Systems, IEEE Journal on Selected Areas n Communcatons, Vol. 1, No., 003, pp [4] G. L and H. Lu, On the Optmalty of the OFDMA Network, IEEE Communcatons Letters, Vol. 9, No. 5, 005, pp [5] C. Mohanram and S. Bhashyam, A Sub-optmal Jont Subcarrer and Power Aallocaton Algorthm for Multuser OFDM, IEEE Communcatons Letters, Vol. 9, No. 8, 005, pp

51 C. K. CHAKRAVARTHY ET AL. 471 [6] G. Manmaran and C. Sva Ram Murthy, A Fault-Tolerant Dynamc Schedulng Algorthm for Multprocessor Real-Tme Systems and Its Analyss, IEEE Transactons on Parallel and Dstrbuted Systems, Vol. 9, No. 11, 1998, pp [7] R. Chen, J. G. Andrews, R. W. Heath and A. Ghosh, Uplnk Power Control n Mult-Cell Spatal Multplexng Wreless Systems, IEEE Transactons on Wreless Communcatons, Vol. 6, No. 7, 007, pp [8] A. J. Page and T. J. Naughton, Framework for Task Schedulng n Heterogeneous Dstrbuted Computng Usng Genetc Algorthms, 15th Artfcal Intellgence and Cogntve Scence Conference, Ireland, 004, pp [9] A. J. Page and T. J. Naughton, Dynamc Task Schedulng Usng Genetc Algorthms for Heterogeneous Dstrbuted Computng, Proceedngs of the 19 th IEEE/ACM Internatonal Parallel and Dstrbuted Processng Symposum, Denver, 005, pp [10] A. S. Wu, H. Yu, S. Jn, K.-C. Ln and G. Schavone, An Incremental Genetc Algorthm Approach to Multprocessor Schedulng, IEEE Transactons on Parallel and Dstrbuted Systems, Vol. 15, No. 9, 004, pp [11] A. Y. Zomaya and Y.-H. Teh, Observatons on Usng Genetc Algorthms for Dynamc Load-Balancng, IEEE Transactons on Parallel and Dstrbuted Systems, Vol. 1, No. 9, 001, pp [1] E. S. H. Hou, N. Ansar and H. Ren, A Genetc Algorthm for Multprocessor Schedulng, IEEE Transactons on Parallel and Dstrbuted Systems, Vol. 5, No., 1994, pp [13] R. Eberhart and Y. Sh, Partcle Swarm Optmzaton: Developments, Applcatons and Resources, IEEE Internatonal Conference on Evolutonary Computaton, Seoul, 001, pp [14] J. Kennedy, The Partcle Swarm: Socal Adaptaton of Knowledge, IEEE Internatonal Conference on Evolutonary Computaton, Indanapols, 1997, pp [15] F. van den Bergh, An Analyss of Partcle Swarm Optmzers, PhD Thess, Unversty of Pretora, 001. [16] J. Blondn, Partcle Swarm Optmzaton: A Tutoral, September 009.

52 Int. J. Communcatons, Network and System Scences, 010, 3, do:10.436/jcns Publshed Onlne May 010 ( Interoperablty of Wreless Networks wth 4G Based on Layer Modfcaton Abstract Dlshad Mahjabeen 1, Amnul Haque Mohammed Sayem 1, Ans Ahmed, Shahda Rafque 1 Electrcal and Electroncs Engneerng, Stamford Unversty Bangladesh, Dhaka, Bangladesh Appled Physcs, Electroncs and Communcaton Engneerng, Unversty of Dhaka, Dhaka, Bangladesh E-mal: dl_shadman@yahoo.com Receved February 6, 010; revsed March 1, 010; accepted Aprl 3, 010 Fourth generaton wreless communcaton systems feel the necessty of transparent and seamless user roamng wth end-to-end connectvty. These systems also demand hgher data rate, hgher moblty support and QoS guarantees due to rapd development of wreless and moble networks. These requrements open potentals for the operators to ncrease ther servce portfolo and for the users to experence context-rch and personalzed servces. Consequently the nteroperablty between dfferent wreless network platforms emerges as a crucal necessty. Here focus s gven to the sgnfcance of the network nteroperablty aspect based on layered approach and ts role n the development towards 4G. Ths paper also gves an overvew of the major 4G features and dfferentatng characterstcs from other generatons. Keywords: 4G, Interoperablty, Reconfgurablty, Cooperatveness, Cross Layer 1. Introducton Dfferent generatons of wreless communcaton mprove the facltes for users day by day. The generatons are classfed nto three groups namely 1G, G and 3G. 1G was completely analog and used for only voce transmsson [1]. G networks were bult manly for voce servces and slow data transmsson. The cellular servces combned wth GPRS became.5g. Ths generaton provdes servces such as Wreless Applcaton Protocol (WAP) access, Multmeda Messagng Servce (MMS) and for Internet communcaton servces such as emal and World Wde Web access. Although G s very popular and successful but t lacks sngle worldwde rado band technology standard as well as neffcent use of spectrum resources for bursty data. 3G networks represent the natural evoluton from prevous standards. In ths case the networks enable network operators to offer users a wder range of more advanced servces whle achevng greater network capacty through mproved spectral effcency. 3G networks offer a greater degree of securty than G predecessors. Ths generaton (3G) allows smultaneous use of speech and data servces and hgher data rates. The servces provded by 3G are wde-area wreless voce telephone, vdeo calls, and wreless data, all n a moble envronment []. However 3G performances may not be suffcent to meet needs of future hgh qualty applcatons. 3G does not provde moblty and servce portablty snce t s based on prmarly a wde-area concept. For ths faclty hybrd network (wreless LAN concept and cell or base-staton wde area network desgn) s requred. We need all dgtal packet networks that utlze IP n ts fullest form wth converged voce and data capablty. And ths great opportunty wll be fulflled by 4G wreless communcatons. In ths paper we manly deal wth nteroperablty the most mportant and specal characterstc of 4G wreless communcatons. Secton II descrbes the characterstcs of 4G and the motvaton for 4G other than 3G. Ths secton also dfferentates 4G from other generatons. Secton III focuses how 4G based on layered modfcaton provde nteroperablty. Secton IV descrbes about the role of cross layer at nteroperable ssue by 4G.. Journey towards 4G from 3G The lmtatons of prevous generaton lead to mproved generaton. Ther characterstcs vary dependng on some demandng features. Some of the dfferentatng characterstcs are gven n Table 1. The frst generaton of cellular networks conssted of analog systems capable of carryng only voce. G were packet swtched transferrng voce only. 3G s bascally a crcut swtched cellu-

53 D. MAHJABEEN ET AL. 473 lar network and so they have ther own gateway to nterpret IP from the back bone network. They also have ther own protocol and nterfaces for communcaton wthn themselves. To make ths problem end, the only soluton s 4G networks. Moreover, 3G s lackng from the followngs: lmtaton of spectrum allocaton, challengng gradually ncreasng bandwdth and hgh data rate for multmeda servce, dffculty to roam across dstnct servce envronment, lack of end envronment, and lack of end to end contnuous transmsson mechansm [3]. To face these challenges, the new level of mechansm, 4G communcatons s ntroduced. 4G s an all IP packet swtched network. Ths generaton s the upgrade strategy n world of wreless communcatons. 4G system s expected to provde a comprehensve and secure IP based soluton wth facltes lke voce, data and streamed multmeda. The key characterstcs of 4G are global moblty, servce portablty, scalablty and seamless handoff. It wll be very frendly to comprehensve lke Anytme, Anywhere, Anyhow and Always-on bass and at much hgher data rates compa- red to prevous generatons. Ths wll secure IP based soluton wth facltes lke voce, data and streamed multmeda. Another specal characterstc of 4G s the nteroperablty wth exstng wreless standards [4]. Ths generaton provdes ntegraton across dfferent network topologes.e., hybrd network archtecture that ntegrates wreless wde area networks wreless. 3. Interoperablty Wth the rapd development of varous wreless communcaton systems worldwde, there are also gradual changes n users expectaton and demand. Consequently the correspondng wreless networks work many fold at ther capacty lmts. So there s every chance of emergency crss and/or dsasters at the peak and crucal perod. Thus nteroperablty can offer network provders wth a possblty to swtch between alternatve wreless access networks. The basc theme of nteroperablty wll yeld the necessty of (user transparent) reconfgurablty and Table 1. Comparson of dfferent generatons. Propertes G 3G 4G Network Archtecture LAN, Wde area cell-based Hybrd - Drvng Archtecture Only voce domnantly voce; also data Converged data and voce over IP Swtchng Packet swtched Crcut and Packet All dgtal wth packetzed voce Rado Access FDMA, TDMA, CDMA WCDMA, CDMA000, IWC-136 MC-CDMA, OFDMA Database HLR,VLR, EIR, AuC EHLR, VLR, EIR, AuC EHLR, VLR, EIR, AuC Data rates 9.6 to 384 kbps Up to Mbps 100Mbps Roamng Restrcted Global Global Compatble Handsets Not comptable to 3G Dual mode TDMA and CDMA Voce and data termnals Comptable to G, G+ and blutooth Multple mode voce, data, vedo termnals Compatble to 3G Multple mode voce, data streamed vdeo at hgher data rates. Applcatons SMS, Internet Internet, SMS Internet, MMS, Mutmeda, HDTV, M TV Bandwdth 5 MHz 5-0 MHz 100 MHz Frequency Band Tr Band800, 900,1800,1900 MHz Dependent on country ( MHz) Hgher frequency bands (-8 GHz) Component desgn Optmzed antenna desgn Optmzed antenna desgn, mult-band adapters Smarter Antennas, software multband and wdeband rados FEC tech Convolutonal codng Convolutonal rate 1/, 1/3 Concatenated codng scheme IP No IP Connecton A number of ar lnk protocols, ncludng IP 5.0 All IP (IP6.0) Copyrght 010 ScRes

54 474 D. MAHJABEEN ET AL. cooperatveness n varous communcaton systems tunnelng towards the 4G journeys Reconfgurablty The reconfgurable nteroperablty can be done at the network level, the user level or both. Ths wll be very helpful for both the network provders and the users perspectve. The reconfgurable nteroperablty wll provde selecton between alternatve wreless access networks. The selecton could be based on several reconfgurable nteroperablty ssues such as: Channel state; Outage probablty; Vertcal handover probablty; Users QoS requrements; Context awareness; Load sharng and dstrbuton between dfferent spatally coexstng wreless networks; Effcent spectrum sharng; Preferred gateway selecton and network dscovery and Congeston control. The mechansm of reconfgurablty refers not only to the physcal layer, but span across the entre protocol stack (ncludng cross-layer optmzatons). The reconfgurable nteroperablty of the heterogeneous 4G system wll lead to more effcent end-to-end connectvty and servce delvery n heterogeneous envronments, easer worldwde roamng and dynamc adaptaton to regonal contexts, enhanced personalzaton and rcher servces. At the network level, the reconfgurable nteroperablty wll offer network provders wth a possblty to choose between alternatve wreless accesses networks at mnmal cost. At the user level, the nteroperablty of the heterogeneous 4G systems wll provde more effcent end-to-end connectvty and servce delvery n heterogeneous envronments, easer global roamng and dynamc adaptaton to regonal contexts, enhanced personalzaton and enrched servces [5]. 3.. Cooperatveness 3G specfy the PHY and MAC of the rado lnk. Ths alone s not adequate to buld an nteroperable broadband wreless network. Interoperable networks nvolve the followng ssues: End-to-end servce such as IP connectvty; Sesson management; Securty; QoS; Moblty; Connectvty ssues; Self organzaton; Authentcaton, Authorzaton, and Accountng. Cooperatveness comes to ensure these ssues. Ths means connectvty between all the enttes of a network n a consstent manner across all access technologes for any servce. In 4G, a cooperatve network (CoNet) conssts of three dstnct layers such as applcaton, connectvty and access that form logcally separate subsystems. Each of the layers can be further dvded nto dfferent sub layers as shown n the Fgure 1. The layers should have well defned nterfaces and be functonally ndependent of each other for an approach s requred to ensure easy adaptaton of heterogeneous access technologes, related technology changes, and flexble support for rapd servce nnovaton. Actually the connectvty layer plays an mportant role for cooperaton across varous realzatons of networks, whch n turn ensures the nteroperablty. Ths layer wll be ndependent of the varous transport technologes used to lnk the nodes of the network together. Fnally, the user wll enjoy seamless roamng across dfferent access technologes and admnstratve domans wthout any manual user nterventon [6]. One of the 4G s major goals s ntegraton, whch offers seamless nteroperablty of dfferent types of wreless networks wth the wre lne backbone. Some of the avalable attempted heterogeneous nteroperable ntegrated archtecture are: a loosely-coupled, Moble IPv6 (MIPv6)-based GPRS/WLAN/LAN heterogeneous network, mplementaton of IPv6-based moblty-enabled network archtecture wth Authentcaton, Authorzaton, Accountng and Chargng (AAAC) servces and support for Qualty of Servce (QoS) [7] Access Network From the pont of vew of access network, 3G access Applcaton Layer Connectvty Layer User Plane Control Plane Servce Applcaton Sub Layer Servce Support Sub Layer Network Control Sub Layer Transport Sub Layer Access Layer Management Plane Fgure 1. Layers of 4G provdng nteroperablty.

55 D. MAHJABEEN ET AL. 475 network uses WCDMA, cdma 000. But these are complcated and requre more protocol for system structure coverage. On the other hand, 4G access network uses the OFDMA, 3 RTT and MIMO antennas. Also hybrd multple access technque s used for hgh speed moble or nomadc user, data or voce traffc, call centre or boundary condtons [8]. But the abovementoned access technques currently do not nteroperate LAS CDMA (Large area synchronzed CDMA) access technque solves ths problem. LAS CDMA wll be compatble wth all current and future standards and there s a relatvely easy transton from the exstng system to LAS CDMA. Lnk ar emphaszes that LAS CDMA wll accommodate all the advanced technology planned for 4G. LAS CDMA wll also further mprove the prevalng the technques lke WCDMA, 3 RTT [1]. 4. Cross Layer In wreless network, nteroperable systems provde coordnaton among layers. Cross-layer desgn or cross-layerng provdes functonaltes assocated wth the orgnal layers to allow coordnaton, nteracton and jont optmzaton of protocols crossng dfferent layers. In order to provde mprovement n terms of some performance metrc, the cross-layer approach to system desgn derves from the nteracton among protocols operatng at dfferent layers of the protocol stack. The man advantage dervng cross layerng paradgm s the modularty n protocol desgn, whch enables nteroperablty and mproved desgn of communcaton protocols. An example of cross layer approach for nteroperablty s shown n Fgure. MAC-PHY Cross Layer: The physcal layer transmts power, whch can be tuned by the Medum Access Control (MAC) layer to ncrease the range of transmsson. NET MAC Cross Layer: Network layer could use NET-MAC-PHY Cross Layer Management Plane Network Layer MAC-PHY Cross Layer Management Plane MAC Layer Physcal Layer NET-MAC Cross Layer Management Plane Fgure. Cross-Layer approach for Interoperablty. MAC layer events lke handoff to reduce Moble-IP hand-off latency for seamless connectvty. NET-MAC-PHY Cross Layer: Ths layer provdes seamless connectvty and enhanced transmsson range [9]. 5. Conclusons For hgher data rates, hgher moblty support and seamless communcaton 4G utlzes a common platform that wll unfy a varety of evolvng access technologes, unnterupted nternetworkng and nteroperablty solutons and adaptve multmode user termnals. Reconfgurable, co-operatve and cross layer archtecture based on layered approach for nteroperablty are mentoned here. The co-net archtecture also provdes end to end servces, securty and self organzaton. More over usng multple descrptons codng at applcaton layer combned wth orthogonal frequency dvson multplexng at the lnk level provdes robustness aganst hostle wreless channels. Negotaton between applcaton, data lnk control and physcal layer s exploted to ncrease user qualty of servce n terms of pcture sgnal to nose rato and bandwdth effcency. 4G networks suffer from the lack of Layer QoS provsonng n heterogeneous networks, manly due to the non-unform nature of the QoS models and servce nterfaces among dfferent wreless technologes. Other problem s the lack of coordnaton of L3 QoS wth L QoS and moblty. All these problems can be solved by ntroducng QoS abstracton layer n between layer and 3 n the control plane whch wll be dscussed n our next paper. 6. References [1] Emergng Wreless Technologes: A Look nto the Future of Wreless Communcatons-Beyond 3G. ACF6-433F-B313-C [] D. I. Axots, F. I. Lazaraks and C. Vlahodmtro, Moblty and Traffc Parameters for Smulatng Interoperatng UMTS and HIPERLAN/ MTMR Enabled Networks, The 57th IEEE Semnual Vehcular Technologcal Conference, Jeju,Vol. 4, 003, pp [3] S. Hussan, Z. Hamd and N. S. Khattak, Moblty Management Challenges and Issues n 4G Heterogenous Networks, ACM, New York, USA, 006. [4] Evoluton of wreless connecton. my/most/mages/stores/dict/polcy [5] L. M. Gavrlovska and V. M. Atanasovsk, Interoperablty n Future Wreles Communcatons Sytems: A Roadmap to 4G, Mcrowave Revew,Vol. 13, No. 1, 007, pp Copyrght 010 ScRes

56 476 D. MAHJABEEN ET AL. [6] Cooperatve Networks of 4G. www-scf.usc/edu/~ssaraf/ EE555.pdf [7] L. Gavrlovska, V. Atanasovsk, V. Rakovc, O. Ognenosk and A. Momrosk, Provdng Interoperablty n Heterogeneous Envronments towards 4G, ELMAR, 50th Internatonal Symposum, Zadar, Vol. 1, 008, pp [8] L. D. Uomo and E. Scarrone, All-IP 4G Network archtecture for Effcent Moblty and Resource Management, 5th Internatonal Symposum on Wreless Personal Multmeda Communcatons, Honolulu, Vol., 00, pp [9] D. Klazovch, M. Devetskots and F. Granell, Formal Methods n Cross Layer Modelng and Optmzaton of Wreless Networks, In: Kotsopoulos, S. and Ioannou, K. Ed., Handbook of Research on Heterogeneous next Generaton Networkng: Innovatons and Platforms, 009, pp. 1-4.

57 Int. J. Communcatons, Network and System Scences, 010, 3, do:10.436/jcns Publshed Onlne May 010 ( Research on Access Network Intruson Detecton System Based on DMT Technology Abstract Lngx Wu, Je Zhan, Qange He, Shuyan He Hunan Unversty of Scence and Technology, Xangtan, Chna E-mal: Receved January 19, 010; revsed March 1, 010; accepted Aprl 7, 010 Analyss s done on the nter-carrer nterference (ICI) that caused by mult-carrer communcaton system frequency offset. The applcaton model of DFT/IDFT n ADSL access network s analyzed further; the hardware detecton and software analyss scheme of the system are proposed for the accessng network. Experments have proved that montorng system can flter the network data flow and carry on statstcal and analyss, achevng real-tme montorng. Keywords: DMT, ICI, Intruson Detecton, DFT/IDFT 1. Introducton The Dscrete Mult-Tone DMT () technology has been appled successfully on the ADSL (Asymmetrc Dgtal Subscrber Lne) transmsson system, and has developed the broadband transmsson system that based on Twst- Par. The problems of network detecton and montorng wll be nherent n the development of network, yet the rapd development of the network has been ahead of the real-tme montorng. To solve t, data detecton system based on DMT technology has been studed systematcally, and data acquston equpment has been devsed, whch can acheve flterng analyss and statstcs of the network data stream wth no nfluence on the user and the phone company end of the lne.. Mult-carrer Communcaton System Modelng QAM (Quadrature Ampltude Modulaton) s the bass of DMT, Model use multple QAM constellaton dagram encoders, and each constellaton dagram encoder use a dfferent carrer frequency, The DMT code element that were formed by summng all the carres transmtted through the channel. If the recever can separate sne waves from cosne waves on dfferent frequences, each wave can be decoded ndependently, the method of encoded and decoded are consst wth the QAM sgnals; to ensure no nterference from f 1 to f n sub-channel, we must make sure that a sne and cosne wave n one sub-channel are orthogonalty wth any other sub-channels, and ts formula s as follow [1,]: T 0 T 0 T 0 cos( nt)cos( mt) dt 0 cos( nt)sn( mt) dt 0 sn( nt)sn( mt) dt 0 n and m are unequal ntegers, and ω s the base rate. By the expresson of orthogonalty, we concluded the each sub-channel frequency must be an ntegral multple of base frequency, and the code element perod T s recprocal of the base frequency or an ntegral multple of the recprocal of the base frequency. Two stuatons would appear: Frst, the frequency offset s an ntegral multple of sub-carrer; second, the frequency offset s not an ntegral multple of sub-carrers; both of two stuatons wll make system characterstcs deterorate. Assumng the number of carrer s lmted, Fgure 1 s a block dagram of a DMT communcaton system model [3], and accordng to the system, we make the followng dscusson: Durng the symbol cycle, Assumng the orgnal data symbol s{a 0,, a 1,,, a n-1, }, after IDFT calculate, we can get: N 1 1 j lk bk, al, exp N l 0 N (1) Therefore, we can get the output sgnal x(t) as follow: x( ) exp( ) N t j fct b p t kt 1 k, k 0 N ()

58 478 L. X. WU ET AL. hgh-speed Data stream splt nto low speed Data stream zz 0, 0, y 0, 0, n-pont DFT zz N-1, N 1, y N N 1 1, b a 0, 0, b0, 0, N-pont parallel/seral IDFT converson b N-1, N-1, bn a 1, N 1, exp(-jπ(fc+δf)t) ( f f) Seral/parallel Converson y(t) c DAC and LPF ADC and BPF exp(j f t) x(t) h(t) n(t) nt () c Fgure 1. Mult-carrer communcaton system modelng. f c represent the carrer frequency, p(t) represent the mpulse response of low-pass flter used n the transmtter system, but there s the frequency devaton Δf at the recevng end, after down-converson and low-pass flter the y(t) sgnal s : 1 () exp( 0) N kt yt j ft c bk, qt k 0 N (3) q(t) represent the combnaton mpulse response get by multplyng low-pass flter of the transmtter and bandpass flter of the recever, θ 0 s the phase dfference between recever local oscllator and RF carrer. If q(t) can meet the Nyqust crteron at the moment kt/n, then we sample y(t) at the same tme. We can get: jfkt yk, exp( j 0) bk, exp N Accordng to the DFT formula, (4) N 1 j km Zm, yk, exp, ( m 0,1,, N 1) k 0 N (5) Substtutng (1) and (4) nto (5), we can get: N1 N1 1 j k( lmft) m, exp( 0 ) l, exp N l0 k0 N Z j a (6) N 1 k Accordng to the sum formula u = 1 N u, (1-6) k 0 1 u can be smplfed: N 1 1 1exp j lmft Zm, exp( j0 ) a l, (7) N l0 j lmft 1 exp N From: 1 exp( j ) [exp( j) exp( j)]exp( j) jsnexp( j) Make: 1 l m ft lmft, N (8) can be express as follow: c N m, exp( 0 ) l, ( ) N l0 jsnexp( j) N 1 1 sn exp( 1 j0 ) al, exp( j ( 1 )) N l0 sn Z j a 1, exp( 0 ) N m l, lm l 0 jsn exp( j ) (8) Z j a c (9) Among of them, lm 1 sn( ( lmft)) ( N1)( lmft) exp j N ( lmft) sn( ) N N (10) 1 sn( ft) N 1 c0 = exp jft N ft sn( ) N N (11) c 0, c 1,, c n-1 are complex weghtng coeffcents, correspondng to nput data symbols a 0., a 1,,, an 1,, then we can get the symbols transmtted n mth sub-channel as follow (N s the number of coeffcents): 1 Z exp( j ) N c a m, 0 lm l, l0 1 exp( j ) c a exp( j ) N c a (1) 0 0 m, 0 lm l, l 0 lm

59 L. X. WU ET AL. 479 The frst tem of the formula s data symbol of weghted mathematcal expectaton, the second tem s the ICI caused by Δf. If Δf = 0, then Z m, = exp(jθ 0 ) a m,, (m = 0, 1,,N 1). Note: Each complex symbol wll be nfluenced by the phase devaton factor θ 0. If Δf 0, the nter-channel nterference (ICI) wll occur. Fgure shows the relatonshp between the real part, magnary part, modulus of the complex weghtng coeffcent and the sub-carrer number N n case of the two knds of Δf T. When the frequency devaton ncreases, the stable zone quckly narrows, the modulus value rapdly ncreases, ndcatng ICI ncreases sgnfcantly 3. ADSL System Based on Mult-carrer Technology Accordng to the model [3,4], assume T s the cycle, we derve the waveform expresson that added up sne and cosne waves : Xn cos( nt) Ynsn( nt) 0t T St () (13) 0 else The waveform shows the nfluence that a sngle subchannel n operate on DMT code element, accordng to Nyqust theorem, samplng the sgnal, samplng frequency s Nf, samplng value s: k k SK Xn cos( n ) Ynsn( n ) Nf Nf (14) nk nk Xn cos( ) Ynsn( ) 0k N 1 N N Make the Dscrete Fourer Transform (DFT) to these N ponts as follow: S nk nk N 1 j mk/n m Xn cos( ) Ynsn( ) e k 0 N N N 1 jnk / N jnk / N jnk / N jnk / N e e e e Xn Yn k 0 j j mk/ N e N(Xn jy n) m n N(Xn jy n) m N-n 0 else (15) From (14) and (15), we conclude that the output can be mapped to a complex number by makng DFT to the sgnal, the value of encoder X-axs (cosne ampltude) represents the real part of the complex number, the value of Y-axs (sne ampltude) represents the magnary part of the complex number, then t s a way to generate DMT code element. If make Inverse Fourer transform to S m, we can deduce S k : S K N 1 1 j mk/n Sme N m0 1 (X Y ) jmk/ N (X Y ) j( Nn) k/ N n j n e n j n e 1 nk nk (X n j Y n )(cos( ) j sn ( )) N N nk nk (Xn jy n)(cos( ) jsn ( )) N N nk nk Xcos( n ) Ysn( n ) 0k N 1 N N (16) 0.8 real lmag abs 0.8 real lmag abs Weghted coeffcent Weghted coeffcent deltaf T = 0.1 deltaf T = 0.3 Fgure. Carrer frequency devaton and synchronzaton features.

60 480 L. X. WU ET AL. The DMT modem can be acheved wth DFT and IDFT. From (13) t can be derved the complex number N(X n - jy n ) to the n th sub-channel, express that a complex number can represents a sub-channel of DMT, N sub-channels have N complex numbers, plus N conjugate complex numbers (X n + jy n ), we can get N complex numbers, from (14),we can get S k through makng IDFT to N complex numbers. So we can get DMT modulaton, demodulaton program, ths program has been appled to ADSL modem. In the ADSL, ATU-C downstream modulator uses 56 wndows, whch s 56 complex numbers, the nterval of wndows s khz. Frequency range s from khz to MHz, accordng to code analyss, the IDFT of downstream DMT can be expressed as: 511 j mk/56 SK Sme k m0, 0,,511 (17) S m s the complex number value or expanded conjugate complex number made by QAM constellaton encodng for each sub-carrer, S k s the tme-doman sample sequence after DMT modulaton, and the tme-doman waveform can be generated after parallel-seral converson and DAC. The upstream DMT modulaton of ATU-R uses 3 wndows, 3 complex numbers represent the codng results of each sub-channel constellaton, the audo nterval s khz, frequency range from khz to 138 khz. Accordng to the code analyss that the DMT of IDFT n the upstream can be expressed as: 63 j mk/64 SK Sme k m0, 0,,63 (18) 4. Research on Intruson Detecton System The prncple s shown n Fgure 3. The structure of collector manly conssts of the DSLAM Smulaton Module [5-7], Modem Smulaton Module and Data Interface Module and so on. The end of ADSL Modem accesses to the DSLAM smulaton module of data acquston equpment, the end of Telecommuncatons Bureau accesses to the ADSL Modem Smulaton Module of data acquston. After the upstream sgnals nput DSLAM USB CY7C68001 Data Interface Module FPGA UTOPIA CTRL_E UTOPIA CTRL_E XTAL Memory MTC MTC XTAL Memory Drver Drver DSLAM Smulaton Module POTS spltter POTS spltter Modem Smulaton Module end-user the Central offce Fgure 3. Hardware block dagram of data acquston system.

61 DPL. X. WU ET AL. 481 UDP datagram PIP data segment Data Processng Protocol AnalyssUAnalyss data Data acquston fle fle data data packet processng database and and data output data analyss layer layer fle fle data data audo audo data data user data Web Web data data mal mal data data I(Telnet (Telnet FTP HTTP SMTP POP IP Phone etc.) etc.) applcaton protocol analyzer layer layer (TCP,UDP) (TCP UDP) Transmsson protocol analyzer layer Transmsson protocol analyzer layer IP IP data segment Applcaton data UDP datagram (IP ICMP (IP ICMP ARP etc.) ARP etc.) Internet protocol analyzer layer Internet protocol analyzer layer IP datagram SIGNAL ACQUIRING DEVICE Fgure 4. System software descrptons. smulaton module, we complete DMT demodulaton and send the demodulated sgnal to the nterface module, then send to the computer by USB nterface. Meanwhle, the upstream sgnal s also sent to the ADSL Modem Smulaton Module, complete the transmsson of upstream data to Telecommuncatons Bureau. After downstream sgnals nput the ADSL Modem Smulaton Module, we complete DMT demodulaton and send the demodulated sgnal to the nterface module, and then send to the computer through USB nterface. Downstream sgnal s also sent to DSLAM Smulaton Module, and complete the transmsson of downstream sgnal to the user. The normal sgnals of consumer are connected drectly wth the dedcated POTS access between the two smulaton modules. So, the devce joned Telecommuncatons Bureau and user ADSL Modem, t wll not affect the user s normal voce and data communcatons, both of them are not aware of the exstence of the devce. Through the UTOPIA nterface [8,9], we extract the cell, and encapsulate the cell to get USB packets, and transmt to the computer through the USB nterface to analyses the data, the process ncludes two parts (data processng and protocol analyzer), n order to restore the data effectvely and accurately from the obtaned data, we must make out the software, accordng to the system request the software herarchy s shown n Fgure Conclusons The dstance of data transmsson on the Twst-Par s lmted, and the varety of crcut characterstcs wll affect the status of crcut connecton whch may cause normal users can t explore the Internet, Meanwhle network data transferrng s bdrectonal and the upstream and downstream data transferrng are asymmetry that makes a great deal of dfference from other wred or wreless audo and vdeo sgnal transmsson. In ths paper, the data acquston system s a data recevng system. In process of the upstream and downstream data processng, there are several techncal dffcultes n data extracton, separaton, storage, etc, and our program can solve these problems well. It brngs new solutons to the data acquston system for detectng network data transmsson and elmnatng network falure, partcularly t can solve the problems of montorng the real-tme. 6. Acknowledgements Ths paper s supported by Natural Scence Foundaton

62 48 L. X. WU ET AL. of Hunan Provncal, Chna (Grant No. 07JJ618). 7. References [1] W. Stallngs, Data and Computer Communcatons, Prentce Hall Upper Saddle Rver, New Jersey, 000. [] T. Pollet and M. Moeneclaey, Synchronzaton of OF- DM Sgnals, IEEE Globecom 95, Sngapore, Vol. 3, November 1995, pp [3] H. V. Poor, Iteratve Multuser Detecton, IEEE Sngal Processng Magazne, Vol. 1, No. 1, 004, pp [4] ITU-T Recommendaton G.99.3, Asymmetrc dgtal subscrber lne transcevers (ADSL), July 00. [5] J. Henanen, RFC1483: Mult-protocol Encapsulaton over ATM Adaptaton Layer 5, RFC Edtor, Unted States, [6] L. X. Wu, Z. P. Huang, G. L. Tang and L. Y. Pan, Desgn and Realzaton of Data Acquston Access Crcut Based on ATM Network, Telecommuncaton Engneerng, Vol. 44, No. 6, 004, pp [7] AFE-0154 Data Sheet rev.1- June [8] L. X. Wu and T. F. Jang, Research and Aealzaton of Real-tme Data Acquston System Based on ADSL Access Network, Engneerng Journal of Wuhan Unversty, Vol. 39, No. 6, 006, pp [9] CY7C68001 EZ-USB SX Hgh-Speed USB Interface Devce, Cypress Semconductor Corporaton, July 003, pp

63 Int. J. Communcatons, Network and System Scences, 010, 3, do:10.436/jcns Publshed Onlne May 010 ( Synchronzaton n Wreless Networks for Practcal MIMO-OFDM Systems Abstract Muhammad Khurram Kyan 1, Muhammad Usman Ahmed, Asm Loan 1 1 Department of Electrcal Engneerng, UET, Lahore, Pakstan Department of Computer Scences, LUMS, Lahore, Pakstan E-mal: {khurramkyan, muahmed}@gmal.com, aloan@uet.edu.pk Receved February 1, 010; revsed March 4, 010; accepted Aprl 5, 010 In ths paper a frequency offset estmaton technque for Wreless Local Area and Wreless Metropoltan Area Networks s presented. For frequency offset estmaton, we have appled a low-complexty frequency offset estmator for smple AWGN channels to fadng channels for MIMO-OFDM systems. Smulaton results have shown that the performance of the proposed estmator s better than the low complexty frequency offset estmator desgned for AWGN channels. Keywords: Synchronzaton, Multple Input Multple Output, Orthogonal Frequency Dvson Multplexng, Wreless Local Area Networks, Wreless Metropoltan Area Networks, Stanford Unversty Interm Channels 1. Introducton A lot of research has been carred out n carrer frequency offset estmaton for Sngle Input Sngle Output (SISO) Orthogonal Frequency Dvson Multplexng (OFDM) systems but comparatvely less work has been done n Multple Input Multple Output (MIMO) OFDM systems. In [1], tmng metrc for frame synchronzaton and frequency offset estmaton n OFDM s proposed n the downlnk. In [], a coarse tmng synchronzaton s carred out by usng autocorrelaton and then Carrer Frequency Offset (CFO) s estmated by performng precse autocorrelaton only on samples that have been compensated for coarse tmng synchronzaton. Both [1] and [] have suffcently explored OFDM but they do not ncorporate MIMO. However, n [3], a novel frequency synchronzaton scheme s presented whch uses repeated pseudo-nose tranng sequences to correct CFO n MI- MO-OFDM systems. Also, n [4], nteger CFO and fractonal CFO are estmated for MIMO-OFDM systems through specal tranng sequences by solvng complex or real polynomal correspondng to the cost functon. Although both [3] and [4] have dovetaled MIMO wth OFDM systems but they lack the practcalty as they have not ncorporated any partcular standard. In ths paper we have extended the work of Luse & Regga- nnn (L & R), [5], by adaptng ther AWGN sngle channel frequency estmator to multpath fadng channels usng IEEE Standards [6], and IEEE 80.11n Standards [7]. The technque used s non-recursve as realstc MIMO scenaros ental changng channels wth nformaton beng sent n bursts correspondng to the duraton of the coherence tme of the channel. The remanng paper s organzed as follows. Secton covers the detals of system model. Secton 3 gves the descrpton of the proposed frequency offset estmator and Secton 4 explans the results obtaned through smulatons. Secton 5 fnally gves the concluson.. System Model A frequency offset can be ntroduced by relatve moton between the transmtter and the recever (Doppler spread) and by the naccuraces n the Local Oscllator (LO). Channel estmaton n MIMO-OFDM system s very senstve to any frequency offset n the down converted sgnal because frequency offset ntroduces a tme dependant factor that degrades the estmaton of channel response matrx H. Therefore accurate frequency offset estmaton would result n a better channel estmate and a more robust system. Generally, a multpath fadng channel s changng therefore the transmsson s done n packets wth the packet length beng governed by the coherence tme of the channel,.e., the tme for whch the channel response

64 484 M. K. KIYANI ET AL. does not change. In packet based communcatons, the channel response matrx H must be estmated for each packet. Ths s generally done by usng a tranng sequence known to the recever. We have used SUI Channels, [8], here as multpath fadng channels and a preamble specfed n IEEE and IEEE 80.11n as a tranng sequence for WMAN and WLAN respectvely. Our proposed algorthm estmates a constant frequency offset over a length of symbols for every packet. In MIMO, the sgnal receved at each recevng antenna s the superposton of the transmtted sgnals that from dfferent transmt antennas. Thus, the tranng sgnal for each transmt antenna needs to be transmtted wthout beng nterfered by the others. Fgure 1 shows three transmsson patterns that avod nterferng wth one another: ndependent, scattered and orthogonal patterns. For the sake of brevty we wll only dscuss the ndependent pattern. The ndependent pattern transmts tranng sgnal from one antenna at a tme whle the other antennas are slent, thus guaranteeng orthogonalty, n the tme doman, between each tranng sgnal. The ndependent pattern s often the most approprate for MIMO-OFDM, snce the preamble s usually generated n the tme doman. To encode the tranng sequence we have used the ndependent pattern and assgned to each transmt antenna a standard tranng sequence. Ths means that at a gven tme only one transmt antenna s transmttng the tranng sequence as shown n Fgure. We have also assumed that the dstance between transmttng antennas s less than λ/ and they all encounter the same channel statstcs. 3. Proposed Frequency Offset Estmaton Technque The tranng sequence symbol s transmtted from the m th transmt antenna and receved on any of the receve antenna. The kth samplng ndex of ths receved symbol can be wrtten as: [ πvkt+ θ] j rk () = hce + nk () for1 k N. (1) m k In the above equaton, h m denotes the complex channel coeffcent between the mth transmt antenna and a specfc receve antenna, v s the frequency offset that needs to be estmated and n k s the complex whte Gaussan nose wth varance N o. In general, there would be multple receve antennas and each receve antenna would have a separate LO that would ntroduce some frequency drft; ths means that frequency estmaton needs to be done separately for each receve antenna. The frequency offset v could possbly be dfferent for dfferent sub-carrers because Doppler nduced frequency shfts depend on the wavelength of the transmsson. Data modulaton can be removed by multplyng r(k) wth c k *, because for PSK constellatons c k c k * = 1 and the tranng sequence z(k) s known(the tranng sequence symbols, taken from IEEE Standards 80.11n and 80.16, are assumed to be a Phase Shft Keyng (PSK) constellaton such as Quadrature Phase Shft Keyng). R * j πvkt + θ zk ( ) = rk ( ) c k = hme + nk ( ). () The autocorrelaton of the z(k) sequence s defned as: p 1 N z( k) z ( k p) ( 1 ) N + p k = p for1 k, p N. (3) ( ) Substtutng () n (3) we get the followng expresson for R(p): Fgure 1. Three dfferent patterns for transmttng tranng sgnals n MIMO-OFDM systems [9].

65 M. K. KIYANI ET AL. 485 For a general case of MIMO systems, N transmt antennas would results n N terms n our proposed correlatons. 4. Smulaton Results Fgure. Space Tme Encodng of the tranng sequence. R jπ pvt p e + n ( p), (4) ( ) where h m s normalzed to be equal to 1 and n''(p) represents the nose related(nose-nose and nose-sgnal) terms after substtuton. We can fnd a frequency offset estmate now usng the formula of L&R gven below: 1 v = π( N + 1) T N arg R( p ). p = 1 The summaton of R(p) n (3) serves to smooth out the nose as t s a movng average flter whch s low pass and deal for nose smoothng. In our MIMO-OFDM system we propose to go one step further and cross-correlate R(p) wth a tranng sequence transmtted from the second antenna n the next tme slot wth the same channel coeffcent h m as shown n Fgure 3.The packet length s assumed to be longer than multple tme slots and the channel remans constant over a packet length. The cross-correlatons gve a greater nose-averagng gan. For ths case the term wthn the summaton n Equaton (3), z(k)z * (k-p) s replaced by auto correlatons and cross- correlatons of the symbol transmtted from frst and second antenna, respectvely: ( ) 1 zk ( ) z k p + z() l z ( l p) +. 4 z( k) z ( l p) + z() l z ( k p) (5) (6) Here we explore the performance of proposed frequency offset estmator for multpath fadng channels usng Mean Square Error (MSE) as a performance metrc. We have used SUI channel 1 and SUI channel 3 as multpath fadng channel for our smulatons. Intally we dscuss the results of WLAN and then the results of WMAN wll be dscussed subsequently. For WLAN we have used 5 GHz lcensed band wth nomnal channel bandwdth of 0 MHz. The transmtted carrer frequency of both the base staton and subscrber staton should have accuracy better than ± as per IEEE Standards n. The value should reman vald over a gven temperature range and tme of operaton.e., ageng of equpment. Keepng aforementoned n vew the maxmum carrer frequency offset comes out to be 00 khz. A packet sze of 1KB s assumed. Unt delay of channel s assumed to be the same as OFDM sample perod. In Fgure 4 the orgnal curves refer to estmatng the frequency offset by takng autocorrelatons of the z(k) sequences whereas the modfed curves refer to usng auto and cross correlatons of z(k) and z (l), respectvely for SUI 1 channel. The curves n ths fgure are generated for the cases of two transmt antennas. The modfed curves, as per the proposed algorthm, provde a log (N) db nose averagng gan n AWGN condtons. However, n the presence of multpath fadng, t s not possble to see the complete nose averagng gan, especally for zk ( m) zl ( m) zk ( ) zl () h m Fgure 3. Auto and cross correlatons of the tranng sequences of two transmt antennas. Fgure 4. Performance of the proposed estmator for SUI 1 channel.

66 486 M. K. KIYANI ET AL. hgh E b /N 0 values. Ths s because sgnal attenuaton due to fades overshadows the effects of nose. Smlarly the performance of the proposed algorthm for MIMO-OFDM system for the case of SUI 3 channel s shown n Fgure 5 below. Same smulaton parameters are used for both SUI 1and SUI 3 channel. For WMAN We have used 3.5 GHz lcensed band wth nomnal channel bandwdth of 3.5 MHz The transmtted carrer frequency of both the base staton and subscrber staton should have accuracy better than ± as per IEEE Standard d and the maxmum carrer frequency offset comes out to be 70 khz. A packet sze of 1 KB s assumed. In Fgure 6 below the orgnal curves refer to estmat- Fgure 5. Performance of the proposed estmator for SUI 3 channel. ng the frequency offset by takng autocorrelatons of the z(k) sequences whereas the modfed curves refer to usng auto and cross correlatons of z(k) and z(l), respectvely for SUI 1 channel. The curves n ths fgure are generated for the cases of two transmt antennas. Smlarly the performance of the proposed algorthm for MIMO-OFDM system for the case of SUI 3 channel s shown n Fgure 7 below. Same smulaton parameters are used for both SUI 1 and SUI 3 channels. The salent aspects of the smulated results are analysed as under:- 1) The modfed curves, as per the proposed algorthm, provde a log (N) db nose averagng gan n AWGN condtons. However, n the presence of multpath fadng, t s not possble to see the complete nose averagng gan, especally for hgh E b /N 0 values. Ths s because sgnal attenuaton due to fades overshadows the effects of nose. ) It s qute evdent that the modfcaton suggested n ths paper can reduce the MSE sgnfcantly for lower values of E b /N 0. The complexty of the system may have ncreased but t may be traded-off for more accurate frequency offset estmaton. 3) Performance of the proposed algorthm s dependent, apart from other factors, on the length of the tranng sequence. The tranng sequence used for WLAN s greater n length as compared to the tranng sequence of WMAN. Resultantly the results of WLAN are better as compared to WMAN especally for hgher values of E b /N 0. 4) Complete executon of the proposed algorthm requres the symbols transmtted n adjacent tme slots to be receved at the recever. 5. Conclusons In ths paper an effcent frequency offset estmaton Fgure 6. Performance of the proposed estmator for SUI 1 channel for WMAN. Fgure 7. Performance of the proposed estmator for SUI 3 channel for WMAN.

67 M. K. KIYANI ET AL. 487 technque for MIMO-OFDM systems n multpath envronment s presented. Smulaton results have shown that synchronzaton problems n MIMO-OFDM systems can be solved wth proposed algorthm whch gves good performance and tends to be lmted only by multpath fadng. Snce our extenson of the smple L & R estmator to the MIMO-OFDM case deals wth data encodng and not wth the fnal estmaton step, we have preserved the optmalty property of the L & R estmate. 6. References [1] C. N. Kshore and V. U. Reddy, A Frame Synchronzaton and Frequency Offset Estmaton Algorthm for OFDM System and ts Analyss, EURASIP Journal on Wreless Communcatons and Networkng, Vol. 006, 006, pp [] T.-H. Km and I.-C. Park, Two Step Approach for Coarse Tme Synchronzaton and Frequency Offset Estmaton for IEEE 80.16d Systems, IEEE Workshop on Sgnal Processng Systems, Shangha, October 007, pp [3] L.-M. He, Carrer Frequency Offset Estmaton n MI- MO OFDM Systems, 4th IEEE Conference on WCOM, Dalan, 1-14 October 008, pp [4] Y. X. Jang, X. H. You, X. Q. Gao and H. Mnn, Tranng Aded Frequency Offset Estmaton for MIMO OFDM Systems va Polynomal Routng, 67th IEEE Vehcular Technology Conference, Sngapore, May 008. [5] M. K. Kyan, M. U. Ahmed and A. Loan, Synchronzaton n Fxed Broadband Wreless Access for Practcal MIMO OFDM systems, 9th IEEE Malaysan Internatonal Conference on Communcatons, Kuala Lumpur, December 009. [6] M. Luse and R. Reggannn, Carrer Frequency Recovery n All-Dgtal Modems for Burst-Mode Transmssons, IEEE Transactons on Communcatons, Vol. 43, No. -4, 1995, pp [7] IEEE P80.11n/D11.0 Draft Standard for Informaton Technology, Telecommuncatons and Informaton Exchange between Systems, Local and Metropoltan Area Networks Part 11: Wreless LAN Medum Access Control (MAC) and Physcal Layer (PHY) Specfcatons, 009. [8] K. V. Erceg, S. Har, M. S. Smth, D. S. Baum, et al., Channel Models for Fxed Wreless Applcatons, IEE- E Task Group Contrbutons, 1 February 001. [9] G. J. Andrews, A. Ghosh and R. Muhamed, Fundamentals of WMAX Understandng Broadband Wreless Networkng, Prentce Hall Communcatons Engneerng and Emergng Technologes Seres, Prentce Hall, February, 007.

68 Int. J. Communcatons, Network and System Scences, 010, 3, do:10.436/jcns Publshed Onlne May 010 ( A Cross Layer Optmzaton Based on Varable-Power AMC and ARQ for MIMO Systems Abstract Shu-Mng Tseng, We-Shun Le Department of Electronc Engneerng, Natonal Tape Unversty of Technology, Tawan, Chna E-mal: shumng@ntut.edu.tw, gavn_le@unhancorp.com Receved February 10, 010; revsed March 9, 010; accepted Aprl 7, 010 To mprove spectrum effcency (SE), the adaptve modulaton and codng (AMC) and automatc repeat request (ARQ) scheme have been combned for MIMO systems. In ths paper, we add varable power subject to power constrant n each AMC mode. We use KKT optmzaton algorthm to get the optmal transmt power and AMC mode boundares. The numercal results show that the average SE s ncreased by about 0.5 bps/hz for MIMO systems wth Nakagam fadng wth parameter m = when SNR s around 15 db and the ARQ retransmsson s twce. Keywords: Transmtter Power Control, Spectrum Effcency 1. Introducton The demand for hgh data rate and qualty of servce (QoS) n wreless networks requres cross layer approach [1]. The adaptve modulaton and codng (AMC) are already consdered for mplementaton n many wreless system standards. In order to mprove the average spectrum effcency (SE), a combnng constant-power AMC scheme n physcal layer and truncated automatc repeat request (ARQ) protocol whch provdes a trade-off between the average codng rate and the probablty of undetected error at the data lnk layer (DLL) n sngle-nput sngle-output (SISO) system []. In ths paper, we add power adaptaton proposed n [3] to the constant-power AMC and ARQ n MIMO systems n [4]. In the proposed adaptve-power AMC and ARQ n MIMO systems, the power can be changed to ncrease average SE. The packet error rate (PER) n [,3] whch s much smaller than the target PER, so we are motvated to ncrease the PER, make PER as close to the target PER as possble. In ths way, the swtchng SNR level of each rate boundary of each rate shft left n the SNR axs and we move to hgher mode earler as usng hgher order modulaton, or hgher code rate and the average SE can be mproved. However the leftmost part of one each SNR regon can have PER exceed lmt, so we need adaptve power to compensate t. In [4] the Lagrangan multplers λ s only one. In the propose method, the Lagrangan multplers λ s dfferent accordng to each AMC mode. Numercal results ndcates that the proposed optmzaton algorthm whch combne adaptve power AMC scheme at physcal layer and truncated ARQ protocol at data lnk layer can ncrease the average SE. Ths paper s organzed as follows: Secton descrbes the system model. Secton 3 presented our proposed scheme wth adaptve power. Numercal results are presented n Secton 4, and our concluson s n Secton 5.. System Model We consder a SISO system whch combnng the AMC scheme wth power control at the physcal layer and the truncated ARQ module at the data lnk layer, as shown n Fgure 1. We assume channel gans reman nvarant durng a packet, but vary from packet to packet. Let R n be the rate of the mode. S denotes the average transmt power, γ denotes the pre-adaptaton receved SNR whch the recever feed back to the transmtter, and Sn ( ) denotes the allocated power n the AMC mode n. The AMC s performed by dvdng the range of the channel SNR nto N + 1 non-overlappng consecutve nterval, denoted by [ n, n 1), n 0,1,..., N, 0, N 1 and N s the number of AMC modes. No data s sent at [ 0, 1) SNR range whch corresponds to the outage mode. Consder of power adaptaton, we modfy the PER expresson n the mode n [,3] as follows:

69 S.-M. TSENG ET AL. 489 PER n 1, 0 Sn ( ) an exp gn, pn S The mode dependent parameters { an, gn, pn} are gven n Table 1. The mode-swtchng SNR values n [,3] (wthout power adaptaton) can be found by assumng Sn ( )/ S 1 and PERn Pt where P t s the target PER (0.01 usually) n (1). If consderng adaptaton power, we have two unknown elements, Sn ( ) s the adaptve power nsde the mode n and s the swtchng SNR to mode n. So we need to solve adaptve power under each mode frst, then we can look for the mode-swtchng SNR values. 3. Adaptve-Power AMC Because the AMC algorthm s the same for all SISO sub-channel, so we only dscuss the AMC scheme wth truncated ARQ protocol n the th sub-channel n ths secton. Here, we propose adaptve-power AMC to fnd the optmal SNR swtchng level that maxmze SE under the PER constrant. The nstantaneous PER s smaller than pn (1) target PER, PER ( ) P n t, ; n n n 1,..., N, 1 where P t denotes the target PER, so average PER wll lower than target PER, too. We now propose that, PER ( ) P n t, and add power factor to make the average SE larger. From (1), we can fnd the power adaptaton wth PER constrant n mode n s Sn( ) 1 a ln( n ) () n n1 S g P n t Usng (), we now have the PER constrant as follows 1 an n1 1 ln( ) p ( ) d Pr( n) g P (3) n n t We want to fnd the close form of the above equaton, so we frst fnd the close form of the followng: (dfference from tradtonal, addton 1/ factor) 1 n p d n1 Pr pow ( ) ( ) n n 1 1 ( ( ) ( n E E )), m 1 mn mn 1 ( m, ) ( m, ) m ( ), m ( m) (4) Input h(t) n Buffer Transmtter X + Recever Buffer Power Control Channel Estmator ARQ Generator ARQ Controller AMC h ˆ( t) Fgure 1. System model. Table 1. Transmsson modes n TM AMC scheme wth convolutonal coded M-QAM modulaton. Mode 1 Mode Mode 3 Mode 4 Mode 5 Mode 6 Modulaton BPSK QPSK QPSK 16-QAM 16-QAM 64-QAM Codng rate 1/ 1/ 3/4 9/16 3/4 3/4 R (bt/sym.) n a n g n γ pn (db) Con-Power Adapt-Power

70 490 S.-M. TSENG ET AL. t when E( x) e / tdt s the exponental ntegral x functon, and has no close form. The optmzaton program can be formulated as Maxmze R Pr( n) N n n1 N N an pow n1gn Pt n1 1 subject to ln( )*Pr ( n ) Pr( n ) (5) The constrant n (5) s from (3) and (4). Usng the KKT soluton to solve, we got that: L(,,, ) 1 n N N 1 an n n pow n1 n1gn Pt R Pr( n) ( ln( ) (Pr ( n) Pr( n)) where are the Lagrangan multplers. The optmal * * * soluton ( 1,,..., N ) and the correspondng Lagrangan multplers, * n must satsfy the followng condtons: L * * * * 1 ) ( 1,, n, 1,..., N) 0, n 1,,, N n (6) 1 a ) ln( )Pr ( ) Pr( n) N N n pow n n1 gn Pt n1 N N 1 Sn p d S N n1 * n * 3 ) ( ) ( ) 4) 0 5 ) n pn, n 1,,, N (7) * * * so that he optmal SNR swtchng level, ( 1,,..., N ) wll larger than the bound pont constrant, pn. The general form of the optmal mode swtchng levels can be wrtten as ln( a / P), * 1 1 * 1 * 1 g1( 1 R1) * * * n 1 n n1 1 n n1 n 1 n * * gg n n1( Rn Rn 1n n 1) g ln( a / P) g ln( a / P), n,..., N we have the optmal mode swtchng levels, checkng the * * * swtchng levels, ( 1,,..., N ) satsfy the constrant n (5), and the optmal mode swtchng levels. Fnally, the proposed algorthm s summarzed n Fgure. (8) Usng [,(11)]decde the target PER by the number of ARQ Change optmal parameter * λ n falure * * * 1 N Use the (8) to decde the ( γ,γ,...,γ ) pass check the constrant n (5) Usng () to decde the value of adaptve-power n each mode Sn( γ) S falure Usng [,(9)] to check the average PER conform the target PER or not pass Fgure. Flowchart. Have the maxmum average SE

71 S.-M. TSENG ET AL Numercal Results Here, we present numercal results for average spectral effcency, power adaptaton and PER usng the optmal soluton. We fnd the constrant n (5) when Nakagam channel m = 1 s a specal case that E( x ) n (4) has no close-form and can t gve the value of constrant n (5). So we use Nakagam channel fadng parameter wth m = nstead. We use the parameters of AMC scheme that s lsted n Table 1 from [5]. In Fgure 3, we can see power swtchng n each mode and the leftmost power n each AMC mode s largest that make the average PER to conform PER constrant n each AMC mode. Fgure shows the flowchart of optmal search algorthm. We use a Nakagam fadng wth parameter m = MIMO channel model, and assume a target PLR of P loss = In Fgure 4, depct the average spectral effcency for TM AMC scheme n Table 1, and we observed that usng the truncated ARQ protocol helps ncrease the system SE ( N ARQ = vs. N ARQ = 0). Also n Fgure 4, we can observe that our proposed scheme SE of N r = 0 has a sgnfcant SE mprovement over the one of N ARQ = n [] Normalzed Power adaptve-power constant-power average SNR(dB) Fgure 3. Adaptve power n each mode for Table 1 AMC scheme for sub-channel, target PER = average SE adaptve power ARQ= adaptve power ARQ=1 adaptve power ARQ=0 constant power ARQ= average SNR Fgure 4. Table 1 AMC scheme MIMO channel, target PER = 0.01.

72 49 S.-M. TSENG ET AL. But the maxmum N ARQ not always brng the best average SE, 11 SNR 1 n Fgure 4, we can see the average SE n MIMO channel at N ARQ = s smaller than N ARQ = 1. Because ncreasng N ARQ, we allevate the error bound to mprove the average SE, but retransmsson the same packet too many tmes wll also degrade the average SE, so we try to balance N r wth target PER, such as N ARQ 1 n Pt 0.01, and have the maxmum average SE n our proposed system. In Fgure 5, we show the dfferent combnatons of two sub-channel for dfferent SNRs n bar dagrams to the ARQ. We consder these 8 combnatons and map them to nteger numbers on x-axs [3]: {1=(c 1 =0,c =0),=(c 1 =1,c =1),3=(c 1 =,c =1) 4=(c 1 =1,c =),5=(c 1 =,c =),6=(c 1 =3,c =) 7=(c 1 =,c =3),8=(c 1 =3,c =3)} where c s the sub-channel. In Fgure 5, we can see the case whch have better SE than the other cases. Ths s consstent wth the conclusons of the Fgure 4. Fgure 6 compares the PER n [] and the PER of our proposed scheme. We can see we ncrease the PER to the maxmum (target PER = 0.01) at the left most of the SNR axs. 4 SE 1 SE average SNR= average SNR=10 SE 5 SE average SNR=15 average SNR=0 10 SE 5 SE average SNR=5 average SNR=30 Fgure 5. Dfferent combnatons of two sub-channel for dfferent average SNRs Average PER adaptve power constant-power Average SNR/dB Fgure 6. PER for Table 1 AMC scheme wth adaptve power.

73 S.-M. TSENG ET AL Conclusons In ths paper, we consdered optmzed rate and power adaptaton at the physcal layer amng at maxmzng the average SE, whle satsfyng a target PER constrant at the data lnk layer n MIMO channel. We proposed that ncreasng the PER makes t approach PER constrant, and the optmal mode swtchng level of each rate wll shft to the left n the SNR axs, so we can use the hgher order modulaton to mprove the average SE. The numercal results shows that the adaptaton power wthn each SNR mode can mprove the average spectrum effcency by 0.4 ~ 0.7 (bts/symbol) n md-snr range for MIMO systems, and ARQ N ARQ 1 s optmal for target PER = 0.01 to have the maxmum average SE. We also show that each transmt antenna can have dfferent maxmum retransmsson number due to ndependence of the sub-channels. 6. Acknowledgements Ths work was presented n part n Internatonal Conference on Advanced Informaton Technologes (AIT010), Tachung County, Tawan, 3-4 Aprl 010. Ths work was supported n part by Natonal Scence Councl, Tawan, under Grant NSC 98-1-E MY. 7. References [1] X. Zhang, J. Tang, H. H. Chen, S. C and M. Guzan, Cross-Layer-Based Modelng for Qualty of Servce Guarantees n Moble Wreless Networks, IEEE Communcatons Magazne, Vol. 44, No. 1, January 006, pp [] Q. Lu, S. Zhou and G. B. Gannaks, Cross-Layer Combnng of Adaptve Modulaton and Codng wth Truncated ARQ over Wreless Lnks, IEEE Transactons on Wreless Communcatons, Vol. 3, No. 5, September 004, pp [3] J. S. Harsn and F. Lahout, Optmzed Lnk Adaptaton for Wreless Packet Communcatons Based on Dscrete-Rate Modulaton and Codng Schemes, IEEE Workshop Sgnal Processng Advances n Wreless Communcatons, Helsnk, June 007, pp [4] A. Jafar and A. Mohammad, A Cross Layer Approach Based on Adaptve Modulaton and Truncated ARQ for MIMO Systems, Processng IEEE Internatonal Conference on Telecommuncatons and Malaysa Internatonal Conference on Communcatons, Penang, May 007, pp [5] A. Doufex, S. Armour, M. Butler, A. Nx, D. Bull, J. McGeehan and P. Karlsson, A Comparson of The HI- PERLAN/ & IEEE 80.11a Wreless LAN Standards, IEEE Communcatons Magazne, Vol. 40, No. 5, May 00, pp

74