Sheet Music


of 52
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
A TURBO DETECTION SCHEME FOR EGPRS A THESIS SUBMITTED TO THE GRADUATE SCHOOL OF NATURAL AND APPLIED SCIENCES OF THE MIDDLE EAST TECHNICAL UNIVERSITY BY ÜLKÜ GÜLMEZ BAŞKÖY IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE IN THE DEPARTMENT OF ELECTRICAL AND ELECTRONICS ENGINEERING SEPTEMBER 2003 Approval of the Graduate School of Natural and Appled Scence Prof. Dr. Canan ÖZGEN Drector I certfy that th the atfe all the requrement a a the for the degree of Mater of Scence. Prof. Dr. Mübeccel DEMİREKLER Head of Department Th to certfy that we have read th the and that n our opnon t fully adequate, n cope and qualty, a a the for the degree of Mater of Scence. Aoc. Prof. Dr. Buyurman BAYKAL Supervor Examnng Commttee Member Aoc. Prof. Dr. Mele D. YÜCEL Aoc. Prof. Dr. Tolga ÇİLOĞLU Aoc. Prof. Dr. Buyurman BAYKAL Aoc. Prof. Dr. Engn TUNCER M.Sc. Hacer SUNAY TÜBİTAK-BİLTEN ABSTRACT A TURBO DETECTION SCHEME FOR EGPRS Başöy Gülmez, Ülü M.Sc., Department of Electrcal and Electronc Engneerng Supervor : Aoc. Prof. Dr. Buyurman Bayal September 2003, 7 page Enhanced Data Rate for Global Evoluton EDGE one of the 3G wrele communcaton tandard, whch provde hgher data rate by adoptng 8- PSK modulaton n TDMA ytem nfratructure of GSM. In th the, a turbo detecton recever for Enhanced General Pacet Rado Servce EGPRS ytem, whch the pacet wtchng mode of EDGE, tuded. In turbo detecton, equalzaton and channel decodng are performed teratvely. Due to 8-ary alphabet of EGPRS modulaton, full tate trell baed equalzaton, a uually performed n GSM, too complex not only for turbo detecton but even for conventonal equalzaton; o uboptmum cheme have to be condered. The Delayed Decon Feedbac Sequence Etmaton DDFSE choen a uboptmal and le complex trell baed cheme and t examned a a conventonal equalzaton technque frtly. It hown that the DDFSE ha a fne tradeoff between performance and complexty and can be a promng canddate for EGPRS. Then t employed to reduce the number of the trell tate n turbo detecton. Maxlog-MAP algorthm ued for oft output calculaton of both SISO equalzer and SISO decoder. Smulaton reult llutrate that propoed turbo detecton tructure mprove bt error rate and bloc error rate performance of the recever wth repect to the conventonal equalzaton and decodng cheme. The teraton gan vare dependng on modulaton type and codng rate of Modulaton Codng Scheme MCS employed n EGPRS. Keyword: Equalzaton, Decodng, Turbo Detecton, Delayed Decon Feedbac Sequence Etmaton DDFSE, Enhanced Data Rate for Global Evoluton EDGE, Enhanced General Pacet Rado Servce EGPRS. v ÖZ GGPRS İÇİN BİR TURBO SEZİMİ YAPISI Başöy Gülmez, Ülü Yüe Lan, Eletr ve Eletron Mühendlğ Bölümü Tez Yönetc : Doç. Dr. Buyurman Bayal Eylül 2003, 7 ayfa Küreel evrm çn gelştrlmş ver hızı, yüe blg aışı ağlayan üçüncü enerayon ablouz letşm tandartlarından brdr. 8 PSK pleme tenğn GSM n zaman bölümlü çoğul erşm altyapıında ullanır. Bu tezde Küreel Evrm çn Gelştrlmş Ver Hızı nın paet anahtarlama modu olan Gelştrlmş Genel Paet Radyo Servler GGPRS çn turbo ezm le çalışan br alıcı üzernde çalışılmıştır. Turbo ezm yöntem anal şfrelemen çözme ve denleştrme şlemn döngülü olara yapar ve alıcı performanını yüeltr. Genellle GSM çn ullanılan tam durumlu afe tabanlı denleştrme yöntemn turbo ezm çn ullanma GGPRS n uyguladığı 8 PSK pleme düşünüldüğünde olduça yüe br heaplama armaşılığı yaratır. K bu yapının alışılagelmş denleştrme ve şfre v çözümleme yöntem çn ble uygulanmaı olduça zordur. Bu nedenle yarı uygun yapılar GGPRS çn düşünülmeldr. Gectrlmş Karar Gerbelemel Dz Ketrme GKGDK yarı uygun ve daha az heaplama armaşılığı unan afe tabanlı yapı olara eçlmştr. GKGDK çn performan ve heaplama armaşılığı araında uygulanablr br lş olduğu göterlmş ve GKGDK nn GGPRS denleştrme çn uygun br aday olableceğ görülmüştür. Turbo ezmnde afe durumlarının azaltılmaı çn GKGDK algortmaı, yumuşa grd/çıtı ezm ve şfre çözme şlemler çn de Max-Log-MAP algortmaı uygulanmıştır. Benzetm onuçları götermştr uygulanan turbo ezm yapıı GGPRS alıcıının bt hata oranını azaltmıştır. Döngülü azanç, GGPRS de ullanılan pleme çeşdne ve şfreleme oranına göre değşmetedr. Anahtar Kelmeler: Denleştrme, Şfre Çözümleme, Turbo Sezm, Gectrlmş Karar Gerbelemel Dz Ketrme GKGDK, Küreel Evrm çn Gelştrlmş Ver Hızı, Gelştrlmş Genel Paet Radyo Servler GGPRS. v ACKNOWLEDGMENTS I would le to expre my grattude to Aoc. Prof. Dr. Buyurman Bayal for h upervon. I would le to than to my huband for h atance, unwaverng upport and patence. Alo, pecal than to my famly; my father İrfan Ze Gülmez, my mother Sarye Gülmez, my ter Hande Gülmez and my brother Oğuzhan Gülmez for ther great love and upport through my whole educaton. v TABLE OF CONTENTS ABSTRACT... ÖZ... v ACKNOWLEDGEMENTS...v TABLE OF CONTENTS...v CHAPTER. INTRODUCTION EQUALIZATION TECHNIQUES Overvew of equalzaton technque Maxmum Lelhood Sequence Etmaton Delayed Decon-Feedbac Sequence Etmaton DDFSE Computatonal Complexty Comparon of MLSE and DDFSE Algorthm TURBO EQUALIZATION Turbo Prncple SISO Decodng/Equalzaton MAP Decodng Algorthm SISO Decoder SISO Equalzer v 4. EQUALIZATION AND TURBO DETECTION FOR EGPRS Modulaton Codng Scheme of EGRRS Modulaton Format of EGPRS Gauan mnmum-hft eyng ary Phae Shft Keyng Moble Rado Channel Recever Structure Channel Etmaton Delayed Decon Feedbac Sequence Etmaton for EGPRS Turbo Detecton for EGPRS CONCLUSION REFERENCES x CHAPTER INTRODUCTION In the lat decade, moble communcaton ha become one of the fatet growng maret epecally wth the ntroducton of the ophtcated dgtal cellular ytem, econd generaton moble rado ytem. The global ytem for moble communcaton GSM one of the econd-generaton tandard, whch ha been a maor ucce tory for the global telecommuncaton ndutry, delverng telephony and low bt-rate data ervce to moble end-uer. A of Aprl 2002, for example, more than 80 countre acceed GSM to provde ervce to more than 680 mllon cutomer []. Today, everal GSM-networ have almot reached ther lmt,.e. the allocated pectrum not uffcent to upport the growng demand for moble communcaton. Addtonally, the demand for more tranmon capacty accelerated becaue of the growth n requred hgh data hgh rate for new ervce n wrele communcaton, e.g. multmeda applcaton and the Internet. The need for hgher data rate fulflled by pecfyng tandard wthn GSM phae 2, called a General Pacet Swtched Servce GPRS and Hgh-Speed Crcut Swtched Data HSCSD a a frt evoluton tep n GSM [2][3]. Thee tandard ncreae the pectral effcency of GSM by employng mult-lot operaton, hence ther capacty of provdng the hgher tranmon data rate lmted and more ophtcated approache are oon requred to meet the demand for fater acce for nformaton. Snce the early 990, ndutry ha been actvely reearchng thrd-generaton moble rado acce ytem and the tandardzaton of the frt releae of thrd generaton cellular ytem wa fnalzed n 999 by ETSI/3GPP. Global ytem for moble communcaton/enhanced data rate for global evoluton GSM/EDGE one of the thrd generaton cellular ytem pecfed by th tandardzaton framewor. EDGE actually an evoluton of the GSM and other TDMA baed 2G moble rado tandard,.e., IS-36, toward gnfcantly hgher pectral effcency wth a novel common phycal layer and ome upper layer feature. Rapd avalablty, the reue of extng GSM and TDMA nfratructure and hgh pectral effcency mae GSM/EDGE potentally attractve a a 3G mgraton path. EDGE mprove the pectral effcency by applyng the modulaton format 8-ary phae-hft eyng 8 PSK, n conuncton wth gauan mnmum hft eyng GMSK already ued n GSM. The modfcaton of the modulaton cheme of gnfcant for equalzaton that fght agant trong Although EDGE a generc ar nterface, t predomnantly ued n conuncton wth GSM a Enhanced Data Rate for GSM evoluton EDGE 2 channel nter-ymbol nterference ISI caued by lmted bandwdth and mult-path propagaton. Optmum equalzaton for mnmzng bt error rate,.e., maxmum lelhood equence etmaton MLSE baed on Vterb algorthm, whch uually utlzed for GSM, requre an extenve computatonal complexty due to ncreae n modulaton ze. Smpler cheme le Decon Feedbac Etmaton DFE tend to perform poorly for 8 PSK nce the ymbol contellaton farly dene caung the algorthm to become overly entve to noe []. In th cenaro, the Delayed Decon Feedbac Sequence Etmaton DDFSE eem to be an attractve canddate nce t generate a tradeoff between complexty and performance. Addtonally, t well uted for mplementaton becaue of t hgh regularty. Sgnfcant mprovement n BER performance are poble wth coded data tranmon ung channel codng. The bac dea of channel codng to ntroduce controlled redundancy nto the tranmtted gnal that exploted at the recever to correct channel nduced error by mean of forward error correcton. Conventonal recever generally nvolve equalzaton and channel decodng eparately. In 995, a new approach called a turbo detecton ha been propoed to effcently combat trong ISI [2]. The bac dea cont n conderng the channel a a non-recurve, non-ytematc convolutonal code of rate one and realzng a ont teratve channel equalzaton and decodng technque, whereby the channel decoder utlzed n order to mprove the performance of the equalzaton proce and vce vera n an teratve regme. The cope of th the to mplement a turbo detecton cheme to examne the performance of the turbo detecton for modulaton codng cheme of 3 EGPRS; the pacet wtched mode of EDGE, wth a reduced complexty equalzer. For th purpoe, the DDFSE employed a the equalzaton technque for 8-PSK modulaton and the tradeoff between the performance and complexty n moble rado channel condton are examned. Then, the performance of turbo detecton for EGPRS, tuded wth the DDFSE approach ued n equalzaton part of turbo detecton. The mplementaton performed wth C programmng language n the development envronment of Mcrooft Vual Studo 6.0. In Chapter 2, an overvew of equalaton technque for TDMA ytem gven and the Maxmum Lelhood Sequence Etmaton MLSE wth the Delayed Decon Feedbac Sequence Etmaton DDFSE equalzaton algorthm decrbed n detal. In order to obtan a meanngful comparon for the computatonal complexty of the MLSE and the DDFSE algorthm, an analy for the computatonal complexty alo produced n th ecton. In Chapter 3, the prncple of turbo detecton frt examned. Then, the MAP algorthm and t varant, log-map and Max-log-MAP are decrbed. The oft n/ oft out SISO equalzer and channel decoder component, whch utlze the MAP algorthm, are tuded fnally. The applcaton of the DDFSE and turbo detecton to EGPRS tuded n Chapter 4. After a hort ntroducton to EGPRS, modulaton and codng cheme a well a moble rado channel are treated. Then channel mpule repone etmaton and preflterng for DDFSE are dcued. DDFSE equalzaton performance nvetgated for GSM channel profle. Fnally, turbo detecton performance tuded by applyng turbo prncple to EGPRS Modulaton Codng Scheme MCS. 4 In Chapter 5, ome concluon are drawn for DDFSE and turbo detecton performance of the propoed recever tructure and ome poble future wor area are ponted out. 5 CHAPTER 2 EQUALIZATION TECHNIQUES Tranmon through a band lmted channel an example of gnfcant nteret practcally where the tranmtted gnal dtorted n tme by the channel. The dperon ntroduced by uch a channel can caue nterference between the tranmtted adacent ymbol, whch called a Inter Symbol nterference ISI. In th cae, the ntantaneou gnal n the recever vualzed a uperpoton of everal ymbol n the pat and n the future. In order to facltate the relable detecton of gnal corrupted by ISI, ome form of ISI compenaton,.e. equalzaton, requred n the recever. Equalaton technque for dgtal modulaton have receved extenve attenton n the lterature for many year. It beyond the cope of th the to addre all thee development n detal, rather an overvew wll be provded of the mot mportant equalzaton technque. And the decrpton of the MLSE and DDFSE algorthm are expreed n ubecton. 2. Overvew of equalzaton technque Two broad categore of equalzer have been documented extenvely n the lterature; ymbol-by-ymbol equalzer and equence etmator. Symbol-by- 6 ymbol equalzer try to compenate ISI drectly and to mae ymbol by ymbol decon on the receved equence, whle equence etmator attempt to fnd the mot probably tranmtted equence regardng receved equence a a whole [8]. Sequence etmator are generally more complex than ymbol-by-ymbol equalzer, but can potentally offer better performance. Lnear equalzer and decon feedbac equalzer are two broad categore of ymbol by ymbol equalzer. The frt one, lnear equalzer LE, the clac tartng pont for equalzaton. It an eentally mple and ntutve oluton to the problem of ISI elmnaton. Conceptually one apple a flter wth the frequency repone, whch the nvere of the channel, thereby creatng a ytem wth an deal effectve frequency repone. The flter coeffcent of the LE can be determned by zero forcng crtera, leadng to Zero Forcng Lnear Equalzaton ZF-LE or by Mean Square Error crtera, leadng to Mean Square Error Lnear Equalzaton MSE-LE. In ZF-LE, the equalzer coeffcent are choen o a to force the combned channel and equalzer mpule repone to zero at all but one of amplng ntant n the pan of the equalzer. In the MSE-LE, however, the coeffcent are choen to mnmze the mean-quare error at the output of the equalzer, o t tae nto account ISI and noe together. Snce the bac operaton of the ZF equalzer to provde gan at frequence where the channel tranfer functon experence attenuaton and vce vera, both gnal and noe are enhanced multaneouly. Th reult n noe enhancement, whch degrade the performance of equalzer gnfcantly. Even more, f channel ha zero on the unt crcle correpondng to nfnte attenuaton n the channel frequency repone, the mean quare error at the output of the equalzer become nfnte. MSE-LE ontly mnmze the noe and the ISI, hence yeld a 7 lower total mn quare error [9]. Addtonally, the MSE-LE ext for the channel that exhbt pectral null nce tranfer functon of the equalzer doe not have correpondng pole. However, at hgher SNR level, when the noe contrbuton low, the MSE equaton of the MSE lnear equalzer approache that of the ZF lnear equalzer, caung both the MSE and ZF equalzer to exhbt mlar charactertc. The problem of the noe enhancement n LE greatly reduced by the Decon Feedbac Equalzaton DFE, the econd category of the ymbol by ymbol equalzer. The DFE employ two flter, a feedforward flter and feedbac flter, n order to elmnate the ISI. The non-lnear functon due to the decon devce ntroduced at the nput of the feedbac flter. The feedforward flter fed only wth preent and future receved gnal ample and elmnate only the pre-curor ISI. The feedbac wor wth the etmate of the receved ample from the decon devce and mtgate the ISI caued by the pat data ymbol,.e. the pot-curor ISI. Snce the feedforward flter elmnate ISI partally and the feedbac flter wor only wth etmated ymbol, noe enhancement n the DFE le gnfcant, when compared to the lnear equalzer. The flter tap n the DFE are aduted mot commonly to mnmze the MSE at output. The DFE le the LE, can tae the form of a zero forcng DFE. However, th may lead to noe enhancement o a better alternatve and mot common the MSE-DFE. Although the problem of the noe enhancement greatly reduced by the DFE, a erou degradaton n the equalzer performance, nown a error propagaton phenomenon, appear. Whenever an erroneou decon fed bac nto the feedbac flter, the flter produce an output etmate, whch erroneou; th lead to another erroneou decon beng fed bac. In th way, error propagate and caue more error n the feedbac loop. 8 The equence etmator conttute the econd cla of the equalzer and contan maxmum lelhood equence etmator and t uboptmum varant. Sequence etmator doe not attempt to remove ISI term from receved gnal. Intead, t etmate the tranmtted equence by earchng the mot lely tranmtted ymbol on the equence of receved gnal ample. Maxmum lelhood equence etmaton MLSE an optmum algorthm for mnmzng the probablty of equence error whle ymbol equalzer are uboptmum one wth repect to th crteron. Although t provde mproved performance a compared to LE or DFE, t hgh computatonal complexty t man drawbac. The complexty of MLSE grow exponentally wth the ze of gnal contellaton, M, and length of channel memory, L a M L. It can be a utable equalzaton technque for the ytem wth low modulaton ze,.e. GSM wth modulaton ze of 2, even f tranmtted gnal dtorted wth evere ISI. However, for ytem wth hgh value of gnal alphabet, MLSE mpractcal to mplement wth long channel memory. For example, f 8-PSK ued a modulaton wth channel mpule repone of length 7 the algorthm complexty become order of 8 6 that The delayed decon feedbac equence etmaton DDFSE [0][8] a reduced-tate equence etmaton algorthm, whch decreae the computatonal complexty of the MLSE. It a uboptmal algorthm and provde a tradeoff between performance and complexty by truncatng the effectve channel memory to µ term, where µ an nteger that can be vared from 0 to the channel memory. By dong o, the complexty of the algorthm can be controlled wth controlled degradaton n performance wth repect to MLSE. In th the, the DDFSE algorthm choen to perform the 9 equalzaton ta for EGPRS and more detaled nformaton gven n ecton Maxmum Lelhood Sequence Etmaton The Maxmum Lelhood Sequence Etmaton MLSE performed by computng the lelhood for every poble tranmtted equence and by electng the equence wth the larget lelhood. The lelhood are obtaned a the product of the probablte of recevng each gnal ample, gven that partcular tranmtted equence. Although the method conceptually mple, the drect oluton of th maxmzaton problem to tet all poble nput equence. Intead of th drect approach, Forney [24] howed that the Vterb algorthm provde the method for MLSE. The prncple of the algorthm to chooe the path through the ISI trell that ha the mnmum accumulated dtance between the receved data equence corrupted by whte-gauan noe and the hypothetcal data equence. The cheme ummarzed n followng lne [8]. MLSE algorthm earche, xˆ the etmate of tranmtted vector x, that maxmze the lelhood functon, ˆ N 0 N x0 x = max[ p r x = p r, Κ, r x, Κ, ] 2- where, N the length of the tranmtted equence and r receved ymbol at tme, whch defned a follow 0 r = L n= 0 x nhn z 2-2 h the channel mpule repone of the channel between tranmtted ymbol and receved one, L channel memory, z are addtve noe ample. Snce the addtve noe whte gauan, t ample are ndependent. Hence, the lelhood functon n equaton 2. correpond to, N = 0 p r x = p r x 2-3 N, Κ, x0 The log-lelhood functon, therefore, N = 0 ln[ p r x] = ln[ p r xn, Κ, x0 ] 2-4 Snce r depend only on the L mot recent tranmtted ymbol, the loglelhood functon can be rewrtten a ln[ p r x] = ln[ p rn xn, Κ, xn L ] ln[ p rn 2, Κ, r0 xn 2, Κ, x0 ] 2-5 If the econd term on the rght hand de of equaton ha been calculated prevouly at epoch N-2, then only the frt term, called the branch metrc, ha to be computed for ncomng gnal ample, r N- at epoch N-. Wth the well-nown equalty 2.6 for p r x, Κ, x wth r n equaton 2.5, L branch metrc at tme, µ,yeld equaton 2.7 [7] p r x L 2, Κ, x L = exp r 2 2 h x 2πσ = 0 2σ 2-6 L 0 2 µ = r h x 2-7 where σ 2 varance of addtve whte gauan noe. Baed on the recuron n equaton 2.5 and the branch metrc n equaton 2.7, the well-nown Vterb algorthm can be ued to determne the mot lely tranmtted equence. 2.3 Delayed Decon-Feedbac Sequence Etmaton DDFSE Unfortunately, the complexty of the MLSE recever grow exponentally wth the channel memory length and ymbol alphabet. When the channel memory length become large, the MLSE recever mpractcal. One oluton to reduce the recever complexty by truncatng the effectve channel memory to µ term, where µ an nteger that can be vared from 0 to L. Thu, a uboptmum decoder obtaned wth complexty controlled by parameter µ. Th the bac dea for Delayed Decon Sequence Etmaton DDFSE [0][8]. The DDFSE recever can be vewed a combnaton of th
Similar documents
View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks