Uplink beamforming for the FDD mode of UTRA

UPLINK BEAMFORMING FOR THE FDD MODE OF UTRA Xa ie Mes e, Ju ie R. Fonollosa, G ego i Vazquez Depa men o Signal Theo y and Communica ions Uni e si a Poli ecnica de Ca alunya c/Jo di Gi ona, 1-3,08034 Ba celona, SPAIN Phone: +34 93 4017052 Fax: +34 93 4016447 e-mail: [ mes e, ono, g ego i]@gps. sc.upc.es ABSTRACT This pape p esen s some link le el simula ion esul s o he e alua ion o aclap i e an ennas in he uplink o he FDD mode o UTRA. Two amilies o algo i hms we e ini ially conside ed, he basic di e ence be ween hem being hei abili y/disabili y o supp ess he con ibu ion om WCDMA di ec ional in e e ing sou ces. Two dis inc schemes we e es ablished as ep esen a i es o each amily and hei pe o mance was e alua ed in p esence o some illus a i e in e e ing scena ios. In he ligh o he esul s i is shown ha ime- e e ence beam o ming algo i hms su e om se e e beampa e n dis o ion e ec s when applied as such. This in u n causes ha sh pe o mance deg ada ion in e ms o aw BER, especially a high SINR le els. I is shown ha hese sho comings a e essen ially caused by he uplink mul iplexa ion o he a ic channel, which1 is seen by he base s a ion as a powe ul in e e ing souce coming om he di ec ion o a i al o he desi ed use . 1. INTRODUCTION The i s aim o he wo k p esen ed he ein was o e alua e he con enience o a ay-p ocessing in e e ence canceling schemes in ypical W-CDMA scena ios. Clea ly, he pe o mance o such a chi ec u es will depend s ongly on bo h spa ial dis ibu ion and empo al s uc u e o he ansmi ed signals. Since he analy ical modeling o hese e ec s becomes somewha bu densome, a i s app oxima ion o he p oblem by means o ex ensi e simula ions seemed mos app op ia e. The e alua ion p ocess was ca ied ou ollowing FRAMES ecommenda ions as desc ibed in [2]. Thus he “Ac ual Value In e ace” (AVI) was chosen as he mos sui able connec ion be ween link-le el and sys em-le el simula ions. The echnique es ablishes ha he link le el simula ion esul s should be measu ed in a bu s -by-bu s basis so ha he sys em simula o unde akes all coding and link le el adap a ion. Thanks o ha , all adio esou lce managemen algo i hms (ha ing an ac i a ion pe iod highe han bu s du a ion) can be accu a ely simula ed on he sys em le el pla o m. 2. (CHANNEL AND SIGNAL, MODEL We deno e x( ) (Pxl column ec o ) he ecei ed snapsho a a pa icula ime ins an . I a&) ep esen s he ansmi ed This wo k was pa ially suppo ed by he Eu opean Coinmission unde ACTS P ojec AC347 SIJNBEAM, he Spanish Go e nmen (TIC96-0500-C10-01, TIC98-0412, TlC98-0703) and he Ca alan Go e nmen CIRlT 1998SGR-0008 1. (analy ic) signal coming om he s- h mobile s a ion, i s con ibu ion o he ecei ed snapsho can be modeled as ollows: being g,, a complex p opaga ion gain and s,,( ) a ime- a ying supe posi ion o s ee ing ec o s. This gene alized s ee ing ec o s,,( ) -as i is commonly e e ed o in he li e a u e- inco po a es he angula in o ma ion o he channel and depends on ime due o he Dopple e ec . In he sequel we will assume ha 1) he ecei ed signal is p ope ly sampled a a mul iple o he chip a e (Nsc sa nples/chip) and 2) pe ec synch oniza ion wi h he use o in e es has been a ained al eady. Now, ga he ing M consecu i e snapsho s o he ecei ed signal a he base s a ion in o an AhP ma ix X, we ha e: x = [x(o) X(TS) ... x((M -l)Ts)lT wi h Ts deno ing he sampling pe iod. Le us now in oduce an MxL segmen o con olu ion ma ix: d((L-1)Nsc +1) d((L - 2)Nsc+l) .’. d(1) d((L -~)Nsc + 2) d((L - ~)Nsc + 2) d((L -1)Nsc + M) d((L - 2)Nsc + M) ... d(M) gene a ed om M+(L-1)Nsc samples o he desi ed use ’s aining sequence d( ) ansmi ed in he Con ol Channel (PDCCH) -o a epe i ion he eo whene e M+(L-1)Nsc is bigge han he pe iod o he aining sequence-. Upon he de ini ion o an LxP ma ix including he wo-dimensional disc e e (1 sample/chip) channel impulse esponse: (4) we can exp ess he spa ial- empo al ecei ed signal ma ix as: X=DH+W (5) whe e W (MxP) includes no only spa ial- empo al backg ound noise bu also he con ibu ion o signals ansmi ed om o he mobile s a ions as well as a ic channels embedded in he desi ed use ’s da a s eam. P o ided ha he leng h o he channel assumed a he ecei e (L) is no su icien ly la ge o cope wi h he whole delay sp ead o he adio channel, he la es e lec ions o he incoming signal a e assumed o be included in his noise e m. 0-7803-5565,-2/99/$10.00 0 1999 IEEE 867 3. UPLINK ALGORITHMS UNDER TEST The main pu pose o he conside ed simula ions was o in es iga e he in luence o di e en ep esen a i e W-CDMA in e e ing scena ios on he pe o mance o con en ional adap i e an enna sys ems. The objec i e was o quan i y he po en ial ad an ages o in e e ence canceling schemes in on o pu ely di e si y combining a chi ec u es. Thus, wo di e en amilies o algo i hms we e conside ed, each one eso ing o a dis inc p esump ion conce ning he spa ial na u e o in e e ence. Figu e 1. Vec o ial Rake Recei e The i s scheme unde conside a ion, he eina e e e ed o as V-Rake Recei e (Figu e l), was chosen as ep esen a i e o he amily o algo i hms which do no ake in o accoun he p esence o di ec ional in e e ing sou ces. The scheme can be basically shown o be he op imum de ec o om a maximum- likelihood poin o iew when only omnidi ec ional Gaussian noise is p esen in he scena io. Assuming ha he componen s o he ma ix W a e join ly Gaussian-dis ibu ed, he ecep o can be de i ed om he maximiza ion o he ollowing likelihood unc ion: ?,I = ec"(X-DH)R-,' ec(X-DH) (6) wi h R, he MPxMP co a iance ma ix o ec(W), and 0 and ec() deno ing he K onecke p oduc and he column-wise ma ix s acking ope a ion espec i ely. A e some manipula ion o (6) we ge : ?,I = ecH(X)R: ec(X) +2Reb]+ (7) + ecH (D)R ec(D) whe e we ha e de ined: y = ecX(DH)R: ec(X) (8) R = (H* 61~ )R;(H~ 61,) wi h Inn he MxM iden i y ma ix. This o mula ion allows sepa a ion o he combining pa o he ecei e om i s subsequen da a-de ec ion pa . P o ided ha noise and in e e ence can be conside ed spa ially and empo ally unco ela ed, y becomes equal o (up o a scala ac o ): y = ecH(DH) ec(X) = [(DHIXX] (9) I we conside he ollowing decomposi ion o he channel and inpu signal ma ices: H=[h,...h,] x =[XI ... x,] he scala y can be exp essed in he ollowing way: P y = zh DHxp p=i Each e m hXDHX can be in e p e ed as he ou pu o a Rake ecei e ma ched o each an enna channel impulse esponse. In conclusion, he op imum ecei e esul s in he scheme he ein ega ded as he V-Rake Recei e (see Figu e 1). Ye one issue emains open: since he channel impulse esponse H is no known, i mus be es ima ed om he incoming da a. An ML channel es ima e can be ob ained aking de i a i es o he cos unc ion (6) wi h espec o ec(H*) and o cing he co esponding g adien o be ze o: PP om whe e we ob ain: ec(H)= [(IL ODH)R;(I, @D)Y(IL @D")R;' ec(X) (13) Finally, o he V-Rake Recei e conside ed he e: H = (D~D)-'D~x (14) The app oxima ion o empo al unco ela ion o in e e ing componen s o he inpu signal seem easonable as long as we deal wi h digi ally modula ed signals sampled a he chip a e. On he con a y, hei spa ial whi eness can no be so easily jus i ied and will depend s ongly on he ac ual scena io unde conside a ion. A mo e sensible app oxima ion, judging by conclusions d awn om measu emen campaigns in [7], is o conside ha he s uc u e o he dis inc gene alized s ee ing ec o s sI in (4) does no a y in obse a ion in e als sho e han he delay sp ead o he mobile adio channel. This is equi alen o he app oxima ion: H = has; ec(H) = so @ ha whe e ho and ep esen he equi alen one-dimensional channel impulse esponse and gene alized s ee ing ec o espec i ely. Subs i u ion o (15) in o (8) leads o: y = (sa 63 ha)H (I, 6 DH)R;' ec(X) (16) o , a e some algeb a: whe e, by i ue o he spa ial- empo al sepa a eness app oxima ion, we ha e decomposed he non-desi ed signal co a iance ma ix in o i s spa ial and empo al componen s R, = R, @ R, . We obse e ha he op imum combining p io o he sequence de ec ion collapses in o wo sepa a e s ages: a spa ial s age, in which he incoming snapsho s a e p ep ocessed by he op imum beam o ming (in he sense ha i p o ides maximum Signal o In e e ence plus Noise Ra io SINR a i s 868 ou pu ); and a i empo al s age, which can be in e p e ed as a single Rake ecei e . In his con ex , he second algo i hm unde es - he so-called Ma ched Desi ed Impulse Response MDIR Recei e [ 8][3] (Figu e 2)- eso s o he sepa a eness app oxima ion in (15) in o de o p o ide he ML es ima ion o he ansmi ed da a, he beam o me weigh s and he equi alen channel he Assuming Gaussiani y o he non-desi ed in e e ences he log-likelihood unc ion o he ecei ed da a a e he spa ial p ocessing can be exp essed (up o a cons an ) as: {=(Xb-Dh)HRi (Xb-Dh) (18) whe e we ha e conside ed ha he channel impulse esponse a e he beam o ming p ocess h, he da a, and he beam o me i sel b a e all de e minis ic quan i ies. As al eady discussed, i seems easonable o assume he spa ial co a iance ma ix is close o he iden i y so ha he ML cos unc ion u ns ou o be a mean squa e e o measu e: { = llXb - Dhll* (19) In o de o a oid he i ial solu ion an addi ional cons ain mus be imposed. Fo ins ance, one may se he powe associa ed wi hi he aining sequence a he ou pu o he beam o me equal o cons an : bXHD(DHD)DHXb = 1 (20) The eby, using he Lag ange mul iplie s me hod he solu ion may be shown o be [3]: XHXb = am,,XHD(DHD)-'DHXb (21) h = (DXD)-'DHXb Thus, he beam o ming weigh ec o is ound as he eigen ec o co esponding o he minimum gene alized eigen alue o he ma ix pencil [x~x, X~D(D D lD /x]. Since his eigen alue is equal o he in e se o he ou pu SINR minus one, he solu ion is yielding op imali y in e ms o SINR be o e he Rake ecei e . 4. SIMULATION ASPECTS The main di icul y ha a ises when simula ing a W-CDMA en i onmen esides in aking in o accoun all possible bi a es, and consequen ly sp ead ac o s, ha migh be in ol ed in an ac ual scena io. In o de o o e come his p oblem wo di e en kinds o use s we e de ined, he di e ence be ween hem being basically he ype o se ice ha hey equi e: High Bi Ra e (LBR) and Low Bi Ra e (HBR) use s. The o me a e supposed o ha e a highe bandwid h demand and, consequen ly, an ac ual mobile ne wo k will be capable o handling a high numbe o LBK use s whe eas he numbe o mobiles equi ing a HBR se ice will be limi ed. Fu he mo e, he highe he bi a e, he mo e ansmi ed powe is needed in o de o p ese e a gi en quali y o se ice. All his sugges s ha LBR use s can be p ope ly modeled as omnidi ec ional Gaussian noise. In ou case, a cons an sp ead ac o (SF) was assigned o all HBR use s i espec i ely o whe he hey cons i u e desi ed o in e e ing sou ces. Conce ning he signal gene a ion in he simula ions, each senso equency-selec i e channel was modeled wi h a apped delay line o ime- a ying coe icien s as shown in (1). Pa icula ly, he models e e ed o as Ou doo o Indoo and Pedes ian and Vehicula in [6] we e conside ed o he gene a ion o he channel equency selec i i y, whe eas he angula app oach p esen ed in 171 was used o cha ac e ize he angula dispe sion o he p opaga ion p ocess. Thus each la - ading componen o he mobile channel was gene a ed using a ay model, whe e he numbe o impinging wa e o ms was se equal o a Poisson andom a iable (mean equal o 25 ays). In addi ion, a Laplacian Powe Angula Spec um was conside ed, along wi h a Gaussian dis ibu ion o he di e en di ec ions o a i al o each use . The Powe Angula Sp ead was ixed o 8 deg ees o all aps and scena ios. The mobile speed was se o 3km/h and 120km/h o he Pedes ian and Vehicula models espec i ely. Acco ding o he Ac ual Value In e ace, he simula ion du a ion was se equal o he minimum powe con ol ac ualiza ion pe iod, namely 0.625ms. n XI@) Beam o ming Design 1 ,, I Me ic I T aining Sequence D h Figu e 2. Ma ched Desi ed Impulse Response Recei e . All he a ay-p ocessing algo i hms we e simula ed on a linea equally spaced a ay o 8 an ennas, in which he in e elemen sepa a ion was se o hal wa eleng h a he ca ie equency (1950 MHz). All mobile angula loca ions we e a bi a ily se by a uni o m andom a iable wi hin [-60,601 deg ees, i.e. 120 deg ees sec o iza ion was assumed. Each use was supposed o gene a e a single Dedica ed Physical Da a Channel (DPDCH) oge he wi h i s associa ed Dedica ed Physical Con ol Channel DPCCH (see 191 o de ails o he modula ion o ma s). 5. SIMULATION RESULTS Two in e e ing scena ios we e simula ed o each powe delay p o ile model, one wi h a single domina ing HBR in e e ence and ano he wi h i e. Apa om he a ay beam o ming algo i hms p esen ed in sec ion 3, a single-senso Rake ecei e was conside ed o compa ison pu poses. Uplink esul s a e depic ed in e ms o aw (uncoded) BER in Figu e 3 and Figu e 4. Al hough simula ions ook in o accoun a high ange o LBR powe ( e lec ed in he Eb/No a io), only esul s wi h Eb/No=lOdB a e p esen ed he ein. No e ha he BER is always exp essed as a unc ion o he ins an aneous Eb/Io pe senso (i.e., ecei ed by a single an enna and 869 measu ed wi hin an ac ualiza ion pe iod o he as powe con ol). Thanks o ha , a con en ional planning ool can di ec ly use hese esul s and he e o e dis ega d he exis ence o adap i e an ennas. Pedes ian, 115 inle eie s. EbINo-10 dB 100 I ' ' 'Q Smole-senso Rake I ... ., .. ............. :I -" ............... ......... 0 5 10 15 20 25 1 o-~ -15 -10 -5 Eblia pe senso [dB1 Figu e 3. Uplink esul s o he Pedes ian Channel Model. Vehicula , 115 In e e e s, EblNo=lO dB P io 5 H 10 0 5 10 15 20 25 lo-+ -15 -10 -5 Ebllo pe senso [dB] Figu e 4. Uplink esul s o he Vehicula Channel Model. Compa ing he pe o mance o he p oposed algo i hms, he MDIR ecei e gene ally a ains he bes esul s in e ms o equi ed Eb/No and Eb/Io o a pa icula aw BER in he ange o in e es (usually om lo-' o 10.' o aw BER). This is a logical esul , since nei he he V-Rake algo i hm no i s single- senso coun e pa ake in o conside a ion he p esence o in e e ence, which has a mos de imen al e ec on he ecei e pe o mance. No e howe e ha o high Eb/No o Ebb0 le els he Mul isenso Rake ecei e ac ually ou pe o ms he MDIR scheme, he eason o ha being basically wo old. Fi s , he MDIR s a egy bases i s ope a ion upon he adap a ion o a na owband beam o me , an app oxima e solu ion o he impai men s in oduced by he ime-dispe si e channel. Second, non-ideal es ima es o he co ela ion ma ices a e bound o o igina e signi ican pe o mance losses, especially when he powe o he ecei ed signals is high enough. The loss in oduced by he app oxima ion by he use o a na owband beam o me ins ead o an a ay o FIR il e s (app oach equi alen o ha o a wideband beam o me ) becomes ully jus i ied by he on -end complexi y educ ion ha he o me en ails. O mo e se ious conce n is he deg ada ion in oduced by impe ec es ima ion o he co ela ion ma ices needed in he MDIR algo i hm. Basically, he eason o his beha io is ha he a ic channel mul iplexed wi h he pilo is seen by he base s a ion as an in e e ing signal coming om he e y same angula di ec ion (no e ha only he aining sequence componen is iden i ied as 'desi ed sou ce' by he beam o ming algo i hm). This leads o a deg ada ion o he desi ed signal es ima ed co a iance ma ix equi alen o a andom poin ing e o when ackling wi h con en ional sou ces. Besides, he highe he desi ed ecei ed powe (and so he measu ed Eb/(No+Io)) he mo e p ecise he es ima ion has o be in o de o p e en he desi ed signal om being cancelled ou by he beam o me [I]. Since accu acy is in his case limi ed by he in e e ence o he a ic channel, a high Eb/(No+Io) a ios he sys em can no a oid supp essing he desi ed signal ins ead o enhancing i . A ay Fac o , Eblla= WE, EbiNo= 3MB 0 -5 -10 -15 --20 D 3 -25 e - -30 -35 -40 -45 -50 -100 -80 -60 -40 -20 0 20 40 60 80 100 Azimulh (deg) Figu e 5. Deg ada ion o he a ay esponse due o he p esence o he a ic channel. Solid and dash-do ed line: spa ial esponse o he global beam o me . Do ed line: spa ial esponse o he eigen ec o co esponding o he maximum eigen alue o he desi ed signal co a iance ma ix. Desi ed signal DOA -20 deg. In e e ing signal DOA 40 deg. The consequences o his e ec on he a ay spa ial esponse can be no iced in Figu e 5, whe e he a ay ac o has been ep esen ed o a Pedes ian scena io in p esence o a high le el o desi ed signal powe . In o de o ob ain mo e insigh in o he beampa e n dis o ion e ec s, a low angula sp ead was chosen (0.01 deg ees). I may ini ially seem ha when he aining sequence is ansmi ed oge he wi h a a ic channel he scheme can s ill es ima e he desi ed signal spa ial signa u e (do ed line) wi h a ce ain deg ee o accu acy. Howe e he global beam o me spa ial esponses show ha small e o s in his es ima e esul in g ea dis o ion e ec s o he global beampa e n, wha in u n b ings abou high deg ada ion o he BER a he mobile s a ion. 870 Se e al possible solu ions o his dis o ion e ec a e cu en ly being unde in es iga ion. They a e based on semi-blind es ima ion echniques and basically seek o ake in o accoun no only he aining sequence bu also o he unknown in e e ing symbols embedded in he modula ed da a s eam [4]. The o mula ion o hese algo i hms is no epo ed he e o lack o space, al hough we do include p elimina y esul s o demons a e hei po en ial. In pa icula , Figu e 6 shows he ou pu SINR o he beam o me as a unc ion o he inpu powe le el (wi h espec o he backg ound noise) in an scena io wi h wo HBR use s ansmi ing wi h SF=8. As he powe o he HBR use s inc eases beyond 10 dB, p ope ou pu SINR alues can no longe be sus ained. Ne e heless, his p ocli i y is success ully o e come by means o he applica ion o a semi- blind echnique based on a Condi ioned ML c i e ion. The in e es ed eade is e e ed o [4] o u he de ails o he algo i hm as well as mo e ex ensi e pe o mance simula ion esul s. Ou pu SINR s Inpu SNR Semi-blind CML g a -10 -20 -20 _-------- - J’ -10 -5 0 5 10 15 20 25 30 Inpu SNR (dB) Figu e 6. Oiu pu SINR e sus Inpu SNR in a scena io wi h wo HBR use s ansmi ing wi h SF=8 and he same powe . Vehicula 115 in e e e s, Eb/No=PO dB 10” 10 ’ I I cc m p 102 g - In addi ion o lhe wo k p esen ed he e, link-le el simula ions o downlink a ay-p ocessing algo i hms o he FDD mode o UTRA ha e also been pe o med (see [5] o de ails). Figu e 7 p esen s pe o mance cu es o wo di e en beam o ming algo i hms: a poin ing scheme (DPA) and a null-s ee ing scheme (DBNS). The DPA app oach seeks o gene a e a maximum may esponse owa ds he di ec ion o a i al o he desi ed use whe eas he DBNS deploymen is pa icula ly in ended o maximize he SINR a he mobile s a ion inpu . Su p isingly, i can be obse ed ha he DPA algo i hm ou pe o ms he DBNS scheme o alues o he ins an aneous Ebb0 abo e 5-8 dB. Once again, his beha io is basically mo i a ed by a poo es ima ion o he desi ed signal spa ial signa u e. As al eady shown, his es ima ion is p ecise enough o p ope ly poin a he desi ed use , bu no so eliable o a oid he desi ed signal supp ession when a maximum SINR s a egy is adop ed. 6. CONCLUSIONS This pape has p esen ed some link-le el simula ion esul s o he uplink and downlink o he cu en UTRA-FDD s anda d de ini ion. Resul s show ha applica ion o classical ime- e e ence beam o ming echniques encoun e s se e e pe o mance p oblems when ope a ing a high inpu SNR le els. A beampa e n dis o ion e ec , p incipally induced by he mul iplexed a ic channel, has been shown o be he ul ima e esponsible o he pe o mance deg ada ion. Finally, semi-blind echniques a e pu o wa d as alid a chi ec u es o ake in o accoun he p esence o he au o-in e e ing channel. Pe o mance esul s o hese echniques will be u he add essed in [4]. 7. REFERENCES [I] Comp on R. T., “Poin ing Accu acy and Dynamic Range in a s ee ed bedm adap i e a ay”, IEEE T ans. on AES, ol. AES-16, 121 Hdmlllinen S., Slanina P., Ha man M., Lappe ellinen A., Holma H., Salonaho O., “A No el In e ace be ween Link and Sys em Le el Simula ions”, P oc. ACTS Mobile Telecomm. Summi , Aalbo g, Denma k, pp. 599-604, Oc obe 1997. 131 Lagunas M.A., Pi ez A. and Vidal J., “Join Beam o ming and Vi e bi Equaliza ion in Wi eless Communica ions”, P oc. 3 1” Asiloma Con e ence on Signals, Sys ems and Compu e s, Paci ic G a e, Cali o nia, No embe 1997. 141 Mes e X., Fonollosa J. R., An 6n C., “Adap i e Beam o ming o HBR se ices in he FDD mode o UTRA”, submi ed o he IEEE Vehicula Technology Con e ence, 1999-Fal1, Ams e dam, The Ne he lands. [SI Mes e X., Fonollosa J.R, “Pe o mance E alua ion o UTRA- FDD’, P oc. FRAMES Wo kshop, Del (The Ne he lands), Janua y 1999. [6] Nikula E., “Ai In e ace Speci ica ion, Laye 1, D a ”, In e media e Deli e able ACO~~/NOK/PTO/DS/R/~I~/~J~. FRAMES p ojec . [7] Pede sen K.I., Mogensen P.E., Fleu y B.H., “A S ochas ic Model o he Tempo al and Azimu hal Dispe sion seen a he Base S a ion in Ou doo P opaga ion En i onmen s”, submi ed o he IEEE T ans. on Vehicula Technology. [SI Pipon E.P., Che alie P., Villa P., Mono J.J, “Join Spa ial and Tempo al Equaliza ion o channels wi h IS1 and CCI- Theo e ical and Expe imen al esul s o a base s a ion ecep ion”, P oc. IEEE Signal P ocessing Wo kshop on Signal P ocessing Ad ances in 191 “Submission o P oposed Radio T ansmission Technologies: he ETSI UMTS Te es ial Radio Access (UTRA) ITU-R R’M Candida e Submission”, ETSI SMG2. Da e o submission: 29/1/1998. A ailable a he ITU WWW h p://www.i u.ch/im / 1976, pp. 280-287. Wi eless Communica ions, Pa is, F ance, Ap il 1997, pp.309-3 12. 87 1