MIMO Schemes In UTRA LTE, A Comparison

Transcription

1 MIMO Schemes In UA LE A Comparson Chrstoph Spege 1 Jens Berkmann Zjan Ba 1 obas Schoand 3 Chrstan Drewes Gudo. Bruck 1 Bertram Gunzemann Peter Jung 1 1 Unverstät Dusburg-Essen Lehrstuh für Kommunkatonsechnk Dusburg Germany Infneon echnooges AG Am Campeon Neubberg Germany 3 Infneon echnooges AG Düssedorfer Landstraße Dusburg Germany Abstract he ong term evouton of UMS (Unversa Mobe eecommuncatons System errestra ado Access abbrevated as UA LE w be based on OFDM (orthogona frequency dvson mutpexng. Furthermore MIMO (mutpe-nput mutpe-output technques have been consdered as a means for the mprovement of wreess connectvty. In wreess systems hgh data rates n the downnk are desrabe: Furthermore wth respect to an effcent mpementaton the downnk requres a thorough assessment. In partcuar the Aamout and the V-BLAS (Vertca Be Labs Layered Space me schemes are seen as nterestng concepts. In ths communcaton the authors w compare these two MIMO schemes w.r.t. the achevabe performance n the UA LE downnk usng up to two transmt and two receve antennas. Furthermore the authors w dscuss the depoyment of spata mutpexng n the UA LE downnk and w show that the performance of successve nterference canceaton (SIC based data detecton technques for MIMO-OFDM s benefca. Index terms Aamout Dversty MIMO OFDM (Orthogona Frequency Dvson Mutpexng Spata Mutpexng Successve Interference Canceaton (SIC UMS UA LE V-BLAS I. INODUCION Mobe rado systems ke UA (Unversa Mobe eecommuncaton System errestra ado Access [1][] drect sequence code dvson mutpe access (DS-CDMA has been used as the basc mutpe access method formng the basc PY and MAC concepts. Despte many advantageous features DS-CDMA technques do not factate a fexbe adaptve engneerng of the used frequency resource. Owng to ts nherent fexbty and ts attractve mpementaton potenta OFDM (orthogona frequency dvson mutpexng has therefore become a preferred canddate for the downnk of the ong term evouton of UA termed UA LE. o aow hgh user data rates whch are requred for the reazaton of wreess mutmeda the combnaton of OFDM wth MIMO (mutpe-nput mutpeoutput antenna schemes has been consdered as a preferred way forward. hs combnaton s termed MIMO-OFDM. he data detecton n MIMO- OFDM can be done n varous ways. In order to mt the addtona mpementaton compexty near equazaton e.g. zero-forcng (ZF and mnmum mean squared error (MMSE equazaton or sera nterference canceaton (SIC s preferred. Besdes transmt dversty.e. the Aamout scheme [3] V-BLAS (Vertca Be Labs Layered Space me [4] a SIC verson of ZF equazaton and the correspondng SIC-MMSE recever are consdered an nterestng concept for ths purpose. In what foows compex base band notaton w be used depoyng matrx-vector cacuus. Dscretetme varabes w be denoted by vectors whch are gven as ower case characters n bod face tacs. Matrces w be denoted by upper case characters n bod face tacs. Compex vaues w be underned. Furthermore ( denotes vector or matrx transposton ( denotes ermtan of a vector or a matrx. hs communcaton s organzed as foows: After ths ntroductory secton Sect. II w present the system mode underyng the anayss. In Sect. III smuaton resuts w be presented. Sect. IV w concude the manuscrpt. II. SYSEM MODEL A. Genera emarks It has been known that MIMO schemes can factate an mproved recever robustness by expotng transmtter (X and recever (X antenna dversty as we as the maxmzaton of data rates by depoyng spata mutpexng (SM. For ths reason K K X antennas are depoyed for communcaton wth at east K K X antennas. Between each par of X/X antennas there s a rado path. he X antennas receve the sgnas of the X antennas. o aow the data detecton n the case of SM K K must be fufed [4]. In SM the overa achevabe data rate at the X s K tmes the data rate of a snge X antenna transmtter. Aamout type X antenna dversty does not mpose any requrements on the number of X antennas /08/$ IEEE 8

2 herefore K > 1 aways yeds addtona X antenna dversty. In what foows a system concept for the SM and a system mode for the Aamout approach w be presented. Each data symbo s assumed to be taken from an M -ary symbo aphabet. B. Spata Mutpexng In ths secton we w depoy the mathematca mode ntroduced n [5]. We w consder a snge sot.e. a snge OFDM symbo perod n a tme transmsson nterva (I. Let us assume that K ( 1 ( k ( K data symbos d d d are transmtted on each subcarrer = 1 L. In what foows we w arrange these K data symbos n the subcarrer specfc data vector ( 1 ( k ( K ( d = d d d = 1 L. (1 Assumng cycc prefxes beng removed at the X nput the frequency doman receved sampe assocated wth subcarrer = 1 L s gven by K ( k ( k k ( k ( k r = d + n = 1 L. ( k = 1 k ( In ( n s the sampe of the nose measured at X antenna k k { 1 K} and assocated wth ( k k the th subcarrer = 1 L. denotes the channe coeffcent assocated wth the th subcarrer and thus wth the subcarrer specfc data vector d of (1. ( k k descrbes the connecton between X antenna k k { 1 K} wth X antenna k k { 1 K}. Owng to ( the subcarrers do not mutuay nterfere.e. ntercarrer nterference (ICI s absent. Wth the subcarrer specfc channe matrx ( 11 ( 1 ( 1 K ( 1 ( ( K = = 1L (3 ( K1 ( K ( K K and wth the subcarrer specfc nose vector ( 1 ( k ( K n = n n n = 1 L (4 ( the receved vector ( 1 ( k ( K ( r = r r r = 1 L (5 assocated wth the th subcarrer s gven by r = d + n = 1 L. (6 he V-BLAS scheme [4] s a vabe detecton approach for SM. he X sets out from r defned ( 1 n (5 and (6 to detect frst data symbo d. hs s done by nung ke n the case of the nusteerng beamformer. he mentoned nung can be performed by appyng a weght vector ( 1 ( k ( K = ( w w w w (7 wth the weghts beng determned accordng to ( 1 k ( k ( K k 0 k ( w = (8 1 k = to the receved vector r. (8 means that the weght vector w s orthogona to the subspace spanned by those contrbutons to r whch have not yet been canceed. (8 can be soved by frst deetng the frst ( k 1 coumns of defned by (3 whch yeds the modfed matrx and then computng the Moore-Penrose pseudo nverse + 1 = (9 ( hen w s gven by the th row of. In the case of the frst data symbo the + s the Moore- Penrose pseudo nverse of the fu matrx of (3 and w 1 s the frst row of +. Wth Q{} a denotng the compex quantzng operaton quantzng a to the nearest aowed data symbo reazaton we fnd ( 1 d = Q{ w1 r } = 1 L. (10 ( 1 Now the nfuence of the detected symbo d s subtracted from r resutng n the modfed vector ( 1 ( 11 ( 1 ( K 1 r = r d = 1 L (11 ( whch contans ess nterference. Snce the nfuence of the frst data symbo has been removed the modfed receved vector r = 1 L contans ess nterference than r. ence the detecton of the remanng symbos can be mproved. hs procedure s then repeated wth the new vector r detectng d (. hs procedure s repeated unt a data symbos have been detected. Obvousy the V-BLAS recever s a ZF (Zero- Forcng based SIC (Successve Interference Canceaton recever aso termed SIC-ZF n what foows. In the above descrpton t s assumed that the symbos are detected n the order of the data vector d. owever as stated n [4] the order of detecton s crtca. Because the sgna-to-nose rato (SN n each data detecton stage vares each data symbo s prone to a dfferent error probabty. ence the orderng of the data detecton effects the + 9

3 overa symbo error. It turns out that an optma strategy s a greedy approach. One hence chooses the partcuar data symbo wth the greatest SN of the remanng undetected data symbos n each detecton stage. hs data symbo s then detected wth the mnmum error probabty of a remanng data symbos. Subtractng the nfuence of the newy detected data symbo from the remanng data symbos yeds an mprovement of ther respectve SN eadng to a favorabe error performance. he SN can be estmated usng the Moore- Penrose pseudo nverse of the modfed channe matrx see (9 aso used to determne the approprate weghts for symbo nung. he sought cardna number of the next data symbos s smpy the row number of ths matrx wth the smaest L vector norm [4]. In V-BLAS recevers the aforementoned separaton s done straght forwardy by matrx nverson of the channe matrx provded that ts nverse exsts. owever ths ZF approach yeds poor resuts f the channe matrx of (3 s condtoned. hs happens when the cross correatons between the dfferent eements of are hgh e.g. when there s a sgnfcant ne-ofsght (LOS component. As an aternatve to the ZF based V-BLAS an MMSE (Mnmum Mean Squared Error based approach factates more favorabe resuts. In the case of MMSE recevers the orthogonaty prncpe s fufed whch guarantees that the estmaton error and the estmates are orthogona n average [6]. he V-BLAS agorthm can be modfed accordngy by repacng (9 wth N 0 = ( ( I + ( k E{ } d (1 where N 0 equa to k B 0 s the spectra nose power ( k densty and E{ d } s the symbo energy. he resutng recever s termed SIC-MMSE n what foows. C. Aamout ype X Dversty If the rado paths between the K K X antennas and the K K X antennas are suffcenty uncorreated the expotaton of dversty s possbe whch makes the recepton more robust aganst the detrmenta effects of nterference nose and fadng. herefore t w be possbe to ncrease the data rate by depoyng hgher order moduaton (OM thus provdng a hgher throughput. A smpe way of reazng such a system s by usng the Space-me Bock Code (SBC dscovered by Aamout [3]. hs scheme combnes dversty wth a ow-compexty recever structure. he key dea of [3] s to transmt two symbos smutaneousy over the two X antennas. At the next symbo perod the same symbos are transmtted agan swappng the X antenna eements for transmsson of each symbo. A smpe codng of the symbos makes t possbe for the recever to detect the receved symbos wth ony sma effort. In order to descrbe the OFDM verson of the Aamout scheme we must consder the transmsson at two consecutve OFDM symbo perods OFDM symbo perod ν and OFDM symbo perod ( ν + 1. In the OFDM symbo perod ( 1 ν the data symbo d s transmtted on the th subcarrer over the frst X antenna and the data ( symbo d s transmtted on the th subcarrer over the second X antenna. Both data symbos form the data vector ( 1 ( ( d = d d = 1 L. (13 the data ( symbo d s transmtted on the th subcarrer over the frst X antenna and the data symbo ( 1 + d s transmtted on the th subcarrer over the second X antenna. Wth (13 wth the subcarrer specfc K channe matrx ( 11 ( ( 1 ( 11 + = = 1 L (14 ( K1 ( K + + ( K ( K 1 + and wth the subcarrer specfc nose vector In the OFDM symbo perod ( ν + 1 (( 1 ν (( 1 ν 1 ( K ( ( K( 1 ( + ν ν + n = n n n n = 1 L. (15 the receved vector (( 1 ν (( 1 ν 1 ( K ( ( K ( 1 ( + ν ν + r = r r r r = 1 L (16 assocated wth the th subcarrer s gven by 30

4 r = d + n = 1" L. Fg. 1. (17 (Quadrature Amptude Moduaton Aamout and 16-QAM V-BLAS as we as 16-QAM SICMMSE w be compared to 56-QAM Aamout respectvey as the V-BLAS and SIC-MMSE schemes have twce the symbo rate for MIMO. In a cases rea decson feedback equazaton (rdfe [5] w be taken nto account when anayzng the V-BLAS and SIC-MMSE performance. Furthermore we w focus on a x MIMO channe wth uncorreated MIMO branches. In a smuatons perfect channe estmaton s consdered. Uncoded BE for QPSK moduaton x-mimo uncorreated MIMO branches In order to detect the receved symbos the maxmum-kehood (ML rue ½ d = arg max e d r d ¾ k (18 d { } = 1" L can be apped [3]. he ML rue n (18 can be smpfed by notng that the channe matrx s aways orthogona regardess of the actua vaues of the channe coeffcents. hen = K k =1 { ( k 1 + ( k } 10 =I = E Fg.. Uncoded BE for 16-QAM moduaton xmimo uncorreated MIMO branches Fg. 3. Uncoded bt error rato (BE performance x spata channe mode wth uncorreated MIMO branches 0 = 1" L. 1¹ (19 herefore (18 becomes ½ d = arg max e d r E d ¾ k d { } (0 = 1" L. r n (0 s the Aamout near equazer n the case of K K ` X antennas. In the case of constant d e.g. when phase shft keyng moduatons are depoyed (0 further smpfes to become (1 d = arg max e d r = 1" L. k d { { }} III. SIMULAION ESULS In [1] a spata channe mode termed SCM has been ntroduced for MIMO reference smuaton purpose. o have a far comparson QPSK (Quadrature Phase Shft Keyng V-BLAS and QPSK SIC-MMSE w be compared to 16-QAM 31 Frst the uncoded bt error rato (BE Pe sha be addressed. Fg. 1 Fg. and Fg. 3 show smuaton resuts for the BE as a functon of the overa sgna to nose rato 10og10 ( Eb N 0 for the V-BLAS and the SIC-MMSE detectors as we as

5 for Aamout X dversty wth dua X antenna dversty. At ow SN vaues the SIC schemes exhbt steeper BE curves than ther near varants. At hgh SN vaues the SIC schemes and the near recever performance curves show the same sopes. he ML recever provdes the steepest BE curves. Nevertheess the SIC-MMSE recever provdes an overwhemngy promsng performance. At hgh SN vaues the near recevers.e. the MMSE and the ZF recevers agan provde the worst performance whereas the SIC versons factate performance mprovements. acheved by usng spata mutpexng athough a hgher SN requrement must be met. herefore both spata mutpexng as we as dversty schemes are seen as competng and vabe means to ncrease the data rate n future mobe rado systems. IV. CONCLUSIONS In ths manuscrpt the authors dscussed spata mutpexng compared wth transmt and receve dversty schemes for MIMO transmsson. In partcuar V-BLAS SIC-MMSE and Aamout type transmsson were taken nto account. Speca focus was ad on the downnk transmsson of UA LE. he SIC-MMSE recever structure was ustrated and ts performance was anayzed n smuatons. It was found that the SIC-MMSE outperforms the V-BLAS scheme. Athough the SIC-MMSE cannot provde recepton as robust as n the case of spata ML recevers t aows a benefcay ow mpementaton compexty and therefore s a vabe canddate for termnas usng MIMO technques. Furthermore smuaton resuts showed that both dversty and spata mutpexng are vabe means to mprove the data rates. Fg. 4. Measured throughput η n bt/s/z for the SIC- MMSE recever Chase combnng (CC or ncrementa redundancy (I x spata channe mode wth uncorreated MIMO branches Fg. 4 shows the evauated throughput η measured n bt/s/z n the case of coded transmsson and SIC-MMSE recepton usng the UMS urbo Code accordng to reease 5 wth varyng code rates. he smuaton resuts consder the hybrd automatc repeat request (-AQ procedure specfed n the UMS standard. Both Chase combnng (CC as we as the seected ncrementa redundancy (I scheme of UMS have been consdered. he chosen code rates vary between 1/3 1/ /3 and 3/4. he data moduaton schemes are ether QPSK or 16-QAM wth maxmum throughput vaues of 3.36 bt/s/z and 6.7 bt/s/z respectvey. It has been found that spata mutpexng technques are nferor w.r.t. the robustness of the transmsson. Nevertheess throughput anayses showed that maxmum throughputs can aso be V. ACKNOWLEDGMEN he authors wsh to gratefu to ther coeagues at Infneon echnooges and at the Lehrstuh für Kommunkatonsechnk of the Unverstät Dusburg-Essen. Parts of ths work have been carred out wthn the scope of the EUEKA MEDEA+ project MIMOWA party funded by the German BMBF. EFEENCES [1] hrd Generaton Partnershp Project; echnca Specfcaton Group ado Access Network: Physca Layer Aspects for Evoved UA ; 3GPP [] hrd Generaton Partnershp Project; echnca Specfcaton Group ado Access Network: equrements for Evoved UA (E-UA and Evoved UAN (E-UAN; (eease 7 ; 3GPP V [3] Aamout S.: A Smpe ransmt Dversty echnque for Wreess Communca-tons. IEEE Journa on Seected Areas n Communcatons vo. 16 (1998 pp [4] Wonansky P. W.; Foschn G. J.; Goden G. D.; Vaenzuea. A.: V BLAS: An Archtecture for eazng Very gh Data ates Over the ch-scatterng Wreess Channe. Proceedngs of the USI Internatona Symposum on Sgnas Systems and Eectroncs (1998. New York/NY pp [5] Schoand.; Spege C.; Berkmann J.; Ba Z.; Drewes C.; Gunzemann B.; Bruck G..; Jung P.: MIMO Succesve Interference Canceaton for UA LE. Proceedngs of the 1th Internatona OFDM-Workshop (InOWo' August 007 amburg. [6] Kay S.M.: Fundamentas of Statstca Sgna Processng Estmaton heory. Page 386 Engewood Cffs: Prentce a