Performance of STBC transmissions with real data

Transcription

1 Performance of STBC transmssons wth real data José A.García-Naya Tago M. Fernández-Caramés Héctor J. Pérez-Iglesas Mguel González-López and Lus Castedo Dpt. Electrónca y Sstemas Unversdad de A Coruña 1571 A Coruña, SPAIN E-mal: {jagarca,tmfernandez,hperez, mgonzalezlopez,lus}@udc.es Davd Ramírez Ignaco Santamaría Jesús Pérez and Javer Vía Dpt. de Ingenería de Comuncacones Unversdad de Cantabra 395 Santander, SPAIN E-mal: {ramrezgd,nacho,jperez, jva}@gtas.dcom.uncan.es José M.Torres-Royo IMS System Engneerng Motorola Inc. 87 Madrd, SPAIN E-mal: Jose.Mguel.Torres@motorola.com Abstract Ths paper presents a comparatve study of three Space-Tme Block Codng (STBC) technques n realstc ndoor scenaros. In partcular, we focus on the Alamout orthogonal scheme consderng two types of Channel State Informaton (CSI) estmaton: a conventonal plot-aded technque and a new blnd method based on Second Order Statstcs (SOS). We also consdered a Dfferental (non-coherent) Space-Tme Block code (DSTBC) that can be optmally decoded wthout CSI estmaton, although t ncurs n a 3 db loss n performance. Expermental evaluaton s carred out wth a flexble and easy-touse MIMO platform at. GHz. Results show the excellent performance of the blnd channel estmaton technque n ether Lne-Of-Sght (LOS) and Non-LOS (NLOS) ndoor scenaros. I. INTRODUCTION Snce the poneerng work of Foschn and Telatar [1], [], multple transmt and receve antennas have been used to drastcally mprove the performance of wreless communcaton systems. Specfcally, snce the work of Alamout [3], and the later generalzaton by Tarokh et al. [], spacetme block codng (STBC) has emerged as one of the most promsng technques to explot spatal dversty n Multple- Input Multple-Output (MIMO) systems. Among space-tme codng schemes, orthogonal space-tme block codng (OSTBC) s one of the most attractve because t s able to provde full dversty gan wthout Channel State Informaton (CSI) at transmsson and wth very smple encodng and decodng procedures. The specal structure of OSTBCs enables optmal Maxmum Lkelhood (ML) decodng usng a smple lnear recever followed by a symbol-by-symbol detector. The CSI requred for coherent detecton of OSTBCs s typcally acqured by sendng a tranng sequence that s known at the recever sde [5]. However, the prce to be pad s reduced spectral effcency, energy loss because tranng sequences do not carry any nformaton and naccurate channel estmates due to the effect of the nose and the lmted number of tranng symbols. Popular approaches to avod the reducton on the spectral effcency nclude the so-called Dfferental STBC (DSTBC) schemes [6] [8] and Untary Space-Tme Modulaton [9], [1]. These schemes do not requre channel knowledge at the recever but they ncur n a performance penalty of 3 db (dfferental codng) and db (untary modulaton) as compared to the coherent ML recever [9]. Moreover, the recever complexty for the untary scheme ncreases exponentally wth the number of ponts n the untary space-tme constellaton. In order to overcome the lmtatons of dfferental codes whle, at the same tme, avodng the spectral effcency reducton of plot-aded technques, several methods for blnd channel estmaton have been proposed [11], [1]. These methods can be dvded nto two groups dependng on whether they explot the hgher-order statstcs (HOS) or the second-order statstcs (SOS) of the sgnals. HOS-based methods exhbt two major drawbacks: they present, n general, a hgher computatonal cost and may requre long streams of data to acheve accurate estmates. On the other hand, SOS-based methods are preferable n practce. Recently, a reduced-complexty SOSbased method for blnd channel estmaton under OSTBC transmssons has been proposed n [13]. Its performance has been evaluated by means of numercal examples, fndng that n most cases t renders accurate channel estmates, provded that n R > 1 receve antennas are avalable. However, for some OSTBCs (ncludng Alamout) some ambgutes appear that have to be avoded, for nstance, usng lnear precodng at the transmtter or resortng to HOS. In ths paper, we focus on the evaluaton of several of the above STBC transmsson technques over realstc ndoor scenaros. To ths end, we make use of a MIMO testbed desgned to operate at the. GHz Industral, Scentfc and Medcal (ISM) band. Due to the lmtatons n the number of transmt antennas, we are constraned to the Alamout code [3] and the dfferental STBC for two transmt antennas [6]. For Alamout coherent decodng, we have employed a plot-aded CSI estmaton technque [5] and the blnd technque proposed n [1], whch avods the ndetermnacy problems of [13] by reducng n a few bts per second the transmsson rate.

2 II. STBC AND DSTBC SCHEMES Throughout ths paper, we wll consder a flat fadng MIMO channel wth n T transmt and n R receve antennas. Ths channel s convenently represented as a matrx H of dmenson n R n T,whereeachofthecoeffcentsh j represents the complex transfer functon between the jth transmtter to the th recever. The transmtted symbols are grouped n blocks of sze M symbols and then encoded usng a Space-Tme Block Code (STBC) nto the codeword matrx S[n] wth dmenson n T L where L s the codeword number of tme slots. Notce that the transmsson rate of ths system s R = M/L. After transmsson through the MIMO channel, the nth block of receved sgnals s represented wth the n R L matrx X[n] =HS[n]+N[n] (1) where N[n] s a n R L matrx representng the spatally and temporally Addtve Whte Gaussan Nose (AWGN). Assumng perfect knowledge of H and takng nto account the Gaussan dstrbuton of the nose, the coherent Maxmum Lkelhood (ML) decodng of S[n] s obtaned after mnmzng the followng crteron [15] Ŝ ML [n] =argmn X[n] HS[n], () S[n] subject to the constrant that the elements of Ŝ[n] belong to a fnte set S. ThssaNP-hardproblemandoptmalalgorthms to solve t, such as sphere decodng, canbecomputatonally expensve [16] [18]. The complexty of the ML recever reduces consderably when resortng to Orthogonal STBCs (OSTBC) n whch the codeword matrces are orthogonal,.e., S[n]S H [n] =I nt (3) where the superscrpt H denotes transpose conjugate and I nt s the n T n T dentty matrx. The ML decodng of OSTBCs s equvalent to M parallel symbol-by-symbol detecton at the output of a lnear recever. The most popular OSTBC s the Alamout code [3], whch transmts M =complex symbols n L =tme slots (.e., the code rate s R =1). The codewords n the Alamout code are constructed as [ ] s1 [n] s S[n] = [n] s [n] s () 1[n] In ths work we restrct ourselves to the Alamout code because of the lmtaton n the number of transmttng antennas of the testbed used n the experments. The use of a platform lmts the use of more sophstcated OSTBCs. Coherent OSTBC decodng requres CSI knowledge at the recever. Ths mposes practcal constrants that can be avoded wth the utlzaton of dfferental schemes. In our comparatve study, we consder the Dfferental Space-Tme Block Codng (DSTBC) for two transmt and two receve antennas descrbed n [15]. Ths partcular type of DSTBC s restrcted to constant modulus sgnals. In ths DSTBC scheme the transmtted codeword matrces (Z[n]) are constructed as follows Z[n] =Z[n 1]S[n] wth Z[] = I nt. When the matrces S[n] are the output of a OSTBC encoder, ML decodng of DSTBC reduces to M parallel symbol-by-symbol detecton at the output of a lnear recever. The advantage of DSTBC s that decodng can be carred wthout the need of knowng the channel at the recever. Ths advantage, however, s obtaned at the expense of a 3 db penalty n comparson wth coherent detecton. III. CHANNEL ESTIMATION IN MIMO-OSTBC SYSTEMS In ths secton we descrbe the channel estmaton technques used n the experments for Alamout decodng. Frstly, we consder the conventonal plot-based technque and, secondly, we descrbe a recently proposed blnd technque. A. Plot-aded channel estmaton We have appled the channel estmaton method descrbed n [5]. Bascally, we need to construct n T orthogonal plot sequences of sze K. [ ] S plot [ ] = 1 s11 s = 1... s 1K (5) s 1 s... s K The plot sequences are desgned to be orthogonal ( ) H δ l l where δ l s the Kronecker delta. Ths orthogonalty among the plot sequences allows us to ndependently estmate each fadng coeffcent h j.specfcally,themnmummeansquare Error (MMSE) estmate of h j s gven by ( ĥ j = x plot j ) H j (6) where x plot s the receved sgnal at the -th antenna when S plot has been transmtted. On the other hand, the transmsson of a plot sequence causes a reducton n the effectve E b /N or, equvalently, a reducton n the effectve transmsson rate. For nstance, f we transmt N D data symbols and K plots durng the n-th frame, the transmtted rate assocated to ths technque s R pl = N D N D + K. B. SOS-based blnd channel estmaton Recently, a new method for blnd channel estmaton under OSTBC transmssons has been proposed n [13]. It s based only on Second Order Statstcs (SOS) and t s able to blndly dentfy the channel (up to a real scalar ambguty) for most of the exstng OSTBCs when the number of receve antennas s n R > 1 [19]. However, some OSTBCs (ncludng the Alamout code used n ths paper) cannot be dentfed by ths method due to an addtonal ambguty, whch must

3 Fg. 1. Schematc dagram of the MIMO platform. be elmnated by resortng to other nformaton (e.g., lnear precodng, non-whte source sgnals, reduced rate, etc.) [13] In ths paper we use a partcularly smple method whch has been recently proposed n [1]. There, t was proved that any OSTBC transmttng an odd number of real symbols, M, s dentfable regardless of the number of recevng antennas. The number of real symbols, M,naOSTBCcodewords { M M for real constellatons, = (7) M for complex constellatons. Therefore, any non-dentfable complex OSTBC can be made dentfable smply by not transmttng one real symbol per OSTBC block. Obvously the transmsson rate s reduced, but ths rate penalty can be controlled by elmnatng only one real symbol each tme B OSTBC codewords are transmtted. In ths case, the transmsson rate s R blnd = BM 1 BM (8) whch tends to one for B 1. Obvously,theresatrade-off between the qualty of the channel estmate and R blnd as a functon of B. Thsssuehasbeendscussedn[1]. IV. MIMO TESTBED In the results that follow, we examne the performance of STBC schemes n realstc scenaros usng real data from ndoor envronments. The real data was obtaned usng a flexble and easy-to-use MIMO testbed, jontly bult at the Unverstes of Cantabra and A Coruña (Span). Ths MIMO testbed s ntended for testng and rapd prototypng of MIMO baseband modules. A schematc dagram of the platform s shown n Fg. 1 and a pcture of the system s shown n Fg.. Its basc operaton s as follows: sgnal generaton, modulaton and space-tme codng at transmsson are carred out off-lne usng MATLAB R.ThetransmttngPCcontansaboardto generate the analog sgnals at an IF of 15 MHz. Snce ths board s equpped wth a large (1 GB) and fast memory, the versatlty of the platform s extremely hgh. The upconverson from IF to the carrer RF frequency of.385 GHz s performed by two Aglent ESG E38C sgnal generators and the sgnals are then transmtted through two prnted dpole antennas. At the recever sde, two downconverters specfcally desgned for ths platform translate the RF sgnal to IF. The IF sgnals are acqured by the receve host PC usng another board wth two ADCs wth a maxmum samplng frequency of 15 MHz. Another fast and hgh capacty (1 GB) memory module s used to store the acqured sgnals. The memory content can be subsequently downloaded nto the hard dsk of the recever host PC where synchronzaton, channel estmaton, demodulaton and decodng are performed off-lne usng MATLAB R. See [] for a detaled descrpton of the MIMO platform. Fg.. A pcture of the MIMO platform. V. EXPERIMENTAL RESULTS In ths secton we compare the performance of the Alamout scheme for plot aded channel estmaton, blnd channel estmaton and the Dfferental STBC that we have descrbed n sectons II and III. The measurements were taken n the laboratory of the Sgnal Processng Group at the Unversty of Cantabra. In the frst experment the transmtters and recevers were approxmately two meters away from each other, wth a clear Lne-Of-Sght (LOS) between them. In the second experment, the recevers were located farther away from the transmtters ( 1 meters) and the transmttng antennas were also moved to avod a clear lne-of-sght (see Fg. 3). To smplfy symbol and frame synchronzaton, we desgned aframestructurecomposedof63preamblesymbolsforframe synchronzaton, up to 6 plot symbols for channel estmaton (for plot-aded technques) and 1 data symbols (see Fg. ). In the preamble we use a pseudorandom sequence (PN) to facltate frame synchronzaton and coarse symbol tmng acquston. Notce that ths frame was selected to smplfy

4 Transmt antennas LOS Experment RX and TX antennas and clear LOS) when the transmtted power per antenna s -1 dbm. Transmt antennas NLOS Experment 1 1 Receve antennas Both Experments BER 1 Fg. 3. Locatons of the TX s and RX s n the two experments. the synchronzaton and estmaton algorthms and not to maxmze throughput. Regardng the modulaton parameters, we employ a QPSK modulaton. The pulse shapng flter s a square-root rased cosne flter wth a roll-off factor of.. The symbol rate s 1Mbaud,sotheRFbandwdths1.MHz.Thesamplng frequency s 8 Msamples/sec. at both the transmtter and the recever. Ths hgh samplng rate was used to smplfy the synchronzaton algorthms. At the recever, we perform carrer offset estmaton and elmnate the carrer modulaton. Afterwards, frame and symbol synchronzaton are carred out by explotng the PN preamble. The fnal baseband observatons are obtaned through matched flterng and samplng at the symbol rate. Preamble (63) Plots (6) Informaton (1) Fg.. Frame structure chosen for the experments. Acqured Sgnal Tmng and frequency synchronzaton Tmng synchronzaton After Alamout decodng plots 16 plots DSTBC Blnd (B1-N5) Blnd (B1-N5) Transmtted power per antenna(dbm) Fg. 6. BER for the LOS scenaro. We have repeated the experment varyng the transmttng power per antenna. For each transmttng power we repeated several tmes the experment. Snce the generaton and codng at the transmtter sde; and the demodulaton, channel estmaton and decodng at the recever sde are carred out off-lne, the tme between two consecutve trals s much larger that the coherence tme of the channel. Wth ths set-up we obtaned the bt error rate (BER) curve versus transmttng power shown n Fg. 6. In ths fgure we compare: The Alamout OSTBC wth plot-aded channel estmaton (labeled as K plots) The Alamout OSTBC wth blnd channel estmaton, labeled as Blnd (BX-NY), where X s the number of Alamout blocks n whch we elmnate one real symbol (to avod the ambguty) and Y s the number of blocks that we use to estmate the correlaton matrx. The DSTBC. Tab. I shows the correspondng rates for the consdered Alamout transmssons. Method Rate 6 Plots Plots.983 Blnd (B1-N5),975 Blnd (B1-N5),975 TABLE I RATE FOR THE DIFFERENT CHANNEL ESTIMATION METHODS. Fg. 5. Symbol constellatons at the recever. Fg. 5 shows the sgnal receved at one antenna (upper left), the sgnal after symbol tmng (upper rght), after carrer frequency offset and symbol tmng (lower left) and after Alamout decodng (lower rght) n the frst scenaro (close As we can see from Fg. 6 and Table I, the blnd technque wth N = 5 blocks practcally acheves the same performance as the plot-aded method wth 6 plots, but transmttng at a hgher rate. Ths mprovement s acheved at the expense of a moderate ncrease of the computatonal cost, snce the blnd technque has to obtan the man egenvector of a 8 8 correlaton matrx. On the other hand, we also observe

5 the expected 3 db loss for the DSTBC and that the plotaded method wth 16 plots losses about. db wth respect to the utlzaton of 6 plots. Usng more than 6 plots s not necessary because t does not mprove system performance. Fnally, f we use less blocks for channel estmaton n the blnd technque, the estmate of the correlaton matrx s worst and ths causes a loss n BER. Specfcally, f we use N = 5 nstead of N =5blocks, the loss s about.9 db. However, the use of a reduced number of blocks for blnd channel estmaton permts the use of shorter frames, whch s especally mportant when the channel coherence tme s smaller. Fgure 7 shows agan the BER versus the transmttng power for the dfferent methods n a NLOS scenaro. The man dfference wth respect to the frst experment s that now we have to ncrease almost 7 db the transmttng power of the RF sgnal generators to attan the same BER, but the comparson among the dfferent STBC transmsson leads to smlar conclusons n ths new scenaro. BER plots 16 plots DSTBC Blnd (B1-N5) Blnd (B1-N5) Transmtted power per antenna(dbm) Fg. 7. BER for the NLOS scenaro. VI. CONCLUSIONS In ths paper we have compared the performance of several STBC systems on real data obtaned from ndoor scenaros usng a MIMO platform at, GHz. In partcular, we have compared the Alamout orthogonal scheme wth coherent and non-coherent demodulaton. The channel was estmated usng two dfferent methods: a conventonal plot-aded technque and a recently proposed blnd algorthm based on SOS. In both LOS and NLOS scenaros, the blnd channel estmaton technque provdes smlar BER performance than the plotaded method, wth a slght ncrease n the effectve data rate and a moderate ncrease n the computatonal complexty of the detector. On the other hand, the dfferental STBC presents, as expected, a 3-dB penalty n comparson wth coherent schemes, but t can be of nterest due to ts smplcty and ts potental advantages n rapdly tme-varyng channels. ACKNOWLEDGMENT Ths work has been supported by Mnstero de Educacón ycencaofspan,xuntadegalcaandfederfundsofthe European Unon under grants number TEC-651-C5-, TEC-651-C5-1, PGIDT5PXIC15PN and FPU grant AP-517. REFERENCES [1] G. Foschn and M. Gans, On lmts of wreless communcatons n a fadng envronment when usng multple antennas, Wreless Personal Communcatons, vol.6,pp ,1998. [] I. E. Telatar, Capacty of mult-antenna Gaussan channels, European Trans. on Telecommuncatons, vol.1,no.6,pp ,Nov.-Dec [3] S. Alamout, A smple transmt dversty technque for wreless communcatons, IEEE J. Sel. Areas Comm., vol. 5, no. 9, pp , [] V. Tarokh, H. Jafarkhan, and A. R. Calderbank, Space-Tme block codes from orthogonal desgns, IEEE Transactons on Informaton Theory, vol.5,no.5,pp ,1999. [5] A. F. Nagub, V. Tarokh, N. Seshadr, and A. R. Calderbank, Aspacetme codng modem for hgh-data-rate wreless communcatons, IEEE J. Select. Areas Commun., vol.16,no.8,pp ,Oct [6] V. Tarokh and H. Jafarkhan, A dfferental detecton scheme for transmt dversty, IEEE Journal on Selected Areas n Communcatons, vol. 18, no. 7, pp , Jul.. [7] B. L. Hughes, Dfferental Space-Tme modulaton, IEEE Transactons on Informaton Theory, vol.6,no.7,pp ,Nov. [8] B. Hochwald and W. Sweldens, Dfferental Untary Space-Tme Modulaton, IEEE Transactons on Communcatons, pp.1 5,Dec.. [9] B. Hochwald and T. Marzetta, Untary space-tme modulaton for multple-antenna communcatons n raylegh flat fadng, IEEE Transactons on Informaton Theory, vol.6,no.,pp.53 56,Mar.. [1] B. Hochwald, T. Marzetta, T. Rchardson, W. Sweldens, and R. Urbanke, Systematc Desgn of Untary Space-Tme Constellatons, IEEE Trans. Inform. Theory, vol.6,no.6,pp ,. [11] C. Budanu and L. Tong, Channel estmaton for Space-Tme Orthogonal Block Codes, IEEE Transactons on Sgnal Processng, vol.5, no. 1, pp , Oct.. [1] P. Stoca and G. Ganesan, Space-Tme block codes: Traned, blnd, and sem blnd detecton, Dgtal Sgnal Processng, vol. 13, pp , Jan. 3. [13] S. Shahbazpanah, A. B. Gershman, and J. H. Manton, Closed-form blnd MIMO channel estmaton for orthogonal space-tme block codes, IEEE Trans. Sgnal Processng, vol.53,no.1,pp ,Dec. 5. [1] J. Vía, I. Santamaría, and J. Pérez, A Suffcent Condton for Blnd Identfablty of MIMO-OSTBC Channels Based on Second Order Statstcs, n Seventh IEEE Workshop on Sgnal Processng Advances n Wreless Communcatons, Cannes,France,July6. [15] E. G. Larsson, P. Stoca, and G. Ganesan, Space-Tme Block Codng for Wreless Communcatons. New York, USA: Cambrdge Unversty Press, 3. [16] U. Fncke and M. Pohst, Improved methods for calculatng vectors of short length n a lattce, ncludng a complexty analyss, Mathematcs of Computaton, vol.,pp.63 71,Apr [17] O. Damen, A. Chkef, and J. Belfore, Lattce code decoder forspace- tme codes, IEEE Comm. Lett., vol., no. 5, pp , May. [18] J. Jaldén, C. Martn, and B. Ottersten, Semdefnte Programmng for Detecton n Lnear Systems Optmalty Condtons and Space-Tme Decodng, n IEEE Internatonal Conference on Acoustcs, Speech, and Sgnal Processng, vol.,aprl3,pp.9 1. [19] J. Vía and I. Santamaría, On the Blnd Identfablty of Orthogonal Space-Tme Block Codes from Second Order Statstcs, Submtted to IEEE Transactons On Informaton Theory. [] D. Ramírez and et al, Aflexbletestbedfortherapdprototypngof MIMO baseband modules, n 3rd Internatonal Symposum on Wreless Communcaton Systems, Valenca, Span, September 6.