An Experimental Downlink Multiuser MIMO System with Distributed and Coherently-Coordinated Transmit Antennas

Transcription

1 An Experimental Downlink Multiuser MIMO System wit Distributed and Coerently-Coordinated Antennas Dragan Samardzija, Howard Huang, Reinaldo Valenzuela and Teodore Sizer Bell Laboratories, Alcatel-Lucent, 791 Holmdel-Keyport Road, New Jersey, 07733, USA dragan, cuang, rav, Abstract In tis paper we present a multiuser MIMO experimental system and te corresponding indoor measurement results tat demonstrate power of distributed and coerentlycoordinated downlink transmit antennas using zero-forcing (ZF) beamforming. We present experimental performance under different conditions and communication scenarios. We consider different transmit power levels, number of transmit antennas, resolution of cannel state information (CSI) quantizer as well as CSI feedback cannel data rates. We demonstrate significant gains even in te presence of practical impairments. I. INTRODUCTION In te MIMO broadcast cannel (MIMO-BC), a transmitter wit multiple antennas sends information over a given bandwidt to multiple receivers. Recent works [1] ave sown tat te sum capacity of te MIMO-BC can be acieved by means of a nonlinear precoding tecnique at te transmitter known as dirty paper coding (DPC) [2]. Using cannel state information (CSI) and a priori knowledge of te users' data signals, dirty paper coding transmits to multiple users simultaneously so tat interference among users' signals is greatly reduced. Unfortunately, tis tecnique as ig computational complexity and is difficult to implement in practice. Beamforming is a simpler, suboptimal tecnique were signals are simultaneously transmitted to multiple users on beams formed by weigting te pases and amplitudes of te transmitted signals. A particular type of beamforming known as zero-forcing (ZF) uses CSI at te transmitter to form noninterfering beams. It as been sown to acieve te sumcapacity of te MIMO-BC asymptotically as te number of users grows witout bound [3] and acieves a significant fraction of te DPC capacity for a finite number of users [4]. ZF beamforming tecniques ave been considered in wireless cellular communication systems for increasing downlink spectral efficiency. Typically, multiple transmit antennas are co-located at a base station so tat beams formed by a given base are non-interfering. However, a user receiving a signal from te beam of one base may experience interference from anoter base's beam. In cannels were te angular spread is sufficiently small, tis intercell interference could be reduced by coordinating base stations so tat a given area is not simultaneously illuminated by beams from different base stations. Alternatively, one could perform ZF beamforming over antennas tat are spatially distributed trougout te network. If CSI is known trougout te network for eac user and te base stations transmit in a coordinated manner, ten ideally te users would experience no intercell interference and te spectral efficiency would be greatly improved. A ig-speed backaul network would be required to connect te base stations. Eiter CSI would need to be transferred among base stations so tat te ZF antenna weigts are calculated locally at eac base, or te CSI would be transferred to a common point were ZF antenna weigts are computed and ten distributed to eac antenna. Realistically, a fully coordinated macro-cellular network would require significant long-term planning. However, near-term applications could include an in-building pico-cell network or a limited outdoor area for otspot coverage. Tese applications would be ideally suited for coordinated transmission since te need for spectral efficiency is ig and te relatively sort distances between base stations makes coordination feasible. In tis paper we describe an experimental prototype tat was built for demonstrating ZF beamforming across spatially distributed antennas. In Section II we present te system model. In Section III we describe te experimental system implementation including te transmitter, receiver, and CSI feedback mecanism. In Section IV we present various experimental results. We conclude in Section V. II. SYSTEM MODEL We consider a narrowband multiantenna downlink cannel modeled as a MIMO-BC wit flat fading, were K users, i.e., mobile terminals, eac equipped wit a single receive antenna, request service from a transmitter wit M distributed antennas. Te discrete-time complex baseband received signal by te kt user is y k = k x + n k, k = 1,, K (1) were k is te kt user's M-dimensional complex row cannel vector, x is te transmitted signal vector, and n k ~ η(0, 1) is te complex additive wite Gaussian noise wit zero mean and unit variance. Te noise is uncorrelated among te users but te signal vector is te same for all users. Under zero-forcing beamforming, te transmitted vector x = Gu were u is te K-

2 dimensional information-bearing signal containing te modulation symbols for eac user, and G = H H (HH H ) -1 is te M x K beamforming matrix were H te K x M cannel matrix. Note tat we assumed tat te transmitter as perfect knowledge of te CSI. In tis case, we can rewrite te received signal by te kt user in (1) as y k = u k + n k, k = 1,, K. Hence te rate acieved by tis user is simply log 2 (1+ν k ), were ν k =E[ u k 2 ] is te power allocated to te kt user. Te transmit power at te mt antenna is G m1 2 ν G mk 2 ν K. Hence te sum rate optimization subject to a transmit power constraint P m on te mt antenna can be written as: v,..., v 1 K k= 1 max log(1 + v ), subject to K vk 0, k = 1,..., K K 2 G, 1,..., k 1 mk vk Pm m= M = Tis optimization as been sown to be a convex optimization [4] wic can be solved using conventional interior point tecniques. k Base Station Te functional block diagram of te base station is presented in Fig. 2. Te base station receives CSI from te mobile terminals. Specifically, mobile terminal k feeds back estimates of te kt row vector of te MIMO cannel matrix H (for k = 1, 2). Having te CSI received, te base station determines te beamforming coefficients. Te output of te generalized beamformer is x = Gu, were x and G is te beamforming matrix. Furtermore, 2 1 u is te input vector were u k is its kt element and it corresponds to te signal dedicated to mobile terminal k (k = 1, 2). Te signal u k consists of te information-bearing signal as well as dedicated pilot k (i.e., reference signal). Dedicated pilot k allows te mobile terminals to estimate te kt column entries of te composite MIMO cannel matrix GH. Te particular estimates are used by te mobile terminals to perform coerent detection of te information bearing portion of te signal u k (for k = 1, 2) witout any explicit knowledge of beamforming tat is performed by te base station. To simplify te estimation procedure, te dedicated pilots are ortogonal in signal space to te information-bearing portion of te signals, and to eac oter. III. EXPERIMENTAL SYSTEM Te experimental system tat is used in tis study is depicted in Fig. 1. It consists of M = 4 distributed downlink transmit antennas and K = 2 mobile terminals, eac wit a single receive antenna. Te transmit antennas are connected to te outputs of te beamformer. Te beamformer resides in a base station tat determines its coefficients and applies tem on te information-bearing signals for eac mobile terminal. Using a pilot-assisted cannel state estimation, te mobile terminals estimate te downlink cannel states between eac transmit and receive antenna. Te quantized estimates, tat correspond to cannel state information (CSI), are fed back to te base station, were tey are used to determine te beamforming coefficients. In te following we provide more details on te functional blocks of tis experimental system. Dedicated Pilot 1 Beamforming Coefficient Calculator u 1 Dedicated Pilot 2 u 2 Generalized Beamformer G Antenna Pilot 1 x 1 Antenna 1 Antenna Pilot 2 x 2 Antenna 2 Antenna Pilot 3 x 3 Antenna Pilot 4 x 4 Antenna 3 Antenna 4 Antenna 1 Uplink CSI Feedback Receiver 1 CSI Feedback 1 Antenna 2 Mobile Terminal 1 Uplink CSI Feedback Receiver 2 CSI Feedback 2 Base Station Antenna 3 Antenna 4 CSI Feedback Mobile Terminal 2 Fig. 1: Experimental system overview. Fig. 2: Base station functional blocks. Before being transmitted over te corresponding downlink transmit antenna, a unique pilot is added to eac beamformer outputs. Te antenna-specific pilots allow mobile terminals to perform pilot-assisted estimation of te MIMO cannel H. Specifically, mobile terminal k estimates te kt row vector of te matrix H. Te quantized estimates correspond to te CSI tat is fed back to te base station. To simplify te estimation

3 procedure, te antenna-specific pilots are ortogonal in signal space to te beamformer outputs and to eac oter. Mobile Terminal Te functional block diagram of te mobile terminal is presented in Fig. 3. It consists of a number of estimators tat implement maximum likeliood estimation [5]. Specifically, in tis implementation te mobile terminal functionality is dedicated to assessing te performance of te distributed antenna beamforming. Te estimators tat correspond to dedicated pilot 1 and 2 are used to estimate te received power of te signal tat is dedicated to mobile terminal 1 and 2, respectively. Ŝ kl denotes estimated power of mobile terminal l signal tat is received at mobile terminal k. For example, in te case wen idealized ZF beamforming and estimation are applied, te received power estimates sould be S ˆ ˆ 12 = S21 = 0 wit S ˆ11 0 and S ˆ22 0, wic is a direct consequence of te zero-forcing criterion. Noise Dedicated Pilot 1 N ˆ k 0 S ˆk1 SINR Calculator SINR k were ˆN 10 and ˆN 20 are estimates of te noise power at mobile terminal 1 and 2, respectively. Te antenna-specific pilots are used by te mobile terminals to perform pilot-assisted estimation of te MIMO cannel H. As said earlier, mobile terminal k estimates te kt row vector ˆ ˆ ˆ k = [ k1 k4] of te matrix H. Te quantized estimates correspond to te CSI tat is transmitted back to te base station, were is used to calculate te beamforming coefficients. Signal Arrangement Te symbol rate of te experimental system is set to Msym/sec, wic is identical to te cdma2000/ev-do cip rate. Te antenna-specific pilots span 128 symbols, lasting usec, and tey are code-multiplexed. Te dedicated pilots ave te identical duration and multiplexing sceme. Te pilot codes are mutually ortogonal and quasi-random wit a low cross correlation. Immediately after te antenna-specific pilots, te dedicated pilots are transmitted. Furtermore, te information-bearing portion (payload) of te signal follows after te dedicated pilots. Te payload duration may vary depending on a particular experimental setup. Te temporal relationsip between te pilots and te payloads is depicted in Fig. 4. Dedicated Pilot 2 S ˆk 2 Antenna Pilot 1 Receive Antenna Antenna Pilot 1 Antenna Pilot 2 Antenna Pilot 3 Antenna Pilot 4 Uplink CSI Feedback ter ˆk1 ˆk 2 ˆk 3 ˆk 4 Syncronization Antenna Pilot 2 Antenna Pilot 3 Antenna Pilot 4 Dedicated Pilot 1 Dedicated Pilot 2 Payload 1 Payload symbols 128 symbols variable lengt Fig. 4: Temporal relationsip between te antenna-specific pilots, dedicated pilots and te payloads. Note tat te beamforming is not performed wile te antenna-specific pilots are being transmitted. Te beamforming is applied during te transmission of te dedicated pilots and payloads. Fig. 3: Mobile terminal functional blocks, k = 1 or 2. As a figure of merit we consider signal-to-noise-plusinterference ratio (SINR) 11 SINR 1 = and 22 SINR 2 = + Nˆ + Nˆ Cannel State Information Feedback Mobile terminal k periodically obtains te cannel state estimates ˆ ˆ ˆ k = [ k1 k4], (k = 1, 2). Te sortest period between new estimates is 256 symbols, i.e., usec. Te sortest period corresponds te case wen no payload is transmitted.

4 Before being transmitted on te uplink, eac component of te vector ˆ k is quantized wit a 15-bit liner quantizer bot for te real (I) and imaginary (Q) component. Te CSI feedback cannel is realized over a cable (CAT5), wic is a igly controllable medium allowing us to assess te system performance under different CSI feedback cannel conditions. In te future we plan to implement a wireless CSI feedback cannel. Implementation Platform and Features Te most of te functional blocks tat are depicted in Fig. 2 and 3 are implemented on a FPGA platform (Xilinx Virtex II 6000), using a small fraction of te available resources. Only te beamformer coefficient calculator is implemented on a floating point DSP (TI 6701). Te mobile terminal front-end is based on a eterodyne arcitecture, followed by a 10-bit analog-to-digital converter. Te carrier frequency is set to 2.1 GHz. Furtermore, te downlink transmitters and eac mobile terminal ave free running oscillators tat are manually tuned. Te measured frequency offsets and clock drifts are negligible compared to te estimation periods and wireless cannel coerence periods. IV. EXPERIMENTAL RESULTS Te experimental results tat are presented in tis section are based on te measurements tat are conducted in our laboratory. Te laboratory floor plan is given in Fig. 5, were te solid lines correspond to its walls. Typical laboratory furniture (bences and wooden cabinets), computers and instruments occupy te laboratory, tus, creating a ric scattering environment. Eac downlink transmit antenna (TX1 to TX4) is placed close to a different corner of te laboratory. Te exact positions of te transmit antennas are given in Fig. 5. Te eigt of eac transmit antennas is 2 m. During te measurements, te mobile terminals (M1 and M2) are placed randomly witin te saded region in Fig. 5. Per eac measurement setup, 100 SINR measurements are recorded eac corresponding to different positions of mobile terminal 1 and m 1.2 m 1.9 m 1.3 m TX1 TX4 7.5 m M1 2 m TX2 TX3 Fig. 5: Laboratory floor plan. For te results tat are presented in tis section, te CSI is fed back every msec (480 times per second), corresponding to 5% overead for te antenna-specific pilots as well as 5% overead for te dedicated pilots. Furtermore, if M2 3 m 0.8 m 0.8 m 1.25 m 3.5 m 1.25 m not stated oterwise, te 15-bit linear quantizer is used for te real (I) and imaginary (Q) component of te cannel state estimate. In all cases te total average transmit power is identical. For example, in a case wen te number of transmit antennas is M = 4 te average transmit power per one antenna is 4 times lower tan in a case wit te single transmit antenna, M = 1. In Fig. 6 we present cumulative distribution function (CDF) of SINR for different transmission scemes. In addition to ZF beamforming tat is described in te previous sections, we present a single user case. In te single user case, te information-bearing signal is sent only to one of te users. Te signal is sent from transmit antenna 1 (TX1). Te single user case would correspond to a time-division multiplexing were transmissions for different users are ortogonalized in time. Furtermore, we considered a case wen no beamforming is applied. Specifically, te information-bearing signal for mobile terminal 1 is transmitted only from transmit antenna 1 (TX1). Concurrently, te information-bearing signal for mobile terminal 2 is transmitted only from transmit antenna 2 (TX2). In oter words, two independent transmissions are appening at te same time witout any interference mitigation sceme being applied. Fig. 6: CDF of SINR for different transmission scemes. Te corresponding CDF of te acievable downlink sum data rates is presented in Fig. 7. Te rates account for te pilot overead. Te pilot overead for ZF beamforming is 10% wile for te oter two scemes it is 5% (because te antennaspecific pilots are not needed). Te results demonstrate significant gains for ZF beamforming. In Fig. 8 we present CDF of SINR for different levels of te transmit power wen ZF beamforming is applied (for M = 4 and K = 2). Note tat even toug te transmit power is varied by 10 db, te cange of te measured SINR is smaller. Tese are te caracteristics of an interference-limited system. We believe tat tis is a consequence of a number of practical impairments. For example, imperfect cannel state estimation, CSI quantization and delay in its feedback are some of te impairments leading to te interference-limited beavior [6]. However, even in te presence of te practical impairments,

5 based on te results in Fig. 6 and 7, ZF beamforming gains are significant compared to te oter solutions tat are considered. of ZF beamforming wit two transmit antennas M = 2, and te 15-bit quantizer is similar or better tan te performance of te same beamforming sceme wit four transmit antennas M = 4, and te 8-bit quantizer. In bot cases te required CSI feedback data rate is comparable, i.e., 28.8 kbps versus kbps. In oter words, we could lower te number of transmit antennas, increase te quantizer resolution, and maintain te performance and te CSI feedback data rates. Fig. 7: CDF of downlink sum data rates for different transmission scemes. Fig. 9: CDF of SINR for different number of transmit antennas. Fig. 8: CDF of SINR for different transmit power levels. In Fig. 9 CDF of SINR is presented for different number of te downlink transmit antennas and ZF beamforming (for M = 2, 3 and 4 and K = 2). As expected, te performance is improved as te number of antennas is increased. As said earlier, in all cases te total average transmit power is identical. In Fig. 10 CDF of SINR is presented for different resolutions of te CSI quantizer and ZF beamforming (for M = 4 and K = 2). Considering tat te CSI is fed back every msec, te presented 15-, 8-, 6- and 4-bit quantizer correspond to te CSI feedback data rates of 57.6 kbps, kbps, kbps and kbps, respectively. Note tat te applied quantizer is linear and not optimal. Terefore te presented results sould be viewed as a lower bound on performance tat may be improved if better CSI quantizers are applied. Let us now consider a tradeoff between te number of transmit antennas and te CSI feedback data rates. For example, based on te results in Fig. 9 and 10 te performance Fig. 10: CDF of SINR for different resolution of te CSI quantizer. V. CONCLUSION We ave presented te multiuser MIMO experimental system and te corresponding indoor measurement results tat ave demonstrated power of distributed and coerentlycoordinated downlink transmit antennas using ZF beamforming. We ave presented experimental performance under different conditions and communication scenarios. We ave considered different transmit power levels, number of transmit antennas, resolution of te CSI quantizer as well as te CSI feedback cannel data rates. We ave demonstrated significant gains even in te presence of practical impairments.

6 ACKNOWLEDGEMENTS Te autors are grateful to Clark Woodwort and Susan Walker for teir support during te implementation of te experimental prototype. References [1] H. Weingarten, Y. Steinberg, S. Samai, "Capacity region of te MIMO broadcast cannel," IEEE Trans. Inform. Teory, vol. 52, no. 9, pp , September [2] M. Costa, "Writing on dirty paper," IEEE Trans. Inform Teory, vol. 29, no. 3, pp , May [3] T. Yoo, A. Goldsmit, "On te optimality of multiantenna broadcast sceduling using zero-forcing beamforming," IEEE J. Select. Areas Commun., vol. 24, no. 3, pp , Mar [4] F. Boccardi, H. Huang, "Zero-forcing precoding for te MIMO broadcast cannel under per-antenna power constraints," IEEE Worksop on Signal Processing Advances in Wireless Communications, July [5] D. Samardzija, N. Mandayam, "Pilot assisted estimation of MIMO fading cannel response and acievable data rates," IEEE Trans on Sig. Proc., vol. 51, pp , Nov [6] D. Samardzija, N. Mandayam, D. Cizik, "Adaptive transmitter optimization in multiuser multiantenna systems: teoretical limits, effect of delays and performance enancements," EURASIP Journal on Wireless Communications and Networking, no. 3, pp , Aug