01220272.pdf

Embed Size (px)

Citation preview

  • 7/27/2019 01220272.pdf

    1/3

    370 IEEE COMMUNICATIONS LETTERS, VOL. 7, NO. 8, AUGUST 2003

    Multiuser Diversity for a Dirty Paper ApproachZhenyu Tu, Student Member, IEEE, and Rick S. Blum, Senior Member, IEEE

    AbstractMultiuser diversity has attracted significant attention

    recently. In this letter, we propose a greedy algorithm based on QRdecomposition of the channel and dirty paper precoding to exploitthe multiuser diversity of the Gaussian vector broadcast channel.Simulations show the approach provides performance which is ex-tremely close to a well-known upper bound on the sum rate. Fur-ther, exploiting multiuser diversity can provide large gain over ap-proaches ignoring this resource.

    Index TermsBroadcast channel, dirty paper, multiple inputmultiple output (MIMO), sum rate.

    I. INTRODUCTION

    I

    N RECENT years, using antenna arrays to form multipleinput multiple output (MIMO) systems has shown promise

    for achieving high capacities over wireless links [1], [2]. TheMIMO enhanced broadcast channel is of great interest lately asobserved by the large number of papers at [3]. Finding simpleways to achieve performance close to capacity is still of greatinterest for the MIMO enhanced broadcast channel. Recent re-sults in [5] demonstrate that the Sato bound [4] is achievable,but do not tell how to achieve this bound. In [6], a simple ap-proach based on QR decomposition is proposed to optimize thesum rate for channel matrices with full row rank. It is referred aszero-forcing dirty-paper (ZF-DP) coding. For a channel matrixto have full row rank, the system must have as many transmitantennas as users which is quite restrictive. In this paper, we ex-tend ZF-DP to a scenario in which the channel does not have

    full row rank. This is accomplished by addressing the orderingand selection of active users to exploit the multiuser diversity.The resulting approach is a simple greedy algorithm referred toas greedy ZF-DP. Simulation results show it provides perfor-mance very close to the Sato bound for some widely acceptedchannel models and provides significant gain over ZF-DP.

    II. SYSTEM MODEL

    We assume antennas are deployed at the transmitter (base)and there are geographically dispersed receivers (mobiles),each of which has one receive antenna. The base is transmit-ting signals intended for all users (mobiles). In typical cases,

    , which means there are more users than transmit antennas.

    Since the propagation environment of each user is generally dif-ferent, the possibility of multiuser diversity [7] exists. The white

    Manuscript received January 22, 2003. The associate editor coordinating thereview of this letter and approving it for publication was Dr. J. Ritcey. Thiswork was supported by the Air Force Research Laboratory under AgreementF49620-01-1-0372 and F49620-03-1-0214, and by the National Science Foun-dation under Grant CCR-0112501.

    The authors are with the Electrical Engineering and Computer Sci-ence Department, Lehigh University, Bethlehem, PA 18015 USA (e-mail:[email protected]).

    Digital Object Identifier 10.1109/LCOMM.2003.815652

    Gaussian noise observed at each antenna is assumed to be inde-

    pendent from the noise observed at other antennas. The channelis assumed to be memoryless and quasi-static and known to boththe transmitter and receivers. The broadcast channel is charac-terized by

    (1)

    where is a complex matrix whose th entry is thepath gain from the th transmit antenna to the th receiver, andis an vector whose th entry represents the received signalat the th receiver (the receiver for user where ).We denote as the th row of which corresponds to thechannel between the base station and the th receiver. In (1),

    is the circularly symmetric complex Gaussian noisevector with identity covariance matrix and is the

    transmitted vector signal with a power constraint , i.e,, where is the signal intended for user .

    A rate vector is said to be achievable if there exists a sequence of codes with

    , where is the probability of error [8]. In thispaper, we attempt to find a simple way to achieve a sum rate

    close to the sum capacity.

    III. PROBLEM FORMATION

    Let be the covariance matrix for user .

    Since the transmitter knows , we can apply thedirty paper results in [9] and [10]. For any fixed permutation of

    denoted by , the achievable ratewith dirty paper precoding is given in [11] as

    (2)

    The dirty paper rate region is the union of all such rate

    vectors over all covariance matrices suchthat and over all permutations. This region appears to be the capacity

    region of the Gaussian broadcast channel based on the resultspresented at a recent meeting [3]. However, simple methods forachieving an arbitrary point in the region are still unavailable.The difficulty lies in handling the interference between users.The conventional approach to address the interference problem,called zero-forcing (ZF), is discussed and compared withZF-DP coding in [6]. It is shown that ZF-DP coding is superiorto ZF scheme in the sense the former can provide no less sumrate than the latter does for any channel realization.

    1089-7798/03$17.00 2003 IEEE

    http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-
  • 7/27/2019 01220272.pdf

    2/3

    http://-/?-http://-/?-
  • 7/27/2019 01220272.pdf

    3/3

    372 IEEE COMMUNICATIONS LETTERS, VOL. 7, NO. 8, AUGUST 2003

    Fig. 1. Ergodic sumrate comparison foruncorrelated channel without full rowrank. r = 1 8 ; t = 3 .

    Fig. 2. Ergodic sum rate comparison for correlated channel without full rowrank. r = 1 8 t = 3 , = = 3 and L = 2 0 .

    Since the approach employed is not optimized, is a lowerbound for . Hence (13) also holds for . Q.E.D.

    If we apply ZF-DP and randomly select and orderrows (we call this ZF-DP with random ordering) of , the re-

    sulting random variables are statistically independent and(see [6] and therein). In this case does not

    depend on at all, so we can not exploit the multiuser diversityof the system if is fixed. Therefore, even with a large numberusers in the system, the sum rate will not benefit.

    In Fig. 1, we plot the ergodic sum rate versus SNR. Since thenoise is normalized, the SNR is actually the power constraint .As we can see, when applying the greedy ZF-DP, the sum rateis pretty close to the Sato bound. We can also see a significant( 4 dB) gain of the greedy scheduling over ZF-DP with randomordering.

    We also test the greedy ZF-DP for correlated fading channels.Here we adopt the channel model used in [7]. Suppose there are

    distinct paths from a mobile to the base station. The channelis uncorrelated at the mobile side and correlated at the base. The

    paths have equally spaced angles of departure from to, where is the angle spread at the base. We also assume

    the base has a linear array of antennas with half carrier wave-length spacing. Then the correlated channel can be representedby

    (14)

    where and are the array response vectors at the re-ceiver and transmit sides respectively. is a complexmatrix whose th column is , and is a complexmatrix whose th row is . Based on the assumptions wehave made, can be modeled as a matrix with i.i.d. complexGaussian entries. The array response vector at the transmitter is

    , where is the angle ofdeparture of th path.

    We plot the ergodic sum rate vs SNR for the correlatedchannel in Fig. 2. As we can see, for the correlated channel, the

    greedy ZF-DP is still close to the Sato bound and outperformsZF-DP with random ordering significantly.

    VI. CONCLUSION

    In this letter, we proposed a greedy algorithm to exploit themultiuser diversity for the vector broadcast Gaussian channelto increase the systems instantaneous sum rate. Our algorithmexhibits good performance for some accepted channel models.We expect each user will get approximately equal rate if the sta-tistical description of each user is identical and the schedulingalgorithm is employed for a time much longer than that overwhich the channel changes. If it is not the case, some additionaltime sharing or scheduling algorithm can be employed to pro-

    vide fairness.

    REFERENCES

    [1] I. E. Telatar, Capacity of multi-antenna Gaussian channels,Eur. Trans.Telecommun., pp. 585595, Nov. 1999.

    [2] G.J. Foschiniand M.J. Gans,On thelimitsof wireless communicationsin a fading environment when using multiple antennas, Wireless Pers.Commun., vol. 6, pp. 315335, 1998.

    [3] DIMACS Workshop on Signal Processing for Wireless Transmis-sion. Piscataway, NJ: Rutgers Univ., Oct. 2002.

    [4] H. Sato, An outer bound on the capacity region of broadcast channel,IEEE Trans. Inform. Theory, vol. IT-24, pp. 374377, May 1978.

    [5] P. Viswanath and D. Tse, Sum capacity of the multiple antennaGaussian broadcast channel and uplink-downlink duality, IEEE Trans.Inform. Theory, submitted for publication.

    [6] G. Caire and S. Shamai, On the achievable throughput of a multi-an-tenna Gaussian broadcast channel, IEEE Trans. Inform. Theory, vol.49, pp. 16911706, July 2003.

    [7] W. Rhee and J. M. Cioffi, On the capacity of multiuser wireless sys-tems with multiple antennas,IEEE Trans. Inform. Theory, vol.47, Aug.2001, submitted for publication.

    [8] T. M. Cover and J. A. Thomas, Elements of Information Theory. NewYork: Wiley, 1991.

    [9] M. Costa, Writing on dirty paper, IEEE Trans. Inform. Theory, pp.439441, May 1983.

    [10] U. Erez, S. Shamai, and R. Zamir, Capacity and lattice strategies forcancelling known interference, in Proc. Int. Symp. Inform. Theory and

    Its Appl., Nov. 2000, pp. 681684.[11] S.Vishwanath, N.Jindal, andA. Goldsmith, Onthe capacity ofmultiple

    input multiple output broadcast channels, in IEEE Int. Conf. on Com-munications (ICC), vol. 3, New York, NY, Apr. 2002, pp. 14441450.

    http://-/?-http://-/?-http://-/?-http://-/?-