Sunteți pe pagina 1din 5

On Single-User Collaborative Random Beamforming

Jia-Hao Wu Ping-Heng Kuo Rong-Terng Juang Pang-An Ting


Information and Communications Laboratory Industrial Technology Research Institute (ITRI) Hsinchu, Taiwan

Abstract In recent years, multi-cell downlink (DL) collaborative multi-input multi-output (MIMO) transmission has drawn a lot of attentions due to its potential benefits in system throughput. However, the requirements on training and feedback may hamper its practical applications. In this paper, we consider a single-user collaborative random beamforming (Co-RBF) scheme to ease the training and feedback burden in a multi-cell environment. The proposed single-user collaborative random beamforming system, either with and without additional transmitter phase adjustments, are described and analyzed from outage probability perspective. From analysis and numerical results, collaborative random beamforming schemes can provide macro diversity gain with little feedback. This implies its feasibility for limited feedback collaborative wireless systems. Keywords- collaborative random beamforming, multi-cell, MIMO, feedback.

interference and noise ratio (SINR) through feedback links, the BS can thereby select the user with the best channel-dependent scheduling metric. As a result, the system spectral efficiency can be increased especially for the number of users is large [5], [6], [7]. It should be noted that the training signal or pilot symbols required does not grow with the number of transmit antennas [5], [8]. Motivated by the opportunistic beamforming technique introduced in [5], we propose a simple collaborative random beamforming scheme which can reduce the burdens on training procedures as well as feedback overhead. In particular, we focus on downlink macro diversity system, where multiple BSs transmit the same signal to the single target user which has been scheduled for collaborative transmission. Note that the proposed scheme can be further improved with transmitter phase adjustment mechanisms, at the expense of additional feedback. The improvement obtained from phase adjustment is characterized analytically via conditional cumulative density functions (CDF) of equivalent channel in this paper. The paper is organized follows. In section II, the system model and assumptions we made are described. By considering a simple two-BS collaboration case, collaborative random beamforming (Co-RBF) scheme with and without phase feedback and adjustments are explained and compared in section III. In section IV, numerical results are shown to verify the gain provided by proposed Co-RBF schemes. Finally, some concluding remarks are drawn in section V. Notation: ()* , ()T , () H denotes the conjugate, transpose

I.

INTRODUCTION

Multi-cell downlink (DL) collaborative beamforming has been a burgeoning research area due to its capability of enhancing the system throughput [1], [2]. With proper synchronization and coordination among multiple base stations (BSs), multi-cell collaborative beamforming can be applied to provide macro diversity for the target user. However, this scheme requires transmit channel state information (CSI) at all BSs that are involved in collaboration. When the number of collaborative BSs, and/or the total number of transmit antennas, is large, the demanded feedback and training overhead become tremendous. Thus, the deployment of multi-cell collaboration with full-CSI beamforming (FCSI-BF) might be infeasible in practice. Many works have been dedicated to deal with feedback reduction issues in MIMO systems. In [3], the authors overviewed the channel quantization schemes for reducing feedback required by close-loop MIMO systems. In [4], a technique that prevents mobile stations (MSs) feeding back channel coefficients related to BSs with unacceptable channel quality is introduced. This provides a way to make a trade-off between sum-rate performance and feedback overhead in multi-cell systems. Nevertheless, the training burden of measuring channel quality from different BSs is still quite large. To cope with the issue of training load and feedback overhead simultaneously, the authors in [5] proposed an opportunistic or random beamforming scheme combined with scheduling to achieve multiuser diversity gain with limited feedback. This technique induces fast and large channel fluctuations seen by multiple users via deliberately varying precoders in a random fashion. Since every user needs to send its received signal to

and conjugate transpose operations, respectively. is the vector norm operation. E {} is the expectation operation with respect to the random variable . All vectors or matrices are expressed in bold face.

II.

SYSTEM MODEL

Consider a multi-cell transmission system with M collaborative BSs, assuming each BS is equipped with N transmit antennas. For simplicity, we assume that each user has only one receive antenna. The received signal of user k (in the cell mk ), which has been scheduled for collaborative transmission, can be written as

yk = k , mk hT , mk s mk + k

k , m hT , m s m + wk k
m mk

(1)

978-1-4244-3574-6/10/$25.00 2010 IEEE

where h k , m C N 1 denotes the channel between user k and the BS m. All channel elements are independent and identicallydistributed (i.i.d) zero-mean complex Gaussian random variables (Rayleigh fading), denoted as CN (0,1) . s m C N 1 is the transmitted signal from BS m and meets per base station power constraint E s m

This is the case of conventional full-CSI beamforming (FCSIBF). Without loss of generality, we let k ,0 = 1 and equation (2) consequently becomes
yk = h k ,0 + k ,1 hT ,1 v k ,1 xk + wk . k

(3)

} = P for all m. As in [6],

k ,m

is a

non-negative scalar that models path loss and shadowing effects. wk ~ CN (0, N 0 ) represents complex Gaussian thermal noise at user k. We also assume channel between different links, i.e. different user and BS pairs, are mutually independent. Furthermore, each BS that involved in the collaborative transmission will transmit signal to the target user k in the form of s k = v k , m xk , m . The v k , m is the beamforming vector at
2 the BS m with v k , m = 1 and xk , m = xk is the information for the user k. We consider slow fading case where channel is almost constant within the scheduling interval. Note that the benefits of the collaborative beamforming may not be obvious in high mobility scenarios due to the incapability of tracking channel variations [5]. Furthermore, all feedback information is assumed to be ideal and error-free.

Here k ,1 denotes the large-scale fading factor which characterizes the relative path loss and shadowing effects that experienced by the signal propagated from collaborative BS 1 and from serving BS 0. In the following, we will use (3) for discussion. In addition, the user index k will be kept to facilitate justifications.
A. Collaborative Random Beamforming without phase feedback As the channel quality between user k and serving BS 0 falls below some given threshold, BS 0 would send a request to its neighboring BS 1 to ask for collaborative transmission. (b) Then BS 1 would generate a set of B random vectors {v k ,1 } as collaborative random beamforming vectors, where b=1,,B. In the training mode, BS 1 would transmit with these random beamforming vectors on separate pilots that pertain to the same channel state while BS 0 simultaneously applies FCSI-BF on these pilots. The equivalent channel observed by the b-th pilot can be expressed as hk(b ) = h k ,0 + k ,1 hT ,1 v (kb,1) . After k estimating the channel gain on these pilots, user k picks the random vector that produces maximum amplitude of equivalent channel and feedback to the collaborative BS 1. Note that the user feedback contains only preferred beamforming vector index that could be encoded with log 2 ( B ) bits.

In the data transmission mode, collaborative BS 1 would help serving BS 0 to transmit data with the beamforming vector according to user feedback. In consequence, the magnitude of equivalent channel can be written as
g = max h k ,0 + k ,1 hT ,1 v (kb,1) . k b =1,..., B
(b)

(4)

Fig. 1. Two-BS collaborative random beamforming scheme.

III.

COLLABORATIVE RANDOM BEAMFORMING

In order to shed light on the proposed collaborative random beamforming scheme and simplify discussions, from now on we will focus on two-BS collaborations case. As shown in Fig. 1, user k is served by BS 0 while BS 1 collaboratively transmits to the user k. The received signal of user k is

It should be mentioned here that the random vectors {v k ,1 } can be arbitrarily selected on an N-dimensional unit-norm sphere in (b) general. However, we consider the case where {v k ,1 } are a set of randomly selected orthonormal vectors to make the analysis more tractable. Obviously, the number of orthonromal random vectors B is less than or equal to N which represents the number of collaborative transmit antennas. When B >N, the additional N random vectors can be represented as the linear combinations of these N orthonormal vectors.
(b ) Since { v k ,1 } are a set of orthonormal vectors, it is easy to
T (b ) (b) (b) find that g = hk = h k ,0 + k ,1 h k ,1v k ,1

yk =

k ,0 hT ,0 v k ,0 + k ,1 hT ,1 v k ,1 xk + wk . k k

(2)

for all b are

Notice that the serving BS 0 needs to transmit pilot signal to user k such that antenna channel gain h k ,0 can be measured and fed back to the BS 0. Since we assume all the training and feedback procedures are ideal, BS 0 can then transmit data by * using the optimum beamforming vector v k ,0 = h k ,0 / h k ,0 .

independent Rician random variables for some given h k ,0 = s . Then the conditional cumulative density function (CDF) of g in (4) can be represented as [10]
B

FG S ( g s ) = 1 Q1

k ,1

s,

k ,1

, gs

(5)

where Q1 (a, b) = x exp


b

x2 + a2 I 0 (ax)dx is the Marcums Q 2

I 0 ( x) exp( x)

I 0 (b) ,xb, exp(b)

(11)

function and I 0 () is the modified Bessel function of the first kind.

B. Collaborative Random Beamforming with phase feedback From (4), we observe that if the random beamforming term T ( h k ,1 v kb,1) can be cophased with the FCSI-BF term h k ,0 of serving BS 0, the magnitude of equivalent channel could be greater than in (4). However, this requires user to feedback additional phase information to the adjacent collaborative BS. Precisely, user k needs to measure and feedback the phase term
b = arg ( hT ,1v (kb,1 ) ) , bk = arg max hT ,1v (kb,1 ) . k k b =1,..., B
k k k

we can argue that when s is very small, the conditional CDF of Co-RBF without phase feedback as shown in (5) will converge to (8). This implies the phase feedback gain will diminish if the FCSI-BF signal transmitted by serving BS is weak. The same argument can be made by directly observing that (4) and (7) will approximately equal when h k ,0 is small. This could happen when FCSI-BF signal comes from serving BS 0 experiences severe large scale fading. We plot the conditional CDF in (5) and (8) for s =0.5 and s=0.05 in Fig. 2 to verify this fact.
(a) 1 Conditional CDF 0.8 0.6 0.4 0.2 0

(6)

To measure the phase term in (6), the collaborative BS 1 shall transmit pilots that are not overlaid with the pilot signal from serving BS 0. After compensating the phase for the beamforming vector v (kb,1 ) at the BS 1, the magnitude of equivalent channel observed by user k in the data transmission mode is given by
k

2 3 4 Equivalent channel amplitude (b)

1 Conditional CDF 0.8 0.6 0.4 0.2 0 Collab. FCSI-BF Co-RBF with phase feedback Co-RBF w/o phase feedback 0 1 2 3 4 Equivalent channel amplitude 5 6

g = h k ,0 + max
The term

b =1,..., B

k ,1 hT ,1 v (kb,1) . k

(7)

k ,1 hT ,1v (kb,1) is Rayleigh distributed and mutually k

independent for all b. In addition, the conditional CDF given h k ,0 = s turns out to be
FG S ( g s ) = 1 exp

( g s)
k ,1

, gs.

(8)

Fig. 2. Conditional CDF in (5) and (8) where B =4, k ,1 = 1 with (a) s = 0.5 (b) s = 0.05

Following, we will use the conditional CDF to compare the collaborative beamforming gain improvement of different schemes from outage probability perspectives.

In Fig. 2, we also plot the conditional CDF of an ideal case that collaborative BS 1 has the full CSI of h k ,1 . Hereafter, we will refer it as collaborative FCSI-BF where neighboring BS can ideally utilize full CSI. In this case, received signal in (3) will become
yk = h k ,0 + k ,1 h k ,1 xk + wk .

C. Comparison From [9], it is easy to find that conditional CDF in (8) is upper bounded by (5) since Q1 (a, b) exp((b a) 2 / 2) .This implies Co-RBF scheme with phase feedback has higher probability to produce large equivalent channel magnitude than that without phase feedback.
In order to see the differences between (5) and (8), we use the following lower bound [9]:
a 2 + b2 Q1 (a, b) exp I 0 ( ab ) , b a 2

(12)

Note that both h k ,0 and h k ,1 are actually chi-distributed random variables with 2N degree of freedoms [10]. For a given h k ,0 = s , the conditional CDF of equivalent channel which is now defined as g h k ,0 + k ,1 h k ,1 is given by [10]
FG S ( g s) = 1 exp

( g s)
k ,1

(9)

1 m = 0 m!

N 1

( g s)
k ,1

, gs.

(13)

By using (9), we can find the upper bound for (5) as follows:
FG S ( g s ) 1 exp( s2 + g 2

k ,1

)I 0 (

2 sg

k ,1

,gs

(10)

Compare (10) with (8) and use the following inequality [2]

Moreover, to see the overall equivalent channel amplitude CDF, one may compute the marginal CDF FG ( g ) = ES {FG s ( g s )} for (8) and (13). Generally, it is difficult to find their close form expressions. However, we can grasp some insights from the following simple lemma.
Lemma 1: Let Y = X 1 + X 2 where X 1 0 and X 2 0 are independent random variables, then FY ( y ) FX ( y ) FX ( y ) .
1 2

Proof: By definition, we have


FY ( y ) =

FY X1 ( y x1 ) PX1 ( x1 ) dx1 =
y

FX 2 ( y x1 ) PX1 ( x1 )dx1

FX 2 ( y )

PX1 ( x1 )dx1 = FX 2 ( y ) FX1 ( y )

(14)

Inequality holds because the CDF is a monotonic increasing function. The lemma implies the CDF of resultant channel amplitude is at least multiplicatively decreased by using collaborative beamforming. In particular, we can view X 1 and X 2 as the corresponding FCSI-BF signal from serving BS and Co-RBF signal from neighboring BS. By using (14), we have
FY ( y ) FX 2 ( y ) FX1 ( y )

The diversity gain from outage probability perspective can also be improved by B-th order where B N) is the number of orthonormal random beamforming vectors. Compare (18) and (20), we can find that when B=N, the collaborative ideal FCSIBF can still have 1 / N ! improvement in outage probability. For the case that B>N, we will show that the gain improvement is insignificant in section IV.
D. Feedback and training overhead From the previous discussions, we can see that collaborative random beamforming scheme with and without phase feedback are always sub-optimum compared with the collaborative FCSI-BF. However, the training and feedback overhead can be significantly reduced especially when the total number of transmit antennas of collaborative BSs is growing.

(15)

which means the CDF ratio between Y = X 1 + X 2 and X 1 is upper bounded by the CDF of X 2 . Note that the bound in (15) is tight when y is small. In the following, we will use (15) to show that how diversity gains can be improved due to multicell collaborative transmission. Since we assume the channel is in slow fading, outage probability may be more suitable for us to evaluate the gain improvement by collaborative beamforming. The outage probability is defined as the probability that channel cannot support for a given rate R [8]. Precisely, we have
Pout Pr log 2 1 + SNR h = F (c SNR 1/ 2 )

{ (

) < R} = Pr { h < ( 2 1)
R

1/ 2

SNR 1/ 2

} (16)

The training and feedback overhead requirements are summarized in Table I. In this table, N, M, and B denote number of transmit antennas per BS, number of BS in collaboration and number of collaboratively transmitted random beamforming vectors, respectively. Additionally, QCSI , Qidx and Q ph represent bit number that required for per antenna channel gain, per vector index and per phase feedback, respectively. Note that in the system model of our concern, user always needs to feedback full CSI to its serving BS in order to perform FCSI-BF. Besides, we assume that feedback information to the neighboring BS should be forwarded by serving BS through backhaul connection. Consequently, we could restrict our attention to the total feedback amount carried through the uplink between the user and its serving BS. In the case of Co-RBF without phase feedback, it is required to allocate pilots for measuring serving BS CSI and equivalent channel quality results from collaborative random vectors. Note that pilots for random beamforming vectors from different collaborative BSs can be overlaid. The feedback information contains antenna CSI of serving BS and the single best vector index that shall be forwarded to collaborative BSs. In the scheme of Co-RBF with phase feedback, the pilots for random beamforming vectors from different collaborative BSs shall be separated to facilitate the measure of phase information and the selection of random beamforming vector index per BS. Moreover, the feedback information contains antenna CSI of serving BS, the best random vector index and the compensated phase term for each collaborative BS. TABLE I. Training and feedback overhead comparison
Training Signal (training unit) Collaborative FCSI-BF Co-RBF w/o phase feedback Co-RBF with phase feedbac Feedback Overhead (bits)

where h denotes the equivalent channel observed by user, F () is the CDF of channel magnitude, c is the constant that relates to R and SNR P / N 0 . For the case that neighboring BS has full CSI for performing collaborative FCSI-BF, we can use (13) and (15) to find
FG (c SNR 1/ 2 ) c2 SNR 1 exp 1/ 2 FS (c SNR ) k ,1 1 c2 SNR 1 m = N m ! k ,1
m

(17)

In high SNR regime, right hand side (RHS) of (17) can be approximated as
exp c2

k ,1

SNR 1

1 c2 SNR 1 m = N m ! k ,1

1 c2 N ! k ,1

SNR N

(18)

which means there is N-th order diversity improvement by using collaborative FCSI-BF at the neighboring cell. For the case that neighboring BS performs Co-RBF with phase feedback, we can use (8) and (15) to obtain
FG (c SNR ) c SNR 1 1 exp FS (c SNR 1/ 2 ) k ,1
2 1/ 2 B

NM N+B N+(M-1)B

NMQCSI NQCSI + Qidx


NQCSI + ( M 1)(Qidx + Q ph )

(19) IV. NURMERICAL RESULTS In this section, we conduct simulations to evaluate the spectral efficiency gain provided by Co-RBF with and without phase feedback. Furthermore, the macro diversity gain provided by using Co-RBF with phase feedback is also verified by un-coded bit error rate performance.

where in high SNR cases, RHS of (19) can be approximated as


1 exp c2
B

k ,1

SNR 1

c2

k ,1

SNR B

(20)

Fig. 3 shows the capacity improvement via different collaborative beamforming schemes for two-BS collaboration cases. The horizontal axis represents the relative large scale fading factor k ,1 in linear scale. We can see spectral efficiency (SE) increases with the number of random beamforming vectors B for Co-RBF schemes. In addition, the Co-RBF with phase feedback significantly outperforms Co-RBF without phase feedback. For comparison, we also plot the collaborative FCSI-BF as an ideal case in the figure. To see the macro diversity gain provided by Co-RBF, we present uncoded bit error rate (BER) performance in Fig. 4 for a multi-cell MIMO orthogonal frequency division multiplexing (MIMO-OFDM) system with two collaborative BSs transmitting the same information to a specific user. The channel model we used is ITU Pedestrian-B channel [11] with zero velocity speed. Antenna configuration is uncorrelated and the large scale fading factor k ,1 mentioned before is assumed to be 1. By observing the slope of the BER curve at high SNR region, increasing the number of random beamforming vectors can generally improve diversity gain. The BER vs. SNR curve of proposed Co-RBF with phase feedback scheme for B=4 have similar slope as in collaborative FCSI-BF case. However, when B is larger than the number of collaborative transmit antennas N, the performance gain is insignificant. Furthermore, performance loss between Co-RBF with phase feedback and collaborative FCSI-BF is reasonable. This implies Co-RBF can make a good tradeoff between training/feedback overhead and the link performance.
5 Collab FCSI-BF B=2,Co-RBF with phase feedback B=4,Co-RBF with phase feedback B=8,Co-RBF with phase feedback B=2,Co-RBF w/o phase feedback B=4,Co-RBF w/o phase feedback B=8,Co-RBF w/o phase feedback Single-BS FCSI-BF

conditional CDF of Co-RBF with phase feedback is upper bounded by Co-RBF without phase feedback and they will be approximately equal when serving BS FCSI-BF term severely degrades. From the outage probability perspective, we show the diversity gain can be improved via both collaborative FCSI-BF and Co-RBF. The macro diversity gain is also verified by numerical simulations. Furthermore, the feedback and training overhead is largely alleviated by using proposed Co-RBF. This implies its feasibility for limited feedback collaborative wireless systems.
10
-1

Uncoded BER QPSK 1/2

10

-2

10

-3

10 BER 10

-4

-5

10

-6

10

-7

10

-8

Single-BS FCSI-BF Collab FCSI-BF Co-RBF with 2 random vec. Co-RBF with 4 random vec. Co-RBF with 8 random vec. 0 1 2 3 4 SNR (dB) 5 6 7 8

Fig. 4. Un-coded bit error rate (BER) performance of different collaborative beamforming schemes. There are 4 transmit antennas per BS and only single receive antenna for the user terminal. Furthermore, we assume the channel is spatially uncorrelated and slow Rayleigh fading. The horizontal axis denotes per BS SNR.

4.5

REFERENCES
M. K. Karakayali, G. J. Foschini, and R. A. Valenzuela, Network coordination for spectrally efficient communications in cellular systems, IEEE Wireless Communications, vol. 13, no. 4, pp. 56-61, 2006. [2] B. L. Ng, J. S. Evans, S. V. Hanly, and D. Aktas, Distributed downlink Beamforming with Cooperative Base Stations, IEEE Trans. Info. Theory, vol. 54, no. 12, Dec. 2008. [3] D. J. Love, R.W. Heath, W. Santipach, and M. L. Honing, What Is the Value of Limited Feedback for MIMO Channels?, IEEE. Comm. Mag. vol. 42, issue 10, Oct. 2004. [4] A. Papadogiannis, H. J. Bang, D. Gesbert, E. Hardouin, Downlink Overhead Reduction for Multi-Cell Cooperative Processing enabled Wireless Networks, IEEE PIMRC, 2008. [5] P. Viswanath, D. N. C. Tse, and R. Laroia, Opportunistic Beamforming Using Dumb Antennas, IEEE Trans. Info. Theory, vol. 48, no. 6, pp.1277-1294, June 2002. [6] N. Sharma and L. H. Ozarow, A Study of Opportunism for MultipleAntenna Systems, IEEE Trans. Info. Theory, vol. 51, no. 5, May 2005. [7] Y. Kim, K-B Song, R. Narasimhan and J. M. Cioffi, Single User Random Beamforming in Gaussian MIMO Broadcast Channels, IEEE ICC, 2005. [8] D. Tse and P. Viswanath, Fundamentals of Wireless Communication, Cambridge University Press, 2005. [9] G. E. Corazza and G. Ferrari, New Bounds for the Marcum QFunction, IEEE Trans. Info. Theory, vol. 48, no. 11, Nov. 2002. [10] J. G. Proakis, Digital Communications, 4-th edition, McGraw-Hill, 2001. [11] ITU-R M.1225, Guidelines for evaluation of radio transmission technologies for IMT-2000, 1997. [1]

Spectral efficiency(bps/Hz)

3.5

2.5

0.5

1 1.5 2 2.5 Adjacent BS Large scale fading factor (Linear)

3.5

Fig. 3. Capacity improvement by using collaborative random beamforming (Co-RBF) with and without phase feedback. There are 4 transmit antennas per BS.

V.

CONCLUSIONS

The single-user collaborative random beamforming schemes with and without phase feedback are proposed in this paper. To ease discussions, we consider a simple two-BS collaboration scenario where the serving BS utilizes full CSI for FCSI-BF and the neighboring BS collaboratively performs random beamforming via a set of orthonormal beamforming vectors. The conditional CDF of proposed Co-RBF schemes and ideal collaborative FCSI-BF scheme are given. The

S-ar putea să vă placă și