Professional Documents
Culture Documents
I. I NTRODUCTION
Manuscript received August 13, 2008; revised April 10, 2009, June 14,
2009; May 7, 2010, and September 15, 2010; accepted September 26, 2010.
The associate editor coordinating the review of this paper and approving it
for publication was Y. J. Zhang.
The authors are with the Department of Electrical Engineering,
Stanford University, Stanford, CA 94305 (e-mail: {hjkwon, ej1130,
cioffi}@stanford.edu).
The material in this paper was presented in part at the IEEE International
Conference on Communications, Glasgow, Scotland, June 2007.
Digital Object Identifier 10.1109/TWC.2010.102810.091716
the BS. Instead, the research using partial CSI at the BS has
been actively investigated for these practical systems.
To enhance the sum-rate with the partial CSI at the BS, opportunistic beamforming has been proposed for multiple-input
single-output (MISO) broadcast channels [6]. This scheme
achieves multi user diversity by supporting the best MS
based on the signal-to-noise power ratio (SNR) feedback
from every MS. Since multi user diversity is best obtained
when the dynamic range of channel fluctuation is high,
opportunistic beamforming multiplies random coefficients at
the BSs transmit antennas to induce intentional channel
variation. This opportunistic beamforming was based on a
single beam and sequentially a new idea to use multiple beams
was proposed in the appendix of [6]. In [7], this idea has
been embodied concretely by using equal power allocation
across multiple beams in order to increase the achievable sumrate. However, this opportunistic beamforming degrades the
performance as the number of MSs decreases. In the low
number of MS regime, it is likely that the selected MSs
channels are misaligned with the randomly generated beams.
As a result, the interference would not be mitigated enough.
[9] has suggested a beam selection scheme where MSs feed
their effective channel gain, which is the inner product of
the channel vector and the beam vector, back to the BS.
Then, using the feedback from the MSs, the BS varies the
numbers of beams during training periods and during data
transmission periods while maintaining equal power allocation
across the beams. [10] also proposed the scheme adapting
the number of active beams to the number of MSs and the
number of antennas which are spatially correlated at the BS.
In [11], based on the signal-to-interference-and-noise power
ratio (SINR) or the effective channel gain feedback from the
MSs, the BS varies power allocation across multiple beams
by performing iterative waterfilling.
To be applicable into current broadband networks, opportunistic beamforming scenarios need to be adapted to both
practical constraints of low complexity and limited feedback,
while achieving high performance. This paper is motivated to
overcome these challengeable issues. When the BS changes
the number of beams or varies the power allocation across
multiple beams, the information should be updated shortly
between a training period and a data transmission period.
However, the frequent update could be burdened to broadband
networks because of high complexity to adapt into channel
variations. Thus, the low complexity process is crucial in order
for a system to be stable. In addition, the constraint of limited
feedback should be also significantly considered because the
c 2011 IEEE
1536-1276/11$25.00
KWON et al.: PREDETERMINED POWER ALLOCATION FOR OPPORTUNISTIC BEAMFORMING WITH LIMITED FEEDBACK
sum-rate highly depends on the quality of feedback information. The finite feedback rate requires each MS to quantize
its CSI, thereby preventing the BS from having the precise
channel knowledge of MSs and degrading the performance.
This issue draws an attention of solid algorithms to reduce
the amount of degradation even with limited feedback.
This paper proposes to enhance opportunistic beamforming
with the predetermined power allocation for MISO broadcast
channels. The BS receives SINR feedbacks from MSs at each
data transmission period and uses these received SINRs to
achieve multi user diversity, i.e., to select the best MS for
each beam, and to support those MSs. Contrary to the power
allocation scheme based on instantaneous CSI feedback from
MSs at each data transmission period, this paper determines
the power allocation a priori based on channel statistics and
the number of MSs, and maintain it throughout multiple
training and data transmission periods. The power allocation
across multiple beams is calculated by using the channel
ergodicity and is updated only when the number of MSs or the
average SNR is changed. Therefore, the proposed scheme is
named opportunistic beamforming with predetermined power
allocation (OBPPA).
The rest of this paper is organized as follows. Section II
describes the system model. Section III formulates the power
allocation search problem for the OBPPA scheme. Section IV
offers various power allocation search algorithms. Section V
analyzes the optimal power allocation strategy and the scaling
law of the OBPPA scheme. Computer simulations in Section
VI evaluate the sum-rate of each power allocation algorithms.
Finally, Section VII provides the conclusion and remarks.
II. S YSTEM M ODEL
This paper considers a MISO broadcast fading channel with
a BS equipped with transmit antennas for MSs, each
having a single receive antenna. The channel is assumed to
be quasi static, i.e., it is invariant during each block, which
comprises a training period and a data transmission period.
The channel input-output relationship for MS is,
= h
b +
= 1, 2, , ,
(1)
=1
85
then
=1 , where is the total power constraint
at the BS. The average SNR is defined as = / 0 . Then,
the SINR of MS for beam is expressed as
SINR, =
h b 2
=
0 + = h b 2
0
+
(2)
2
For example, let an auxiliary variable =
=1 ,
where is i.i.d. with the standard normal distribution and
is the number of . This corresponds to 1 for
in (2). The weighting factor is
satisfied with the following
two conditions: > 0, and
=1 = 1. Then, based on
[15], the distribution of is approximated as
( ) 21 1
2
1
()
, > 0.
(3)
2 =1
( 21 )
Since corresponds to , the distribution of also
corresponds to the distribution of with the normalized power
coefficient in (2).
Using the approximate distribution of , the probability
density function (PDF) of SINR, , (), can be derived by
conditioning on as
() =
() ()
(4)
0
0
(
)
( ) 21 1
( )
1
2
+
2
0
1
()
= 1
2
( 21 )
0
=1
) 1 1 (
) 21
(
0
1 2
1
1
( )
()
= 1
+
2
2
2
=1
(5)
86
=1
=1
1
log(1 + ) ()
(), (6)
sum ({ }
=1 )
subject to 0 ,
=1 .
(7)
TABLE I
E XHAUSTIVE - SEARCH ALGORITHM
Initialization:
such that
1 =
for all
)
calculate sum (
Result:
= arg max sum (
)
) = sum (
)
sum (
TABLE II
L INE - SEARCH ALGORITHM
Initialization:
= [0 0 0], [1 1 1], or [ 0 0]
)
calculate sum (
Recursion:
1 =
such that
for all
)
calculate sum (
)
= arg max sum (
if sum ( ) sum ( )
) = sum (
)
=
, sum (
KWON et al.: PREDETERMINED POWER ALLOCATION FOR OPPORTUNISTIC BEAMFORMING WITH LIMITED FEEDBACK
V. M ATHEMATICAL A NALYSIS
This section analyzes the performance of the proposed
scheme from two perspectives. Theorem 1 shows how the
sum-rate is affected when the number of MSs and the number
of BSs transmit antennas are fixed but the average SNR
increases. Theorem 2 shows how the sum-rate scales when
the average SNR is fixed but the number of MSs and the
number of BSs transmit antennas change.
A. The Optimal Power Allocation Strategy
Theorem 1 When the number of MSs and the number of
transmit antennas at the BS are fixed, the optimal power
allocation of the OBPPA scheme to maximize the average sumrate at the high SNR region is to concentrate the BSs transmit
power on a single beam.
Proof: See Appendix A.
For the fixed number of MSs and BSs transmit antennas,
using multiple beams has intuitively two opposite effects on
the performance. First, it enables the simultaneous support
of multiple MSs, thereby increasing the sum-rate. On the
other hand, it increases the interference, thereby decreasing
the sum-rate. In the high SNR region, the interference effect
is dominated on the sum-rate. Thus, the BS should use a single
beam instead of multiple beams to minimize the interference
in order to increase the sum-rate. The Theorem 1 aligns well
with this intuition.
B. The Scaling Law
Theorem 2 When the SNR is fixed, the sum-rate of the
OBPPA scheme scales as log log , where is the
100
2.5
80
Number of Searches
120
All Beams
Single Beam
ExhaustiveSearch
LineSearchFA
LineSearchFS
LineSearchZP
BisectionSearch
1.5
0.5
10
20
30
40
Number of Mobile Stations, N
60
40
20
0
0
20
40
Number of Mobile Stations, N
50
120
100
6
Number of Searches
3.5
Initialization:
that are
1 = based on their biases
sort all
, = 1, , , = 1, =
denote them as
Recursion:
) sum (
)
if sum (
= ( + )/2
) > sum (
)
elseif sum (
= ( + )/2
if 1
) > sum (
)
if sum (
=
break
Result:
, sum (
) = sum (
)
=
TABLE III
B ISECTION - SEARCH ALGORITHM
87
4
All Beams
Single Beam
ExhaustiveSearch
LineSearchFA
LineSearchFS
LineSearchZP
BisectionSearch
10
20
30
40
Number of Mobile Stations, N
80
60
40
20
50
0
0
20
40
Number of Mobile Stations, N
88
11
120
10
7
6
5
All Beams
Single Beam
ExhaustiveSearch
LineSearchFA
LineSearchFS
LineSearchZP
BisectionSearch
3
2
0
10
20
30
40
Number of Mobile Stations, N
60
40
20
50
3
2.5
2
1
0.5
0
0
20
40
Number of Mobile Stations, N
performs well for the low SNR region but poorly for the
high SNR region. On the other hand, when it starts with concentrated power allocation, the performance of linear search
algorithm is opposite. This indicates that the optimal power
allocation depends on the SNR region. This can be verified
from observing the cross-over points between the sum-rates of
uniform power allocation and concentrated power allocation.
As the SNR increases, the cross-over point moves to the right.
Therefore, for a fixed number of MSs, it is implied that the
optimal power allocation is to concentrate total power to fewer
beams as the SNR increases.
The bisection search algorithm achieves the sum-rate almost
equal to that of the exhaustive search algorithm for all SNRs
and all number of MSs. Simultaneously, as opposed to the
other power allocation search algorithms, the number of
searches is relatively low for all SNRs and all numbers of
MSs. Even though the use of bisection search algorithm is not
concretely validated in a mathematical form, this numerical result for both the sum-rate and the number of searches presents
how practically effective the bisection search algorithm is.
As a result, figures demonstrate that the proposed OBPPA
scheme significantly improves the average sum-rate over the
opportunistic beamforming scheme that uses all beams with
uniform power allocation. This is especially shown when the
SNR is high. Due to the approximation used in (6), it could
be possible that the sum-rate is affected when the number of
MSs is quite small. However, as the number of MSs increases,
this effect becomes negligible and the sum-rate curves support
the mathematical analysis in Sec. V.
B. Effect of the Finite Feedback Rates
Computer simulation compares how finite feedback rates
affect the sum-rate of both the proposed OBPPA scheme
and the beamforming scheme that varies the power allocation
based on instantaneous channel gain feedback, abbreviated as
the OBVPA scheme. In the OBPPA scheme, the BS finds
power allocation in advance by using the bisection search
algorithm and support the selected MSs based on instantaneous SINR feedback where the MSs quantize their SINRs
based on [19]. The BS uses this SINR feedback only to
achieve multi user diversity during data transmission periods.
OBPPAunlimited
OBVPAunlimited
OBPPA3bit
OBVPA3bit
OBPPA2bit
OBVPA2bit
OBPPA1bit
OBVPA1bit
1.5
Fig. 4.
20
40
60
Number of Mobile Stations, N
80
100
Effect of the feedback load on the sum rate for SNR = 0 dB.
9
8
80
Number of Searches
3.5
100
7
6
5
OBPPAunlimited
OBVPAunlimited
OBPPA3bit
OBVPA3bit
OBPPA2bit
OBVPA2bit
OBPPA1bit
OBVPA1bit
4
3
2
1
Fig. 5.
20
40
60
Number of Mobile Stations, N
80
100
Effect of the feedback load on the sum rate for SNR = 10 dB.
KWON et al.: PREDETERMINED POWER ALLOCATION FOR OPPORTUNISTIC BEAMFORMING WITH LIMITED FEEDBACK
When the average SNR is large enough, the PDF in (10) can
be expanded by using the Taylor series as
}
{
1+
(1 + )
+ ()
1
() =
0
1 + { }
.
(11)
14
12
10
8
OBPPAunlimited
OBVPAunlimited
OBPPA3bit
OBVPA3bit
OBPPA2bit
OBVPA2bit
OBPPA1bit
OBVPA1bit
Fig. 6.
89
20
40
60
Number of Mobile Stations, N
80
100
Effect of the feedback load on the sum rate for SNR = 20 dB.
VII. C ONCLUSION
This paper proposed the opportunistic beamforming with
predetermined power allocation (OBPPA) scheme to maximize the sum-rate for broadcast channels over finite feedback
channels. The OBPPA scheme predetermines power allocation
solely based on channel statistics and the number of MSs, and
maintains the power allocation throughout even multiple transmission periods. This paper proposed three power allocation
search algorithms. The bisection search algorithm performed
as well as the exhaustive search algorithm while achieving the
almost same sum-rate and requiring less complexity.
The OBPPA scheme could be extended into a hybrid
algorithm using both exhaustive search and bisection search.
In advance, analytical sum-rates can be calculated based on
the exhaustive search. Then, the bisection search obtains the
best sum-rate adapted to practical scenarios.
A PPENDIX A
P ROOF OF T HEOREM 1
The average sum-rate with multiple beams is calculated
as follows. If the transmit power is allocated across multiple
beams as [ 1 2 ]
with the weighting scalar
=
,
(8)
2
1 + = h b
1 +
2
where = h b 2 and =
= h b
are redefined, respectively. For notational brevity,
let =
h b 2 , then 2 (2)/2 and =
= .
Accordingly, the following equations are derived.
{ } =
{ } =
= ( ). (9)
SINR, =
sum,
log (1 + ) 1
1
.
0
=1
(13)
On the other hand, the average sum-rate with a single beam
is calculated as follows. When the power is concentrated to a
single beam, there is no interference. The SNR of MS for
beam 1 is
SNR,=1 =
h b1 2 = 1 .
(14)
0
The PDF and CDF of SNR,=1 are respectively as follows.
(
(
)
)
1 () = exp
, 1 () = 1 exp
.
(15)
From (13) and (16), the ratio of the sum-rate with multiple
beams to the sum-rate with a single beam is
lim
sum,
= 0.
sum,1
(17)
By conditioning on as , the SINR, follows an exponential distribution with the mean /(1 + ). Using the
conditional distribution, the PDF and CDF of the SINR,
can be written as follows:
1 + (1+)
() =
(),
0
(1+)
() = 1
().
(10)
0
A PPENDIX B
P ROOF OF T HEOREM 2
In [7], the opportunistic beamforming using equal power
allocation scales as log log , as long as does not grow
faster than log . Since equal power allocation is one of
many possible power allocation, the sum-rate of the proposed
OBPPA scheme is lower bounded by the sum-rate using the
equal power allocation.
90
R EFERENCES
[1] I. E. Telatar, Capacity of multi-antenna Gaussian channels, Bell Labs
Technical Memorandum, 1995.
[2] G. Caire and S. Shamai (Shitz), On the achievable throughput of a
multiantenna Gaussian broadcast channel, IEEE Trans. Inf. Theory, vol.
49, no. 7, pp. 1691-1706, July 2003.
[3] W. Yu and J. M. Cioffi, Sum capacity of Gaussian vector broadcast
channels, IEEE Trans. Inf. Theory, vol. 50, no. 9, pp. 1857-1892, Sep.
2004.
[4] H. Weingarten, Y. Steinberg, and S. Shamai(Shitz), The capacity region
of the Gaussian multiple-input multiple-output broadcast channel, IEEE
Trans. Inf. Theory, vol. 52, no. 9, pp. 3936-3964, Sep. 2006.
[5] T. Yoo and A. Goldsmith, On the optimality of multi-antenna broadcast
scheduling using zero-forcing beamforming, IEEE J. Sel. Areas Commun., vol. 24, no. 3, pp. 528-541, Mar. 2006.
[6] P. Viswanath, D. Tse, and R. Laroia, Opportunistic beamforming using
dumb antennas, IEEE Trans. Inf. Theory, vol. 48, no. 6, pp. 1277-1294,
Nov. 2002.
[7] M. Sharif and B. Hassibi, On the capacity of MIMO broadcast channel
with partial side information, IEEE Trans. Inf. Theory, vol. 51, no. 2,
pp. 506-522, Feb. 2005.
[8] B. Hassibi and T. L. Marzeta, Multiple antennas and isotropicallyrandom inputs: the received signal density in closed form, IEEE Trans.
Inf. Theory, vol. 48, no. 6, pp. 1473-1484, June 2002.
[9] J. Wagner, Y. C. Liang, and R. Zhang, On the balance of multiuser
diversity and spatial multiplexing gain in random beamforming, IEEE
Trans. Wireless Commun., vol. 7, no. 7, pp. 2512-2525, July 2008.
[10] E. Jorswieck, P. Svedman, and B. Ottersten, On the Performance
of TDMA and SDMA based opportunistic beamforming, IEEE Trans.
Wireless Commun., vol. 7, no. 11, pp. 4058-4063, Nov. 2008.
[11] Y. Kim, K. Song, R. Narasimhan, and J. M. Cioffi, Single user random
beamforming in Gaussian MIMO broadcast channels, in Proc. IEEE
Inter. Conf. on Comm., Seoul, Korea, vol. 4, pp. 2695-2699, May 2005.
[12] A. Paulraj, R. Nabar, and D. Gore, Introduction to Space-Time Wireless
Communications. Cambridge, 2003.
[13] W. Nelson, Applied Life Data Analysis. John Wiley and Sons, 1982.
[14] A. Castano-Martinez and F. Lopez-Blazquez, Distribution of a sum of
weighted central chi-square variables, Commun. in Statistics - Theory
and Methods, vol. 34, pp. 515-524, 2005.
[15] S. Gabler and C. Wolff, A quick and easy approximation to the
distribution of a sum of weighted chi-square variables, Statistics Hefte,
vol. 28, pp. 317-325, 1987.
[16] M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions
with Formulas, Graphs, and Mathematical Tables. New York: Dover,
1972.
[17] J. M. Cioffi, Advanced Digital Communication, EE379C Stanford University Course Reader, http://www.stanford.edu/class/ee379c.
[18] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge, 2004.
[19] E. W. Jang, C.-S. Hwang, and J. M. Cioffi, Flexible AMC tables for a
wireless network, in Proc. IEEE Inter. Conf. on Comm., Beijing, China,
pp. 3378-3382, May 2008.
[20] S. Lloyd, Least squares quantization in PCM, IEEE Trans. Inf. Theory,
vol. 28, no. 2, pp. 129-137, Mar. 1982.
[21] J. G. Proakis, Digital Communications, 4th edition. McGraw Hill, 2001.
[22] M. Sharif and B. Hassibi, Scaling laws of sum rate using time-sharing,
DPC, and beamforming for MIMO broadcast channels, in Proc. IEEE
Inter. Symp. of Info. Theory, Chicago, IL, p. 175, June 2004
Hyukjoon Kwon [S06] received his B.S. degree
in electrical engineering from Seoul National University, Korea, in 2001. He also received his M.S.
degree and Ph.D. degree all in electrical engineering
from Stanford University, Stanford, CA, in 2006 and
2010, respectively. From 2001 to 2004, he worked
for Corecess, Inc. His Ph.D. thesis was about MIMO
systems with cooperation over finite feedback rates.
He is currently with Marvell Semiconductor, Inc.