Professional Documents
Culture Documents
PHD Seminar (Encs8051) New Power Allocation Strategy Among Mimo Spatial Subchannels - Beam-Nulling
PHD Seminar (Encs8051) New Power Allocation Strategy Among Mimo Spatial Subchannels - Beam-Nulling
Mabruk Gheryani
List of Tables 3
List of Tables 4
1 Introduction 7
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2 Literature Survey . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Problem Statement and Objective . . . . . . . . . . . . . . . . 10
1.4 Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2 Preliminaries 12
2.1 Channel Model . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2 Power Allocation Among Spatial Subchannels . . . . . . . . . 14
2.2.1 Equal Power . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2.2 Water-Filling . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.3 Eigen-Beamforming . . . . . . . . . . . . . . . . . . . . 18
1
3 Recent Research Results 22
3.1 Beam-Nulling . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.2 Comparisons Among the Four Schemes . . . . . . . . . . . . . 26
3.2.1 At low SNR . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2.2 At medium SNR . . . . . . . . . . . . . . . . . . . . . 33
3.2.3 At high SNR . . . . . . . . . . . . . . . . . . . . . . . 34
2
List of Tables
3
List of Figures
4
Notations and Abbreviations
• XH : hermitian of X
• XT : transpose of X
• tr(X): trace of X
• det(X): determinant of X
• {x}: a set of x
• E(x): expectation of x
• BF: beamforming
• BN: beam-nulling
• MIMO: multiple-input-multiple-output
5
• SISO: single-input-single-output
• ST: space-time
• WF: water-filling
6
Chapter 1
Introduction
1.1 Motivation
Since the discovery of multiple-input-multiple-output (MIMO) channel ca-
pacity, a lot of research efforts have been put into this field [1] [2].
To combat channel quality variation and thus further improve system
performance such as power efficiency, error rate and average data rate, it has
been recognized that adaptive techniques proposed for single-input-single-
output (SISO) channel [3] [4], can also be applied to improve MIMO chan-
nel capacity. In this method, a feedback channel is utilized to provide CSI
from the receiver to the transmitter. According to the feedback of CSI, the
transmitter will adjust transmission parameters, such as power allocation,
modulation, coding rate, etc. This is conditioned by the fact that the chan-
nel keeps relatively constant until the transmitter receives the CSI and then
transmits the next data block accordingly.
The ideal scenario is that the transmitter has full knowledge of channel
7
state information (CSI) fed back from the receiver and the CSI keeps con-
stant before the transmitter sends information to the receiver. With such a
perfect CSI feedback, the original MIMO channel can be converted to multi-
ple uncoupled SISO channels via singular value decomposition (SVD) at the
transmitter and the receiver [1]. In other words, the original MIMO chan-
nel can be decomposed into several orthogonal “spatial subchannels” with
various propagation gains. To achieve better performance, various strategies
to allocate constrained power to these subchannels can be implemented de-
pending on the availability of CSI at the transmitter [6]- [8]. In this study,
we propose to develop new scheme with limited CSI feedback.
8
nels were compared in [12]. The comparison shows that for Rayleigh channels
without shadowing, space-time WF-based scheme gains little in capacity over
spatial WF-based scheme. However, for Rayleigh channels with shadowing,
space-time WF-based scheme achieves higher spectral efficiency per antenna
over spatial WF-based scheme. A WF-based scheme using imperfect CSI in
MIMO systems was studied in [13].
For the WF-based scheme, the feedback bandwidth for the full CSI grows
with respect to the number of transmit and receive antennas and the perfor-
mance is often very sensitive to channel estimation errors. To overcome these
disadvantages, various beamforming techniques for MIMO channels are also
investigated intensively. Beamforming is a linear signal processing technique
that control the complex weights of the transmit and receive antennas jointly
to optimize the signal-to-noise ratio (SNR) in one direction [10]. In other
words, beamforming can increase the sensitivity in the direction of wanted
signals but decrease the sensitivity in the direction of interfering signals.
In an adaptive beamforming scheme, complex weights of the transmit
antennas are fed back from the receiver. If only partial CSI is available at
the transmitter such as the eigenvector associated with the strongest spa-
tial subchannel, eigen-beamforming [7] is optimal in this case. The eigen-
beamforming scheme only allocates power to the strongest spatial subchan-
nel but can achieve full diversity and high signal-to-noise ratio (SNR). Also,
in practice, the eigen-beamforming scheme has to cooperate with the other
adaptive parameters to improve performance or/and data rate, such as con-
stellation and coding rate.
There are also other beamforming schemes based on various criteria.
9
For example, an optimal eigen-beamforming space-time block code (STBC)
scheme based on channel mean feedback was proposed in [7]. A MIMO sys-
tem based on transmit beamforming and adaptive modulation was proposed
in [8], where the transmit power, the signal constellation, the beamforming
direction, and the feedback strategy were considered jointly. The analysis of
MIMO beamforming systems with quantized CSI for uncorrelated Rayleigh
fading channels was provided in [9].
10
1.4 Organization
The rest of the seminar is organized as follows. In Chapter 2, preliminaries
are presented. In Chapter 3, current research results are provided. Finally
in Chapter 4, we will conclude the seminar and present the research schedule
for the remaining tasks.
11
Chapter 2
Preliminaries
12
x1 y1
hmn
xn ym
xN t yN r
i.e., E(xi ) = 0 and E|xi |2 = 1. By using the linear model, the received
signals can be written as
y = Hx + z (2.1)
z is the additive white Gaussian noise (AWGN) vector with i.i.d. symmetrical
complex Gaussian elements of zero mean and variance σz2 .
The singular-value decomposition of H can be written as
H = UΛVH (2.2)
13
assume that the rank of H is r (r ≤ Nt ). That is, the number of non-zero
singular values is r.
From (2.2), the original channel can be considered as consisting of r
uncoupled parallel subchannels. Each subchannel corresponds to a singular
value of H. In the following context, the subchannel is also referred to as
“spatial subchannel”. For instance, one spatial subchannel corresponds to
λi , ui and vi .
If the transmitter has no knowledge about the channel, the most judicious
strategy is to allocate the power to each transmit antenna equally. In this
case, the received signals can be written as
s
P
y= Hx + z (2.3)
Nt
14
The associated instantaneous channel capacity with respect to H can be
written as [1] Ã !
Nt
X P 2
Ceq = log 1 + λ (2.4)
i=1 Nt σz2 i
In the following figure, numerical results of ergodic (average) channel
capacity for 2 × 2, 3 × 3 and 4 × 4 Rayleigh flat fading channels are shown
in Figure 2.2.
Equal power
25
2x2
3x3
4x4
20 5x5
Capacity (bit/s/Hz)
15
10
0
0 5 10 15 20 25
SNR (dB)
15
2.2.2 Water-Filling
If the transmitter has full knowledge about the channel, the most judicious
strategy is to allocate the power to each spatial subchannel by water-filling
principle [1]. It allocates more power when a spatial subchannel has larger
gain (i.e. {λi }) and less when a subchannel gets worse. With V at the
transmitter and U at the receiver, the original MIMO channel is converted to
r uncoupled parallel SISO channels. The WF scheme is shown in Figure 2.3.
Binary
Info. P1x1 Ant-1
source Binary
Info.
Out
Ant-1
Power Allocation
Constellation
Mapper ~y
Detector
S/P H
V U
Prxr Ant-Nr
Ant-Nt
Nt-r 0's
Channel
Estimation
Feedback
16
and σz2 variance. Following the method of Lagrange multipliers, the optimal
Pi can be found as [1]
à !
1 σ2
Pi = max − z , 0 , ∀i = 1, 2, . . . , r (2.6)
L ln 2 λi
where L is the Lagrange multiplier. The instantaneous channel capacity is
for this spatial subchannel is
à !
Pi
Cwf,i = log 1 + 2 λ2i (2.7)
σz
Then the total channel capacity with respect to H is
r
X
Cwf = Cwf,i (2.8)
i=1
The WF scheme maximizes the channel capacity by power allocation over
spatial subchannels. Since λ1 ≥ λ2 . . . ≥ λr , we have P1 ≥ P2 . . . ≥ Pr and
Cwf,1 ≥ Cwf,2 . . . ≥ Cwf,r . In practice, if each spatial subchannel requires
the same error rate performance, the spatial subchannel with larger gain
√
(i.e. Pi λi ) can have higher rate; while if each spatial subchannel has the
same rate, the spatial with larger gain will have better performance. The
application of these two approaches will depend on the type of service. For
example, the important data needs high quality but the voice can tolerate
low quality.
In the following figure, numerical results of ergodic (average) channel
capacity for 2 × 2, 3 × 3 and 4 × 4 Rayleigh flat fading channels are shown
in Figure 2.4.
In the WF scheme, the feedback bandwidth for the perfect CSI grows
with respect to the number of transmit and receive antennas. In practice,
the bandwidth for CSI feedback is often very limited. The large feedback
bandwidth also affects the transmission efficiency.
17
Water−filling
25
2x2
3x3
4x4
20 5x5
Capacity (bit/s/Hz)
15
10
0
0 5 10 15 20 25
SNR (dB)
2.2.3 Eigen-Beamforming
18
the transmission beamformer. The Eigen-beamforming scheme is shown in
Figure 2.5.
Ant-1 Binary
Binary Info.
Info. Out
source
Ant-1
Px1
Constellation
Mapper
~
y1
H
v1 u1 Detector
Ant-Nr Channel
Estimation
Ant-Nt
Feedback
where z1 is the additive white Gaussian noise vector with i.i.d. symmetrical
complex Gaussian elements of zero mean and variance σz2 .
The eigenvector associated with the maximum singular value from the
19
receiver side, i.e., u1 , is applied as the receiver beamformer. Then we have
√
ỹ1 = uH
1 y1 = P λ1 x1 + z̃1 (2.10)
where z̃1 is Gaussian with zero mean and variance σz2 . As can be seen from
(2.10), only the spatial subchannel associated with the maximum singular
value λ1 is applied.
The associated instantaneous channel capacity with respect to H can be
written as à !
P
Cbf = log 1 + 2 λ21 (2.11)
σz
In the following figure, numerical results of ergodic (average) channel capacity
for 2×2, 3×3 and 4×4 Rayleigh flat fading channels are shown in Figure 2.6.
The eigen-beamforming scheme can save feedback bandwidth and is op-
timized in terms of SNR [10]. However, since only one spatial subchannel is
considered, this scheme suffers from loss of channel capacity [17], especially
when the number of antennas grows. Also, as can be seen from the figure,
the gap of capacity between different numbers of antennas is nearly the same
for all the SNR region.
20
Beamforming
10
2x2
9 3x3
4x4
8 5x5
7
Capacity (bit/s/Hz)
0
0 5 10 15 20 25
SNR (dB)
21
Chapter 3
3.1 Beam-Nulling
Inspired by the eigen-beamforming scheme, we will propose a new beamforming-
like scheme called “beam-nulling” (BN). This scheme uses the same feedback
bandwidth as beamforming. That is, only one eigenvector is fed back to the
transmitter. Unlike the eigen-beamforming scheme in which only the best
spatial subchannel is considered, in the beam-nulling scheme, only the worst
spatial subchannel is discarded. Hence, the loss of channel capacity as com-
pared to the optimal water-filling scheme can be reduced. The Beam-Nulling
scheme is shown in Figure 3.1.
In this scheme, the eigenvector associated with the minimum singular
value from the transmitter side, i.e., vNt , is feedback to the transmitter.
By a certain rule, a subspace orthogonal to the weakest spatial channel is
22
Binary
Info.
source x1 Ant-1
Ant-1
Binary
Info.
Constellation
~y Out
Detector
Mapper
S/P H
Φ U
xNt-1 Ant-Nr
Ant-Nt
Channel
Estimation
Feedback
23
constructed. That is, the following condition should be satisfied.
ΦH vNt = 0 (3.1)
The Nt × (Nt − 1) matrix Φ = [g1 g2 . . . gNt −1 ] spans the subspace. Note that
the rule to construct the subspace Φ should also be known to the receiver.
An example to construct the orthogonal subspace is presented as follows.
We construct an Nt × Nt matrix
A = [vNt I0 ] (3.2)
where ze is additive white Gaussian noise vector with i.i.d. symmetrical com-
plex Gaussian elements of zero mean and variance σz2 . With the condition in
24
(3.1),
B
VH Φ = (3.6)
0T
where
v1H g1 v1H g2 ... v1H gNt −1
... ..
v2H g1 ... .
B=
.. .. ... ..
(3.7)
. . .
H H
vN g
t −1 1
... . . . vN g
t −1 Nt −1
e 0 and z
where y e0 are column vectors with the first (Nr − 1) elements of y
e and
ze, respectively, and Λ0 = diag[λ1 , λ2 , . . . , λ(Nt −1) ]. From (3.8), the associated
instantaneous channel capacity with respect to H can be found as
NX
à !
t −1
P
Cbn = log 1 + λ2 (3.9)
i=1 (Nt − 1)σz2 i
25
Beam−nulling
25
2x2
3x3
4x4
20 5x5
Capacity (bit/s/Hz)
15
10
0
0 5 10 15 20 25
SNR (dB)
26
for any SNR.
Let C̄eq , C̄bf , and C̄bn denote the ergodic channel capacities of equal
power, beamforming and beam-nulling, respectively. That is,
where E() denotes expectation with respect to H. Let ρ = P/σz2 denote SNR.
Differentiating the above ergodic capacities with respect to ρ respectively, we
have
X Nt
∂ C̄eq 1
= E Nt (3.13)
∂ρ i=1 ρ + λ2
i
∂ C̄bf 1
= E (3.14)
∂ρ ρ + λ12
1
NX
t −1
∂ C̄bn 1
= E Nt −1
(3.15)
∂ρ i=1 ρ+ λ2i
The differential will also be referred to as “slope”. The second order differ-
entials are listed as follows.
∂ 2 C̄eq X Nt
1
= E − µ ¶2 (3.16)
∂ρ2
i=1 ρ+ N t
λ2 i
∂ 2 C̄bf 1
= E − ³ ´2 (3.17)
∂ρ2 ρ+ 1
λ21
27
∂ 2 C̄bn NXt −1
1
= E − µ ¶2 (3.18)
∂ρ2
i=1 ρ + Nλt −1
2
i
As can be seen, the above ergodic capacities are concave and monotonically
increasing with respect to ρ. With the fact that λ1 ≥ λ2 . . . ≥ λNt , it can
be readily checked that the slopes of ergodic capacities associate with equal
power and beam-nulling are bounded as follows.
Nt ∂ C̄eq Nt
E Nt ≥ ≥E (3.19)
ρ + λ1 ∂ρ ρ + λNNt
t
Nt − 1 ∂ C̄bn Nt − 1
E Nt −1 ≥ ≥E (3.20)
ρ + λ1 ∂ρ ρ + λN t −1
(N −1) t
28
2x2
EQ
12
WF
BF
BN
10
Capacity (bit/s/Hz)
0
0 5 10 15 20 25
SNR (dB)
29
3x3
16 EQ
WF
BF
14 BN
12
Capacity (bit/s/Hz)
10
0
0 5 10 15 20 25
SNR (dB)
(a) 0 − 25 dB
3x3 3x3
6 16
EQ EQ
WF 15 WF
BF BF
5
BN 14 BN
13
4
Capacity (bit/s/Hz)
Capacity (bit/s/Hz)
12
3 11
10
2
9
8
1
7
0 6
0 2 4 6 8 10 12 14 16 18 20 22
SNR (dB) SNR (dB)
(b) 0 − 10 dB (c) 12 − 22 dB
30
4x4
22
EQ
20 WF
BF
18 BN
16
Capacity (bit/s/Hz)
14
12
10
0
0 5 10 15 20 25
SNR (dB)
(a) 0 − 25 dB
4x4 4x4
6 20
EQ EQ
WF 19 WF
BF BF
5
BN 18 BN
17
4
Capacity (bit/s/Hz)
Capacity (bit/s/Hz)
16
3 15
14
2
13
12
1
11
0 10
0 1 2 3 4 5 6 7 8 16 17 18 19 20 21 22 23 24
SNR (dB) SNR (dB)
(b) 0 − 8 dB (c) 16 − 24 dB
31
5x5
25
EQ
WF
BF
20 BN
Capacity (bit/s/Hz)
15
10
0
0 5 10 15 20 25
SNR (dB)
(a) 0 − 25 dB
5x5 5x5
7 25
EQ EQ
WF 24.5 WF
6 BF BF
BN 24 BN
5 23.5
Capacity (bit/s/Hz)
Capacity (bit/s/Hz)
23
4
22.5
3
22
2 21.5
21
1
20.5
0 20
0 1 2 3 4 5 6 7 8 22 22.5 23 23.5 24 24.5 25 25.5 26
SNR (dB) SNR (dB)
(b) 0 − 8 dB (c) 22 − 26 dB
32
As can be seen from all these figures, the water-filling has the best capac-
ity at any SNR region. The other schemes perform differently at different
SNR regions as discussed in the following context.
At low SNR, the beamforming is the closest to the optimal water-filling, e.g.,
the SNR region below 3 dB for 3 × 3 fading channel, the SNR region below
3.2 dB for 4 × 4 fading channel, and the SNR region below 3.5 dB for 5 × 5
fading channel. At low SNR, the water-filling scheme can only allocate power
to one or two spatial subchannels. Especially, if only one spatial subchannel
can be use, the beamforming scheme is just the optimal water-filling, which
can be seen from the capacity of 2 × 2 fading channel in Figure 3.3.
33
3.2.3 At high SNR
At high SNR, the equal power scheme is the closest to the optimal water-
filling, e.g., the SNR region over 16 dB for 3 × 3 fading channel, the SNR
region over 20.5 dB for 4 × 4 fading channel, and the SNR region over 23.5
dB for 5 × 5 fading channel. As can be seen from the figures, at high SNR,
the equal power scheme will converge to the water-filling scheme.
In summary, the application of the above four schemes shall depend on the
SNR region and the availability of CSI. At medium SNR, the proposed beam-
nulling scheme can achieve larger capacity than the beamforming scheme with
the same feedback bandwidth.
34
Chapter 4
In this chapter, we will conclude the study and present our remaining tasks
together with their schedule.
4.1 Conclusions
Via singular-value decomposition, the original MIMO channel is converted
to uncorrelated spatial subchannels. Based on the concept of spatial sub-
channels, we studied various power allocation strategies for various channel
state information scenarios, such as equal power, water-filling, beamforming.
Inspired by the beamforming scheme, we proposed a novel scheme called
“beam-nulling”. Using the same feedback bandwidth as beamforming, the
new scheme exploits all spatial subchannels except the weakest one and thus
achieves significant high capacity near the optimal water-filling scheme at
medium signal-to-noise ratio. Additionally, the capacities of equal power,
beamforming and beam-nulling were compared through theoretical analysis
35
first and then numerical results of these three schemes are also compared
with the optimal water-filling scheme. The comparison showed that at low
signal-to-noise ratio, beamforming is the closest to the optimal water-filling,
at medium signal-to-noise ratio, beam-nulling is the closest to the optimal
solution, and at high signal-to-noise ratio, equal power is the closest to the
optimal solution.
36
beam-nulling scheme can be further extended, respectively. The extended
schemes will implement or discard k spatial subchannels and are referred to
as “multi-dimensional” (MD) beamforming and “multi-dimensional” beam-
nulling, respectively. The theoretical analysis and numeric results in terms
of capacity will be provided to evaluate the new extended schemes. We will
study the performance of the extended scheme. Similarly, we will study
how to cooperate with the other schemes and thus improve the performance
further.
37
Journals
[1] Mabruk Gheryani, Z. Wu and Y. Shayan, “Design of Adaptive MIMO
System Using Linear Dispersion Code,” A transaction paper has been sub-
mitted to IEEE Trans. on Wireless Communications.
[2] Mabruk Gheryani, Y. Shayan, X. Wang and Z. Wu, “Error Perfor-
mance of Linear Dispersion Codes,” A transaction letter has been submitted
to IEEE Trans. on Wireless Communications.
4.4 Schedule
The remaining tasks are scheduled in Table 4.1 and the associated Gantt
chart is also shown in Figure. 4.1.
38
2007 2008
ID Task Name
Q1 Q2 Q3 Q4 Q1 Q2
3 Wrap-ups
4 Thesis Writing
39
Bibliography
[4] A. J. Goldsmith and S.-G. Chua, “Variable rate variable power MQAM
for fading channels,” IEEE Trans. Commun., vol. 45, no. 10, pp.
12181230, Oct. 1997.
40
[9] B. Mondal and R. W. Heath, Jr., “Performance analysis of quantized
beamforming MIMO systems,” IEEE Transactions on Signal Processing
, vol. 54, no. 12, Dec. 2006.
[14] X. Zhang and B. Ottersten, “Power allocation and bit loading for spatial
multiplexing in MIMO systems,” IEEE Int. Conf.on Acoustics, Speech,
and Signal Processing, 2003. Proceedings (ICASSP ’03) vol.5 pp. 54-56,
Apr. 2003.
[16] S. Serbetli and B. Ottersten, “Power allocation and bit loading for
spatial multiplexing in MIMO systems,” IEEE Int. Conf.on Acoustics,
Speech, and Signal Processing, 2003. Proceedings (ICASSP ’03) vol.5 pp.
54-56, Apr. 2003.
41
Rayleigh MIMO channels,” IEEE Trans. Wireless Comm., vol.3, no.4,
pp. 12851294, July 2004.
42