Download as pdf or txt
Download as pdf or txt
You are on page 1of 61

PhD Proposal

Design of Adaptive MIMO Wireless


Communication Systems
Mabruk Gheryani
Supervisor: Dr. Yousef R. Shayan
August 1, 2007
Abstract
Design of Adaptive MIMO Wireless
Communication Systems
Mabruk Gheryani, PhD
Concordia University, 2007
Since the discovery of MIMO channel capacity, a lot of research works have
been done in this eld. Space-time (ST) codes are the most promising technique for
MIMO systems. However, in most applications, the channel state information (CSI)
is assumed to be known to the receiver but unknown to the transmitter. To further
improve the system performance, the transmitter shall adapt the transmission rate
based on the level of CSI fed back from the receiver. Our overall goal in this study
is to develop adaptive MIMO schemes that can adapt the transmission rate based on
the level of CSI and meanwhile satisfy the given quality of service (QoS).
First, a tight upper bound of error probability at high signal-to-noise ratio
is derived for full-rate linear dispersion code and the bound is veried by simulation
results. The theoretical results demonstrate the relationship between the error proba-
bility, the constellation size and the space-time symbol rate, which will be a guideline
for adaptation.
Secondly over a Rayleigh fading channel, the probability density function of
signal-to-interference-noise ratio of a MIMO transceiver using full-rate linear disper-
sion code and linear minimum-mean-square-error receiver is derived. With these
i
theoretical results as a guideline, we study the design of adaptive systems with dis-
crete selection modes. An adaptive algorithm for the selection-mode adaptation is
proposed. Based on the proposed algorithm, two adaptation techniques using con-
stellation and space-time symbol rate are presented, respectively. To improve the
average transmission rate, a new adaptation design is developed, which is based on
joint constellation and space-time symbol rate adaptation. Simulation results and
theoretical analysis are provided to verify our new design.
As future work, new beamforming techniques and adaptation strategy will
be further investigated. Additionally, overall adaptation design for a concatenated
system will be studied.
ii
Contents
Abstract i
List of Tables v
List of Figures vi
Notations and Abbreviations vii
Chapter 1 Introduction 1
1.1 Research Background . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Literature Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Organization of the proposal . . . . . . . . . . . . . . . . . . . . . . . 5
Chapter 2 Performance Analysis of linear dispersion codes 7
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 Performance Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.4 Simulation Results and Discussions . . . . . . . . . . . . . . . . . . . 14
2.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Chapter 3 New Adaptive MIMO System using full rate linear disper-
sion code with Selection Modes 17
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.2 Adaptive Transceiver . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.2.1 The Adaptive Transmitter . . . . . . . . . . . . . . . . . . . . 18
3.2.2 The Statistics of SINR with the MMSE Receiver . . . . . . . . 18
3.3 Design of Adaptive Transceiver . . . . . . . . . . . . . . . . . . . . . 26
3.3.1 Adaptation Using Variable Constellations . . . . . . . . . . . 27
iii
3.3.2 Adaptation Using Variable ST Symbol Rate . . . . . . . . . . 33
3.4 Joint Adaptation Technique . . . . . . . . . . . . . . . . . . . . . . . 39
3.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
Chapter 4 Conclusions and Remaining Works 45
4.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.2 Remaining Works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
Bibliography 48
iv
List of Tables
3.1 adaptive constellation with ST symbol rate =1, 2, 3 and 4 . . . . . . 32
3.2 adaptive ST symbol rate when constellation size=BPSK, QPSK, 8PSK
and 16QAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.3 Joint Adaptation of ST symbol rate and constellation size . . . . . . 41
4.1 Schedule for the remaining tasks. . . . . . . . . . . . . . . . . . . . . 47
v
List of Figures
2.1 Adaptive Transmission System Model . . . . . . . . . . . . . . . . . . 8
2.2 Simulation result and the associated upper bound at high SNR. . . . 14
3.1 Comparison between the theoretical PDF of SINR and Monte Carlo
Simulation when N
r
= N
t
= 2 at P/
2
z
= 20dB . . . . . . . . . . . . . 24
3.2 Comparison between the theoretical PDF of SINR and Monte Carlo
Simulation when N
r
= N
t
= 4 at P/
2
z
= 20dB . . . . . . . . . . . . . 25
3.3 Adaptive Constellation Size when ST symbol rate=1 . . . . . . . . . 28
3.4 Adaptive Constellation Size when ST symbol rate=2 . . . . . . . . . 29
3.5 Adaptive Constellation Size when ST symbol rate=3 . . . . . . . . . 30
3.6 Adaptive Constellation Size when ST symbol rate=4 . . . . . . . . . 31
3.7 Adaptive ST symbol rate when Constellation Size is BPSK . . . . . . 34
3.8 Adaptive ST symbol rate when Constellation Size is QPSK . . . . . . 35
3.9 Adaptive ST symbol rate when Constellation Size is 8PSK . . . . . . 36
3.10 Adaptive ST symbol rate when Constellation Size is 16QAM. . . . . . 37
3.11 Joint Adaptation of ST symbol rate and constellation size . . . . . . 40
3.12 Fading Regain for adaptive constellation size . . . . . . . . . . . . . . 43
3.13 Average spectral eciency comparison for the three adaptive schemes. 44
4.1 Gantt chart for the remaining tasks. . . . . . . . . . . . . . . . . . . . 47
vi
Notations and Abbreviations
X: upper bold letter for matrix
x: lower bold letter for column vector
X
H
: hermitian of X
X
T
: transpose of X
: Kronecker product
diag[x]: a diagonal matrix with x on its main diagonal
tr(X): trace of X
det(X): determinant of X
vec(X): a column vector formed by stacking the column vectors of X in order
: Euclidean norm
{x}: a set of x
P(x): probability of event x
E(x): expectation of x
AWGN: additive white Gaussian noise
BER: bit error rate
BLAST: Bell-labs layered space-time
CSI: channel state information
FDFR: Full Diversity Full Rate
vii
LDC: linear dispersion code
MIMO: multiple-input-multiple-output
ML: maximum-likelihood
MMSE: minimum mean square error
PDF: probability density function
QoS: quality of service
SINR: signal-to-interference-noise ratio
SISO: single-input-single-output
SNR: signal-to-noise ratio
ST: space-time
STBC: space-time block code
STTC: space-time trellis code
ST Turbo TC: space-time turbo trellis code
SVD: channels via single value decomposition
TCM: trellis-coded modulation
T-BLAST: thread BLAST
V-BLAST: vertical BLAST
WF: water-lling
i.i.d.: independently identically distributed
viii
Chapter 1
Introduction
1.1 Research Background
In the future wireless communications, one major challenge is to design an exible
system that can adapt the transmission rate to the channel condition while the target
quality of service (QoS) is also satised [1].
Recent signicant advances in wireless communications is the so-called multiple-
input-multiple-output (MIMO) technology [2][3]. It makes use of multiple transmit
and receive antennas to improve the data rate and performance over fading channels.
Since the advent of MIMO technology, tremendous research and development eorts
in academia and industry have been invested, and this investment is ever increasing.
To date, MIMO technology has been widely used in modern wireless communica-
tion systems, such as WLAN and 3G cellular systems, and is recognized as the most
important enabling technology for future wireless communication systems.
Information theory have demonstrated that a signicant gain in capacity over
fading channels can be obtained in MIMO systems [2][3]. Furthermore, the use of
multiple antennas increases the diversity to combat fading [4][5][6].
To realize the promised theoretical capacity and diversity of MIMO wireless
channels, we propose to develop new adaptive MIMO wireless communication schemes
with maximal data rate for various level of CSI feedback while target QoS is satised.
1.2 Literature Survey
After the discovery of capacity of MIMO systems, a lot of research eorts have been
put into this eld[2][3]. To exploit the signicant capacity and diversity, space-time
1
(ST) codes are the most promising technique for MIMO systems. In most appli-
cations, the channel state information (CSI) [7] is assumed to be known or can be
estimated at the receiver but unknown to the transmitter.
To combat channel quality variation and thus further improve system perfor-
mance such as power eciency, error rate and average data rate, adaptive transmission
can be applied. In this method, a feedback channel is utilized to provide CSI from the
receiver to the transmitter. According the feedback of CSI, the transmitter will ad-
just transmission parameters, such as power allocation, modulation, coding rate, etc.
This is conditioned by the fact that the channel keeps relatively constant before the
transmitter receives the CSI and then transmits next data block accordingly. That
is, the channel is slow. Adaptation algorithms can be classied into two categories:
One approach is designed with a power constraint to optimize the throughput while
maintaining a target bit error rate (BER); the other approach aims to optimize the
performances, i.e., BER, with a constant throughput. In our study, we focus on the
rst category.
Historically, the potential of adaptation transmission was recognized by Cavers
[8], but probably due to hardware constraints, lack of accurate channel estimation and
unavailability of feedback, it received little interest at that time. Recently the advent
of high-speed devices and the capability of transceiver reconguration contribute to
a renewed interest in adaptive techniques [9][10].
The idea of adaptation transmission can also be applied in MIMO systems.
With a perfect CSI feedback [7], the original MIMO channel is converted to multiple
uncoupled single-input-single-output (SISO) channels via single value decomposition
(SVD). To optimize the system throughput, the so-called water-lling (WF) principle
is performed on the multiple SISO channels. Numerous schemes have been proposed
based on this optimal solutions. For example, over time-invariant MIMO channels,
it is known that the optimal performance (ergodic capacity) is attained by power
water-lling across channel eigenvalues with the total power constraint [2]. Also, for
time-varying MIMO channels, the optimal performance is obtained through power
water-lling over both space and time domains with the average power constraint [11].
The space-time WF-based scheme and the spatial WF-based scheme for MIMO fading
channels 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 eciency per antenna over spatial WF-based
scheme. A WF-based scheme using imperfect CSI in MIMO systems was studied in
2
[13]. In [14], a so-called QoS-based WF was proposed to solve the power allocation
problem for a given xed bit error rate threshold. It is worthy of noting that at high
signal-to-noise ratio (SNR), WF-based schemes can not benet any more [ref?].
In the WF-based scheme, the feedback bandwidth for the perfect CSI grows
with respect to the number of transmit and receive antennas and the performance
is very sensitive to the channel estimation errors. To overcome these disadvantages,
various beamforming techniques are also investigated intensively. In these schemes,
complex weights fed back from the receiver are applied to transmit antennas. For ex-
ample, an optimal eigen-beamforming STBC scheme based on channel mean feedback
was proposed in[15]. A MIMO system based on transmit beamforming and adaptive
modulation was proposed in [16], 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 [17].
The above schemes often need near-perfect CSI feedback for adaptation cal-
culation. In practice, feedback channel has narrow bandwidth, the feedback CSI is
often not prompt and the CSI estimation is not accurate. All these factors make the
CSI at the transmitter imperfect. In this case, adaptive schemes with selection modes
are often more preferable, which require only partial CSI at the transmitter.
For the MIMO communication system, the structures of most existing ST cod-
ing designs mainly fall into two categories, either trellis structure or linear structure.
ST codes with trellis structure, such as the space-time trellis codes (STTCs) [5] and
space-time turbo trellis codes (ST Turbo TCs) [18][19][20], can achieve full diversity
and large coding rate. However, their computational complexity grows exponentially
with respect to the number of states and transmit antennas, they are often designed
by hand and the trellis structure is not exible for rate adaptation. ST codes with
linear structure can also be referred to as linear dispersion codes (LDCs), such as
STBC[6][21], Bell-labs layered space-time (BLAST) architectures [22][23][24][25][26].
The LDC allows a variety of decoders including simple linear techniques, higher data
rate and exibility. However, the error performance of these high-rate LDCs is often
less satisfactory.
To achieve better performance, the idea of concatenated coding schemes is
often applied to MIMO communications recently. By combining two or more relatively
simple constituent codes, a concatenated coding scheme can achieve large coding gain
with a moderately complex decoding. Additionally, such a coding structure also allows
exible and simple design. In MIMO communications, a ST code is concatenated with
3
a conventional outer code serially. Such a concatenated coding system often possesses
many advantages: On the one hand, the outer code can provide large coding gain
and time diversity; on the other hand, the inner space-time code provides guaranteed
spatial multiplexing and diversity [27]. Together, they enable a variety of design
targets in performance, bandwidth eciency, complexity, and tradeos among them
[28]. Although any structure of ST codes can be a potential candidate for the inner
ST modulation, a particular desirable choice is linear dispersion codes (LDCs) instead
of ST codes with trellis structure. This is because the LDC is simple, exible with
relatively low complexity.
For such a concatenated MIMO system [29], several discrete parameters are
available for adaptation, such as constellation size (i.e., bit-loading), active transmit
antennas and coding rate of the outer code. For example, adaptive modulation with
antenna selection combined with STBC was discussed in [30][31][32][33] [34]. The
advantage of this scheme using STBC is to simplify the design of an adaptive mod-
ulation system. However, this scheme is not exible for dierent rates which is the
key requirement in the future wireless communications.
To achieve these requirements, LDCs are applied in our system. The ST symbol
rate of the LDC together with the other parameters can be adjusted for exible rate
and throughput improvement.
1.3 Objectives
The overall goal of this study is to develop adaptive MIMO schemes that can improve
the average transmission rate according to the level of CSI and meanwhile satisfy the
given quality of service (QoS). To accomplish our goal, the following tasks will be
carried out.
1. Derive Upper Bound Of Linear Dispersion Code
In general, it is dicult to nd the codeword error probability. However, the
pair-wise error probability (PEP) can be used in the codeword design. That is,
the Euclidean distance between the received signals associated with any pair of
codewords shall be maximized by minimizing the PEP between any pair of code-
words. In this task, the upper bounds of error probability for high signal-to-noise
ratio (SNR) and low SNR are obtained, respectively. The bounds demonstrate
the relationship between error probability and, space-time symbol rate and the
constellation size. The relationship will be a guideline for adaptation. The task
4
has been accomplished and will be described in Chapter 2.
2. New adaptive MIMO system with selection modes
In this task, Over a Rayleigh fading channel, the probability density function
of signal-to-interference-noise of a MIMO transceiver using full-rate linear dis-
persion code and linear minimum-mean-square-error receiver is derived. With
these theoretical results as a guideline, new adaptation design with selection
modes is studied. New parameter, i.e., ST symbol rate is introduced to improve
transmission rate together with constellation for the given QoS. This task has
been accomplished and will be described in Chapter 3.
3. New beamforming technique
When more CSI is available at the transmitter, beamforming techniques can be
applied. In this task, new beamforming techniques will be studied for MIMO
adaptation.
4. New adaptation strategy
With the adaptation techniques, more ecient new strategy shall be studied
for MIMO adaptation.
5. Overall adaptation for concatenated system
In this task, we will study the design for overall MIMO concatenated system.
Especially, when turbo principle is applied to the concatenated system, design
for overall adaptive system will be studied.
1.4 Organization of the proposal
The rest of the proposal is organized as follows.
In Chapter 2, For full-rate linear dispersion code, tight upper bounds of the
pair-wise error probability at high SNR and low SNR are obtained and veried. The
theoretical results show the relationship between the error probability and the con-
stellation size and the space-time symbol rate, which will provide guidelines for adap-
tation.
In Chapter 3, rst we study the probability density function of signal-to-
interference-noise for a MIMO transceiver using full-rate linear dispersion code and
linear minimum-mean-square-error receiver over a Rayleigh fading channel. With the
5
statistics as a guideline, we study design of the adaptive transceiver with selection
modes. Two adaptation techniques using constellation and space-time symbol rate
are studied, respectively. To improve the average transmission rate, a new adaptation
design is proposed, in which constellation and space-time symbol rate are considered
jointly. Theoretical analysis and simulation results are provided to verify the new
design.
Finally, in Chapter 4, we will conclude our proposal and present the research
schedule for the remaining tasks.
6
Chapter 2
Performance Analysis of linear
dispersion codes
2.1 Introduction
The rich and mature knowledge on the conventional outer codes lets us focus on
the adaptive design of the inner ST modulator. Although any existing space-time
code can be a potential candidate for the inner space-time modulation, a particu-
lar desirable choice is linear dispersion codes (LDCs). This is because it subsumes
many existing block codes as its special cases, allows suboptimal linear receivers with
greatly reduced complexity, and provides exible rate-versus-performance tradeo
[24]. Hence, in our research, we focus on the LD space-time modulator in the adap-
tive MIMO transmission system. Below, we rst introduce our system model and
then provide the performance analysis for the inner ST modulator. The analytical
results will provide design guideline for the adaptive system with selection modes.
2.2 System Model
In this study, a block fading channel model is assumed where the channel keeps con-
stant in one modulation block but may change from block to block. That is, the
channel is not necessarily constant within a coding frame which often consists of a
large number of modulation blocks. Furthermore, the channel is assumed to be a
Rayleigh at fading channel with N
t
transmit and N
r
receive antennas. Lets denote
the complex gain from transmit antenna n to receiver antenna m by h
mn
and collect
them to form an N
r
N
t
channel matrix H = [h
mn
], known perfectly to the receiver
7
but unknown to the transmitter. The entries in H are assumed to be independently
identically distributed (i.i.d.) symmetrical complex Gaussian random variables with
zero mean and unit variance. The Adaptive MIMO Transmission System is shown in
Figure 2.1.
M
L
Ant-Nt
Constellation
Mapper
S/P
M1
1
Ant-1
ML
Or
MMSE
1
Nt
Nt
Ant-1
Ant-Nr
De-Mapper
Binary
Info.
source
Binary
Info.
Out
2
Q
L /T
Feedback
Selection
Mode
Figure 2.1: Adaptive Transmission System Model
8
In this system, the information bits are rst mapped into symbols. After that,
the symbol stream is parsed into blocks of length L. The symbol vector associ-
ated with one modulation block is denoted by x = [x
1
, x
2
, . . . , x
L
]
T
with x
i

{
m
|m = 0, 1, . . . , 2
Q
1, Q 1}, i.e., a complex constellation of size 2
Q
, such as
2
Q
-QAM). The average symbol energy is assumed to be 1, i.e.,
1
2
Q
2
Q
1

m=0
|
m
|
2
= 1.
Each block of symbols will be mapped by the ST modulator to a dispersion matrix
of size N
t
T and then transmitted over the N
t
transmit antennas over T channel
uses.An N
t
T codeword matrix is constructed as [24].
X =
L

i=1
M
i
x
i
+
L

i=1
N
i
x

i
(2.1)
Where M
i
, N
i
are the dispersion matrices for the i-th symbol. For simplicity, the
following model will be considered in this study, i.e.,
X =
L

i=1
M
i
x
i
(2.2)
where M
i
is dened by its L N
t
T dispersion matrices M
i
= [m
i1
, m
i2
, . . . , m
iT
].
The so-obtained results can be extended to the model in (2.1). With a constellation
of size 2
Q
, the data rate of the space-time modulator is R
m
= Q L/T bits per
channel use. Hence, one can adjust ST symbol rate L/T, constellation size Q, to
meet dierent requirements on data rate and performance. Since the ST modulation
is linear, suboptimal linear receivers can be used for demodulation [24][25]. It can also
be observed that the space-time mapping schemes used in the existing layered space-
time architectures, e.g., [22][23], are LD modulation. Hence the proposed adaptive
MIMO Transmission System with LD ST modulation subsumes existing layered space-
time schemes as special cases. At the receiver, the received signals associated with
one modulation block can be written as
Y =

P
N
t
HX+Z =

P
N
t
H
L

i=1
M
i
x
i
+Z (2.3)
where Y is a complex matrix of size Nr T whose (m, n)-th entry is the received
signal at receive antenna m and time instant n, Z is the additive white Gaussian
noise (AWGN) matrix with i.i.d. symmetrical complex Gaussian elements of zero
mean and variance
2
z
, and P is the average energy per channel use at each receive
antenna. It is often desirable to write the matrix input-output relationship in (2.3)
9
in an equivalent vector notation. Let vec() be the operator that forms a column
vector by stacking the columns of a matrix and dene y = vec(Y), z = vec(Z), and
m
i
= vec(M
i
), then (2.3) can be rewritten as
y =

P
N
t
HGx +z =

P
N
t

Hx +z (2.4)
where H = I
T
Hwith as the Kronecker product operator and G = [m
1
, m
2
, . . . , m
L
]
will be referred to as the modulation matrix. Since the average energy of the signal
per channel use at a receive antenna is assume to be P, we have tr(GG
H
) = N
t
T.
Denoting

h
i
= Hm
i
as the i-th column vector of

H, the above equation can also be
written as
y =

P
N
t
L

i=1

h
i
x
i
+z (2.5)
2.3 Performance Analysis
After we dene our system model as in (2.5). we will start to nd the upper bound on
the probability of error. Basically, most of the LD codes have follow (2.5) with dier-
ent design for modulation matrix.The average pairwise error probability conditioning
on

H is give by
P
e
(pairwise/

H) = P(d
E
i
< d
E
i+1
/s
i+1
sent) (2.6)
where d
E
i
is the Euclidean distance related to the signal vector s
i
and equal to
d
E
i
= x

SNR
N
t

Hs
i
(2.7)
We can start from [24] to nd the upper bound. In [24]the average pairwise
error probability is obtained by choosing x as Gaussian in (2.5). Then, the error
result is averaged between an independent x and x by applying a union bound to
this average pairwise probability of error which yields an upper bound on probability
of error of a signal constellation.
P
e
2
RT1
E
_
det

I +

H
H

1
_
(2.8)
where E[ ]is the expectation over the channel matrix

H and =
SNR
2Nt
. Now,
10
we need to nd the distribution of the modied

H. The modied

H depends on H
and G.In the Full Diversity Full Rate(FDFR) design [28][26], the entries of G should
satisfy GG
H
= I
N
t
T
to preserve the channel capacity. So that, the entries of the
modied

H is still CN(1, 0),
Note also that
det
_
I
NrT
+

H
H
_
= det
_
I
L
+

H
H

H
_
Lets dene
W =
_
_
_

H
H
N
r
T < L

H
H
H N
r
T L
(2.9)
and
n = max(N
r
T, L), m = min(N
t
T, L), L = min(N
t
T, N
r
T)
W is an m m random non-negative denite matrix and thus has real non-
negative eigenvalues.The distribution law of W is called the Wishart distribution
with parameters m and n. The unorder eigenvalues have the joint density function
[2] [36][37][38]
P

(
1
....
m
) = (m!K
m,n
)
1
exp
_

i
_

nm
i

i<j
(
i

j
)
2
(2.10)
where K
m,n
is a normalization factor, where
i
are the eigenvalues of W. We can
write (2.8) in terms of the eigenvalues of W as.
P
e
2
RT1
m

i=1
_

r
i
=0
[1 +
i
]
1
f

(
i
)d
i
(2.11)
where f

i
(
i
) denotes the pdf of the i
th
nonzero eigenvalue of the W [2]. If we let
f

() denote the PDF of any unordered


i
for i = 1, ....., m, then (2.11) leads to
P
e
2
RT1
_

0
[1 + ]
m
f

()d (2.12)
The probability distribution function f

() of an unordered eigenvalue of a Wishart


distribution matrix was given in [2] and can be written as
f

() =
1
m
m

i=1

i
()
2

nm
exp() (2.13)
11
where

k+1
() =
_
k!
(k + n m)!
_1
2
L
nm
k
() k = 0, ...m1
where L
nm
k
() is the associated Laguere Polynomial of order k [39] dened as
L
nm
k
() =
k

c=0
(1)
c
_
_
k + n m
k c
_
_

c
c!
Equation (2.13) can be written as,
f

() =
1
m
m1

k=0
k!
(k + n m)!
[L
nm
k
()]
2
(2.14)
Substituting (2.10) into (2.11) we have that
P
e

2
RT1
m
_

0
(1 + )
m

m1

k=0
k!
(k + n m)!
[L
nm
k
()]
2
d (2.15)
Lets dene
K1(k) =
k!
(k + n m)!
(k +n

)
2
2k
k!
K2(i) =
(2i)!(2k 2i)!
i![(k i)!]
2
(k + n

)
K3(d) =
(2)
d
d!
_
_
2k + 2n 2m
2k d
_
_
(2.16)
where n

= n m + 1. Then, we can write (2.12) as


P
e

2
RT1
m
m1

k=0
K1(k)
k

i=0
K2(i)
2k

d=0
K3(d) I (2.17)
where I is the intgration part dene as
I =
_

0
(1 + )
m
exp()
nm+d
d
I = ( )
m
_

0
(u + )
m
exp()
nm+d
d (2.18)
12
where u =
1
. To compute the above integration,we make use of the result in [39].
_

0
x

(x + u)

exp (x) dx
=

2
1
u
+
2
( + 1) exp
_
1
2u
_
W
p,s
_

1
_
where W( ) is the Whittaker function dened by Gradshteyn and Ryzhik
[39] then
I = ( )
m
( )
dn+2m
2
exp
_

2
_

(d +n

)W
p,s
_

1
_
(2.19)
where p = (
dn
2
) and s = (
d+n2m+1
2
)
Substituting (2.19) into (2.17) we have,
P
e
2
RT1
( )
m
exp
_

2
_
m1

k=0
K1(k)
k

i=0
K2(i)
2k

d=0
K3(d)(d + n

) ( )
2mnd
2
W
p,s
_

1
_
(2.20)
The above upper bound probability of error does not show the diversity ad-
vantage which is the important measure of code performance. We instead examine
the performance at High SNR.
Then, (2.17) can be further upper bounded by
P
e

2
RT1
m
( )
m

m1

k=0
K1(k)
k

i=0
K2(i)
2k

d=0
K3(d)
_

0
exp()
n2m+d
d (2.21)
Lets dene the integration by I
I =
_

0
exp()
n2m+d
d
I = (d + n 2m) (2.22)
Substituting I into (2.21) we get the upper bound at high SNR
13
P
e
K
k,i,d
2
QL1
m
_
SNR
2N
t
_
m
(2.23)
where
K
k,i,d
=
m1

k=0
K1(k)
k

i=0
K2(i)
2k

d=0
K3(d)(n 2m + d)
2.4 Simulation Results and Discussions
In this section, we verify our derivation by simulation. In the simulation, N
t
= N
r
=
T = 2, BPSK and QPSK constellations were assumed. At the receiver, the optimal
maximum-likelihood (ML) detector was applied.
5 10 15 20 25
10
9
10
8
10
7
10
6
10
5
10
4
10
3
10
2
SNR(dB)
P
e
High SNR Upper bound for BPSK&QPSK LD 2TX & 2RX with ML receiver


SIM BPSK
SIM QPSK
UP BOUND BPSK
UP BOUND QPSK
Figure 2.2: Simulation result and the associated upper bound at high SNR.
In Figure 2.2, the simulation results and the associated theoretical upper
bounds derived at high SNR are compared. As can be seen from the Figures, both up-
per bounds t well with simulation results. The simulation results and the associated
theoretical bounds have the same slopes.
14
Some observations related to the above performance analysis will be useful
for future study. The rst observation is that from (2.23), the P
e
(Q, L, SNR) is
monotonically increasing with respect to ST Symbol rate and the constellation size.
This means that if we increase number of layers, the P
e
will increase. We have the
same observation for constellation size.
The second observation is described as follows. The above upper bound shows
that the diversity order of the FDFR MIMO scheme is m = min(N
r
T, L), where
L = min(N
t
T, N
r
T). For example, if N
r
N
t
and T N
t
then the diversity order
will be N
t
T. When we set N
t
= N
r
and T = N
t
then the diversity order will be N
t
N
r
.
To satisfy full rate and full diversity, the minimum value for T is N
t
and
L = N
t
T must be satised. For given N
t
, If T is increase, L will be increased so
that the symbols interference from other layers will be increased and the associated
diversity will be reduced. By these facts, the optimum value for T is N
t
. we introduce
sample design that perform full rate full diversity as following
1. Choose the number of symbols in each block to satisfy
L min(N
t
, N
r
)T
, and choose T = N
t
.
2. Step one grantee full rate, to achieve full diversity, choose the coecients of
modulations matrices to maximize the entries of the diagonal of

H
H
L


H
L
.
3. To perverse the capacity, make sure these coecients satises these conditions.
i
trace(GG
H
) = N
t
T
ii
trace(M
H
k
M
j
) =
_
_
_
1 when k = j
0 otherwise
For example, in the system with 2 2 if N
t
= N
r
= T, then we have four Modulation
matrix which are
M
1
=
_
_
a
1
a
2
a
3
a
4
_
_
M
2
=
_
_
a
5
a
6
a
7
a
8
_
_
M
3
=
_
_
a
9
a
10
a
11
a
12
_
_
M
4
=
_
_
a
13
a
14
a
15
a
16
_
_
15
Then,we can write the modulation matrix G as follows
G = [m
1
, m
2
, m
3
, m
4
]
G =
_

_
a
1
a
5
a
9
a
13
a
2
a
6
a
10
a
14
a
3
a
7
a
11
a
15
a
4
a
8
a
12
a
16
_

_
where a
1
, a
2
, . . . , a
16
are constants , their values depend on the design of LDC. In our
design, if we choose
a
1
= 1, a
2
= 0, a
3
= 0, a
4
= 1, a
5
= 1, a
6
= 0, a
7
= 0, a
8
= 1
a
9
= 0, a
10
, 1, a
11
= 1, a
12
= 0, a
13
= 0, a
14
= 1, a
15
= 1, a
16
= 0
then

H =
_

_
h
11
h
11
h
12
h
12
h
21
h
21
h
22
h
22
h
12
h
12
h
11
h
11
h
22
h
22
h
21
h
21
_

_
the diagonal entries of

H
H
L


H
L
are equals to

N
t
m=1

N
r
n=1
|h
mn
|
2
which prove our design achieve full diversity.
As mentioned before, the error probability of an FDFR LDC is a function of ST
symbol rate and constellation size. With this fact, we can maximize the transmission
rate by adjusting the ST symbol rate and constellation size jointly while maintaining
the target QoS as described in the next chapter.
2.5 Conclusions
In this chapter, our adaptive MIMO system model with linear dispersion code is
introduced. For full-rate linear dispersion code, a tight upper bound of the pair-wise
error probability at high signal-to-noise ratio is derived and veried by simulation
results. The theoretical results show the relationship between the error probability
and the constellation size and the space-time symbol rate. The relationship will
provide guidelines for adaptation.
16
Chapter 3
New Adaptive MIMO System
using full rate linear dispersion
code with Selection Modes
3.1 Introduction
For the reasons mentioned in Chapter 1, we study the adaptive system with discrete
selection modes in this chapter. With the upper bound of pair-wise error probability
obtained in the last chapter as a guideline, we design an adaptive MIMO system with
discrete selection modes. The associated MIMO transceiver uses an LDC as the ST
modulator and the minimum mean square error (MMSE) detector at the receiver for
simplicity. As mentioned before, dierent from existing adaptive systems, the new
design adds a new adaptive parameter referred to as ST symbol rate. As can be seen
from the following discussions, by adding this new parameter, the overall throughput
of the system is increased.
3.2 Adaptive Transceiver
In this section, we will introduce our adaptive MIMO transceiver, which uses a LDC
as the ST modulator and the MMSE receiver.
17
3.2.1 The Adaptive Transmitter
In our design example, the ST modulation is LDC with dispersion matrices given by
M
(k1)Nt+i
= diag[f
k
]P
(i1)
(3.1)
for k = 1, 2, . . . , N
t
and i = 1, 2, . . . , N
t
, P is the permutation matrix of size N
t
and
given by
P =
_
_
0
1(N
t
1)
1
I
N
t
1
0
(Nt1)1
_
_
(3.2)
where f
k
denotes the k-th column vector of F. F = [f
mn
] is a Fast Fourier Transform
(FFT) matrix and f
mn
is calculated by
f
mn
=
1

N
t
exp(2j(m1)(n 1)/N
t
) (3.3)
3.2.2 The Statistics of SINR with the MMSE Receiver
As a suboptimal receiver for the LDC, linear minimum-mean-square-error detector is
more attractive due to its simplicity and good performance [40][41].
However, the performance analysis in this case is still decient. Most of the
related works address only the V-BLAST [22][42] scheme, a special case of the full-rate
LDC. For example, the case of two transmit antennas was analyzed in [43] and the
distribution of the angle between two complex Gaussian vectors was presented. The
layer-wise signal to interference plus noise ratio (SINR) distribution for V-BLAST
with successive interference cancellation at the receiver was provided in [44].
The main goal of this section is to study the statistics of SINR for full-
rate LDCs [24]-[28] using linear minimum-mean-square-error(MMSE) receiver over
a Rayleigh fading channel, which will benet future studies, such as error-rate prob-
ability.
We consider a general system model as shown in Section 2.2. In our study,
N
r
N
t
is assumed. For simplicity, we choose T equal to N
t
and L equal to N
t
T.
Equation (2.4) can also be written as
y =

P
N
t

h
i
x
i
+

P
N
t

j=i

h
j
x
j
+z (3.4)
In the sequel, the i-th column of

H, denoted as

h
i
, will be referred to as the
signature signal of symbol x
i
.
18
Without loss of generality, we consider the estimation of one symbol, say
x
i
. Collect the rest of the symbols into a column vector x
I
and denote

H
I
=
[h
1
, .., h
i1
, h
i+1
, ..., h
L
] as the matrix obtained by removing the i-th column from

H.
A linear MMSE detector is applied and the corresponding output is given by
x
i
= w
H
i
y = x
i
+ z
i
. (3.5)
where z
i
is the noise term of zero mean. The corresponding w
i
can be found as
w
i
=
_

h
i

h
H
i
+R
I
_
1

h
i

h
H
i
_

h
i

h
H
i
+R
I
_
1

h
i
(3.6)
where R
I
=

H
I

H
H
I
+
2
z
I. Note that the scaling factor
1

h
H
i
(

h
i

h
H
i
+R
I)
1

h
i
in the coef-
cient vector of the MMSE estimator w
i
is added to ensure an unbiased estimation
as indicated by (3.5). The variance of the noise term z
i
can be found from (3.5) and
(3.6) as

2
i
= w
H
i
R
I
w
i
(3.7)
Substituting the coecient vector for the MMSE estimator in (3.6) into (3.7), the
variance can be written as

2
i
=
1

h
H
i
R
1
I

h
i
(3.8)
Then, the SINR of MMSE associated with x
i
is 1/
2
i
.

i
=
1

2
i
=
_
P
N
t
_

h
H
i
R
1
I

h
i
(3.9)
In our system model, all the symbols has the same SINR, i.e.,
1
=
2
= .........
L
=
By using singular value decomposition (SVD), (3.9) can be written as
=
_
P
N
t
_

h
H
i
U
1
U
H

h
i
(3.10)
where U
H
is an N
2
t
1N
2
t
1 unitary matrix and the matrix is (N
2
t
1)(N
2
t
1)
with nonnegative numbers on the diagonal and zeros o the diagonal. Lets dene
h

= U
H

h
i
19
which is the transformed propagation vector with components h

l
, l = 1, ...........N
r
N
t
.
Equation (3.10) can be written as
=
_
P
N
t
_
N
r
N
t

l=1
|h

l
|
2

l
(3.11)
with

l
=
_
_
_

l
=
1

2
z
(
l
+ 1) l = 1, ......., r L 1

2
z
l = r + 1, ........., N
r
N
t
(3.12)
where =
P
Nt
2
z
and r is the rank of

H
I

H
H
I
which is less than N
2
t
1.
The vector h

has the same statistics as the original vector



h
i
[2]. For analytical
purpose, we can replace |h

l
|
2
by |

h
il
|
2
. Now, we can write (3.11) as
=
r

i=1

|

h
il
|
2
(
l
+ 1)
+
NtNr

r+1
|

h
il
|
2
(3.13)
The probability density function (PDF) of can be found using the moment
generating function (MGF) as follows [36]. First, we nd the conditional (on the
eigenvalues) MGF of as
M
/
(s) = [M

( s)]
NrNtr
r

l=1
M

_
s
(
l
+ 1)
_
(3.14)
the f

i
(
i
) denotes the PDF of the i
th
nonzero eigenvalue of the

H
I

H
H
I
. If we let f

()
denote the PDF of any unordered
i
for i = 1, ....., m,, then (3.14) can be written as
M
/
(s) = [M

( s)]
N
r
N
t
r
_
M

_
s
( + 1)
__
r
(3.15)
Further,
M
/
(s) =
1
(1 s)
N
r
N
t
r

1
_
1
s
(1+ )
_
r
(3.16)
we can nd the probability density function (PDF) of conditionally on by using
inverse Laplace transform for (3.16) as [46]
P
/
() =
( )
N
r
N
t
(N
r
N
t
)
()
N
r
N
t
1
exp(


)
(1 + )
r
1
F
1
(N
r
N
t
r, N
r
N
t
, ) exp() (3.17)
20
where
1
F
1
(., ., .) is Kummers conuent hypergeometric function [39] and dened as
1
F
1
(a, b, x) =

n
(a)
n
(c)
n
x
n
n!
where ()
n
=
(+n)
()
.
Lets dene
K

=
( )
NrNt
(N
r
N
t
)
()
NrNt1
exp(


)
Then equation (3.17) can be written as
P
/
() = K

n
(N
r
N
t
r)
n
(r)
n

n
n!

n
(1 + )
r
exp() (3.18)
Now, we can nd the probability density function (PDF) of as follows.
P

() =
_

0
P
/
()f

() d (3.19)
f

() was given in [2] and can be written as


f

() =
1
r
r

i=1

i
()
2

NrNtr
exp() (3.20)
where

k+1
() =
_
k!
(k + N
r
N
t
r)!
_1
2
L
N
r
N
t
r
k
()
k = 0, ...r 1
where L
NrNtr
k
() is the associated Laguere Polynomial of order k [39]. Equation
(3.20) can be written as
f

() =
1
r
r1

k=0
k!
(k +N
r
N
t
r)!
[L
NrNtr
k
()]
2
(3.21)
Lets dene
K1(k) =
k!
(k + N
r
N
t
r)!
(k + n

)
2
2k
k!
K2(i) =
(2i)!(2k 2i)!
i![(k i)!]
2
(k + n

)
21
K3(d) =
(2)
d
d!
_
_
2k + 2N
r
N
t
2r
2k d
_
_
where n

= N
r
N
t
r + 1. Then we can write (3.19) as
P

() =
K

r
r1

k=0
K1(k)
k

i=0
K2(i)
2k

d=0
K3(d)
_

0
(1 + )
r

NrNtr+d
1
F
1
(N
r
N
t
r, N
r
N
t
, ) exp()d (3.22)
The term (1 + )
r
can be written as
(1 + )
r
=
r
r

v=0
_
_
r
v
_
_

vr

v
Then equation (3.22) can be written as
P

() =
K


r
r

n
K(n)
r

v=0
K(v)K1(k)
k

i=0
K2(i)
2k

d=0
K3(d)
r1

k=0
_

0

N
r
N
t
r+d+v+n
exp() d (3.23)
with
K(n) =
(N
r
N
t
r)
n
(r)
n
n!
and
K(v) =
_
_
r
v
_
_

vr
The general form of the integration of (3.23) can be found in [39]
_

0
x

exp(x)dx = !
1
where
= N
r
N
t
r + d + v + n
22
Then (3.23) can be written as
P

() =
K


r
r

n
K(n)
r

v=0
K(v)
r1

k=0
K1(k)
k

i=0
K2(i)
2k

d=0
K3(d)
N
r
N
t
+rdvn1
(N
r
N
t
r + d + v + n)! (3.24)
Further,
P

() =
K

r
r
NrNt+1
r1

k=0
K1(k)
k

i=0
K2(i)
2k

d=0
K3(d)

d
r

v=0
K(v)
v

n
K(n)(N
r
N
t
r + d +v +n)! (3.25)
Lets dene
K(v, d) = (N
r
N
t
r + d + v)!
(N
r
N
t
r +d + v + 1)(d + v r + 1)
(d + v + 1)(N
r
N
t
+d + v + 1)
`
K(v)
and
`
K(v) =
_
_
r
v
_
_
Then (3.25) can be written as
P

() =
K

rN
r
N
t
1
r
r1

k=0
K1(k)
k

i=0
K2(i)
2k

d=0
K3(d)
d
r

v=0
K(v, d) (3.26)
This is the PDF of SINR for our system over Rayleigh fading channels.
We verify our derivation by simulation. In the simulation, N
t
= N
r
= T = 2
and N
t
= N
r
= T = 4 were assumed. In Figure 3.1 and Figure 3.2, the theoretical
PDFs of the SINR in (3.26) and results by Monte Carlo simulation were compared
for 2 2 and 4 4 channels, respectively at P/
2
z
= 20dB. Simulation results match
to the analytical result very well.
23
0 10 20 30 40 50 60 70 80 90
0
5
10
15
x 10
4


Monte Carlo Simulation
Theoretical PDF
2
10
Figure 3.1: Comparison between the theoretical PDF of SINR and Monte Carlo
Simulation when N
r
= N
t
= 2 at P/
2
z
= 20dB
24
0 10 20 30 40 50 60
0
1
2
3
4
5
6
7
x 10
4


Monte Carlo Simulation
Theoretical PDF
2
10
Figure 3.2: Comparison between the theoretical PDF of SINR and Monte Carlo
Simulation when N
r
= N
t
= 4 at P/
2
z
= 20dB
25
3.3 Design of Adaptive Transceiver
The general idea of adaptive technique with selection modes is to choose from a set of
available adaptive transmission rates. Based on some certain strategy, the transmitter
is informed with necessary information that will be used to increase or decrease the
transmission rate depending on the channel condition (i.e., CSI). For adaptive system
with the selection modes, the SNR will be considered as a proper metric. In this case,
the adaptive algorithm is proposed as follows.
1. Find the SNR at the receiver;
2. Find the BERs associated with the SNR for each mode for the BER curves
obtained from experiments.
3. Select a proper transmission mode with the maximum rate while maintaining
the given target BER.
We can describe the selection of transmission modes as follows.

opt
= arg max

n
(BER

n
() BER
target
)] (3.27)
where
n
, 1, 2, . . . , N is a transmission mode, BER
n
() is the BER of the adaptive
scheme using the transmission mode
n
for given SNR .
opt
is the optimal trans-
mission mode with the maximum data rate for given SNR while the target BER
is satised. Let R

n
denote the rate of the transmission mode
n
. Without loss of
generality, we assume R

1
< R

2
< . . . < R

N
.
Below, we consider the average transmission rate with the above algorithm.
Let

n
denote the minimum SNR for the given transmission mode
n
and target
BER, i.e.,

n
= arg min

[BER
n
() BER
target
)] (3.28)
Then, the average transmission rate is

R =
N

n=1
R

n
_

n+1

n
p

()d (3.29)
where p

() is the probability density function (pdf) of the SNR and

N+1
= .
The solution for the above optimization problem can be solved using Lagrange multi-
pliers. However, due to the structure of both the objective function and the inequality
constraint an analytical solution is extremely dicult to nd as can be seen from SINR
26
distribution given in the last section. Therefore, we will nd the optimal SNR regions
by using simulation results.
In our simulations, we assume the same system model as Section 2.2 with
N
t
= N
r
= 4. First, we will start our adaptation using the constellation size while
each set has xed ST symbol rate. Secondly, we will change the ST symbol rate
with xed constellation size. Finally, we will adapt these two parameter jointly to
maximize the throughput while maintaining the target BER which is equal 10
3
in
our design example.
3.3.1 Adaptation Using Variable Constellations
Although the system design for continuous-rate scenario provide intuitive and use-
ful guidelines[16], the associated constellation mapper requires high implementation
complexity. In practice, using discrete constellation is preferable. That is, Q takes
only integer number Q = 1, 2, 3, ..... For a given adaptive system, we can adjust the
constellation size to maximize the transmission rate at the same time satisfying the
target BER. The above algorithm is applied to the case. Although any constellation
can be used, we only use BPSK (Q = 1), QPSK (Q = 2), 8PSK (Q = 3) and 16QAM
(Q = 4) as examples. Simulation results are shown in Figure 3.3 - Figure 3.6, where
each Figure has its own ST symbol rate.
27
4 2 0 2 4 6 8 10 12 14 16
10
4
10
3
10
2
10
1
10
0
SNR(dB)
B
E
R


8PSK1layer
QPSK1layer
BPSK1layer
16QAM1lyaer
Figure 3.3: Adaptive Constellation Size when ST symbol rate=1
28
2 0 2 4 6 8 10 12 14 16 18 20
10
4
10
3
10
2
10
1
10
0
SNR(dB)
B
E
R


BPSK2layer
QPSK2layer
8PSK2layer
16QAM2layer
Figure 3.4: Adaptive Constellation Size when ST symbol rate=2
29
5 0 5 10 15 20 25 30
10
4
10
3
10
2
10
1
10
0
SNR(dB)
B
E
R


BPSK3Layer
QPSK3layer
8PSK3layer
16QAM3layer
Figure 3.5: Adaptive Constellation Size when ST symbol rate=3
30
0 5 10 15 20 25 30 35
10
3
10
2
10
1
10
0
SNR(dB)
B
E
R


BPSK4layer
QPSK4layer
8PSK4layer
16QAM4layer
Figure 3.6: Adaptive Constellation Size when ST symbol rate=4
We can nd the SNR region for each constellation by curve-tting technique or
simply by reading the SNR points corresponding to a target BER. The BER versus
SNR relationship can be approximated by the following expression.
BER = a
R
m
,Q
exp(b
R
m
,Q
SNR) (3.30)
where R
m
and Q are the ST symbol rate and the constellation size respectively, and
a
R
m
,Q
and b
R
m
,Q
are constant which can be found by curve-tting technique. We
summarize our simulation results in Table 3.1. Note that,
L
T
Q
in Table 3.1 is the
minimum SNR for the given transmission mode.
31
Table 3.1: adaptive constellation with ST symbol rate =1, 2, 3 and 4
MODE Constellation size ST symbol rate Total Rate bits/ch use
L
T
Q
1 BPSK 1 1
1
1
=-0.6309
2 QPSK 1 2
1
2
=-0.1893
3 8PSK 1 3
1
3
=3.384
4 16QAM 1 4
1
4
=11.7479
MODE Constellation size ST symbol rate Total Rate bits/ch use
2
Q
1 BPSK 2 2
2
1
=0.8385
2 QPSK 2 4
2
2
=1.4058
3 8PSK 2 6
2
3
=5.3886
4 16QAM 2 8
2
4
=15.4452
MODE Constellation size ST symbol rate Total Rate bits/ch use
3
Q
1 BPSK 3 3
3
1
=3.1014
2 QPSK 3 6
3
2
=4.4833
3 8PSK 3 9
3
3
=8.9696
4 16QAM 3 12
3
4
=26.5898
MODE Constellation size ST symbol rate Total Rate bits/ch use
4
Q
1 BPSK 4 4
4
1
=8.1509
2 QPSK 4 8
4
2
=14.2812
3 8PSK 4 12
4
3
=24.2533
4 16QAM 4 16
4
4
=30.8208
32
3.3.2 Adaptation Using Variable ST Symbol Rate
In other existing schemes, only the orthogonal designs, such as Alamouti scheme, are
applied as the ST modulation. In this case, the most convenient adaptive parameter
is the constellation size. For our adaptive scheme, the application of LDC makes
another adaptive parameter available, i.e., ST symbol rate. In this section, we x the
constellation size but adjust the ST symbol rate for adaptation. Additionally, one
advantage of using ST symbol rate is that it is easier to change ST symbol rate than
constellation size for adaptation. The same algorithm can be applied to ST symbol
rate.
Note that, this system with 4 transmit antennas can have 16 choices of ST
symbol rates, i.e., (
1
4

16
4
). For convenience and less complexity, we use 4
choices, i.e.,
L
T
= 1, 2, 3, 4. In the following context, the integer of
L
T
is referred as
layer. The simulation results are shown in Figure 3.7 - Figure 3.10.
33
2 0 2 4 6 8 10
10
4
10
3
10
2
10
1
10
0
SNR(dB)
B
E
R
BER BPSK 4X4


3Layer
4layer
1layer
2layer
Figure 3.7: Adaptive ST symbol rate when Constellation Size is BPSK
34
2 0 2 4 6 8 10 12 14 16 18 20
10
6
10
5
10
4
10
3
10
2
10
1
10
0
SNR(dB)
B
E
R
BER 4PSK 4X4


2layer
3layer
4layer
1layer
Figure 3.8: Adaptive ST symbol rate when Constellation Size is QPSK
35
0 5 10 15 20 25
10
4
10
3
10
2
10
1
10
0
SNR(dB)
B
E
R
BER 8PSK 4x4


1layer
2layer
3layer
4layer
Figure 3.9: Adaptive ST symbol rate when Constellation Size is 8PSK
36
0 5 10 15 20 25 30 35
10
4
10
3
10
2
10
1
10
0
SNR(dB)
B
E
R
BER 16QAM 4X4


3layer
1lyaer
2layer
4layer
Figure 3.10: Adaptive ST symbol rate when Constellation Size is 16QAM.
37
We summarize these results in Table 3.2.
where
L
T
Q
In Table 3.2 is the minimum SNR for the given transmission mode.
Table 3.2: adaptive ST symbol rate when constellation size=BPSK, QPSK, 8PSK
and 16QAM
MODE Constellation size ST symbol rate Total Rate bits/ch use
L
T
Q
1 BPSK 1 1
1
1
=-0.6309
2 BPSK 2 2
2
1
=0.8385
3 BPSK 3 3
3
1
=3.1014
4 BPSK 4 4
4
1
=8.1509
MODE Constellation size ST symbol rate Total Rate bits/ch use
i
2
1 QPSK 1 2
1
2
=-0.1893
2 QPSK 2 4
2
2
=1.4058
3 QPSK 3 6
3
2
=4.4833
4 QPSK 4 8
4
2
=14.2812
MODE Constellation size ST symbol rate Total Rate bits/ch use
i
3
1 8PSK 1 3
1
3
=3.384
2 8PSK 2 6
2
3
=5.3886
3 8PSK 3 9
3
3
=8.9696
4 8PSK 4 12
4
3
=24.2533
MODE Constellation size ST symbol rate Total Rate bits/ch use
i
4
1 16QAM 1 4
1
4
=11.7479
2 16QAM 2 8
2
4
=15.4452
3 16QAM 3 12
3
4
=26.5898
4 16QAM 4 16
4
4
=30.8208
38
3.4 Joint Adaptation Technique
As shown in the above two techniques, adaptation by adjusting either constellation
size or ST symbol rate can increase the average transmission rate satisfying the given
QoS as compared to non-adaptive MIMO schemes. However, we can further improve
the average transmission rate by applying a joint adaptation. The idea of the joint
adaptation is to choose the best combination of constellation size and ST symbol
rate. For the given target-BER, a scheme with the joint adaptation can improve the
average transmission rate signicantly as compared to that with only one of the above
two techniques.
Our adaptation in this case works as follows.
max
N

n=1

n
()(Q
n
, (
L
T
)
n
) for BER
Target
where (Q
n
, (
L
T
)
n
) is the specic rate associated with a specic fading region and
where

n
() is the probability of
n
in the region
n
. We will used the joint adap-
tation technique by choosing the best one from the available curves, which has the
maximum throughput. The simulation results are shown in Figure 3.11.
39
0 5 10 15 20 25 30 35
10
6
10
5
10
4
10
3
10
2
10
1
10
0
SNR(dB)
B
E
R
BER BPSK,QPSK,8PSK,16QAM and 1,2,3,4 ST symbol rate for uncoded LDC with MMSE IC


1layer
1layer
2layer
3layer
4layer
2layer
4layer
2layer
4layer
3layer
2layer
1bitBPSK1
2bitQPSK1
4bitQPSK2
6bitQPSK3
6bit8PSK2
9bit8PSK3
8bitQPSK4
8bit16QAM2
12bit8PSK4
16bit16QAM4
Figure 3.11: Joint Adaptation of ST symbol rate and constellation size
40
We note from Figure 3.11 that we can reduce the gap between the selection
modes further by adding more choices of the transmission rates. We conclude the
result in Table 3.3, where
L
T
Q
is the minimum SNR for the given transmission mode.
Table 3.3: Joint Adaptation of ST symbol rate and constellation size
MODE Constellation size ST symbol rate Total Rate bits/ch use
L
T
Q
1 BPSK 1 1
1
1
=-0.6309
2 QPSK 1 2
2
1
=-0.1893
3 QPSK 2 4
2
2
=1.4058
4 QPSK 3 6
2
3
=4.4833
5 8PSK 3 9
3
3
=8.9696
6 8PSK 4 12
3
4
=24.2533
7 16QAM 4 16
4
4
=30.8208
From Figure 3.11, we observe the following observations:
If the ST symbol rate is reduced, the slope of the associated BER curve becomes
steeper, which suggests a larger diversity;
If the constellation size is reduced, the BER curve will go down but with the
similar slope, which suggests the diversity keeps the same but the coding gain
is enlarged.
There exists a tradeo between diversity gain and multiplexing gain [27]. However,
this tradeo can not provide insight for the adaptive system with discrete constel-
lations. From the above observations, we nd that we can improve data rate by
using the two adaptive parameters jointly. Specically, in some cases, we can adjust
constellation size to improve rate and performance; which in the other cases, we will
adjust ST symbol rate, i.e., multiplexing gain, for adaptation. To proceed, we have
the following proposition.
41
Proposition 1: The average data rate in the adaptive system with selection
modes can be improved by adding more possible selection modes providing higher
data rate than the corresponding original mode at the same SNR region.
Proof: Let us dene the SNR regions of our adaptive system using one set of
selection modes as follows.

1

1
< <
2
associated with R
1

2

2
< <
3
associated with R
2
.
.

i

i1
< <
i
associated with R
i
as shown in Figure 3.12 a. If we add more possible selection modes, the SNR regions
will be changed as follows.

1
< <
X
i
1
associated with R

2

X
i
1
< <
X
i
2
associated with R

i

X
i
j1
< <
X
i
j
associated with R

i
as shown in Figure 3.12 b.
The total average rate for original scheme can be written as

R =

i
R
i
_

i
2

i
1
P
n
()d
The total average rate when for the scheme with more selection modes can be
written as

A =

i
(R
i
_

i
x

i
1

n
()d + R

i
_

i
2

i
x

n
()d)
It is obvious that

A >

R
42
Ri
R1
R2 R3

1
2

i
Ri
R1
R2 R3

2
3

x2

x1

xi
Figure (a) adaptive constellation size only
Figure (b) adaptive constellation size and ST Symbol Rate
Figure 3.12: Fading Regain for adaptive constellation size
In Figure 3.13, we compare the average spectral eciency (ASE) for the three
adaptation techniques. As can be seen from Figure 3.13, The ASE of the adaptive
scheme with the joint adaptation outperforms the other two schemes signicantly
from 0dB to 25dB. At high SNR (larger than 25 dB), three schemes have the same
performance. As can been seen from the Figure, if there are more available rate
choices, the ASE can be improved further.
43
5 0 5 10 15 20 25 30 35 40
0
2
4
6
8
10
12
14
16
18
20
SNR[dB]
A
v
e
r
a
g
e

S
p
e
c
t
r
a
l

E
f
f
i
c
i
e
n
c
y

b
p
s
/
c
h
a
n
n
e
l

u
s
e


ASE Joint Adaption (Q,L/T)
ASE changing constellation size(BPSK,QPSK,8PSK,16QAM with L/T=4
ASE changing ST symbol rate(L/T=1,2,3,4) with 16QAM
Figure 3.13: Average spectral eciency comparison for the three adaptive schemes.
3.5 Conclusions
In this chapter, rst we studied the statistics of signal-to-interference-noise for a
MIMO transceiver using full-rate linear dispersion code and linear minimum-mean-
square-error receiver over a Rayleigh fading channel. The associated probability den-
sity function of the signal-to-interference-noise is derived, which will benet the future
study, such as error-rate probability. With the statistics as a guideline, we study the
design of the adaptive transceiver with selection modes. An adaptive algorithm for
the selection-mode adaptation is proposed. Based on the proposed algorithm, two
adaptation techniques using constellation and space-time symbol rate are studied,
respectively. To improve the average transmission rate, a new adaptation design is
proposed. In the new design, constellation and space-time symbol rate are considered
jointly. Theoretical analysis and simulation results are provided to verify our new
design.
44
Chapter 4
Conclusions and Remaining Works
In this chapter, we will conclude the proposal and present our remaining tasks with
their scheduled.
4.1 Conclusions
We have studied MIMO adaptive systems based on partial channel state information.
This proposal gives an introduction of MIMO adaptation with selection modes and
the research results will provide background for further research.
In the proposal, our adaptive MIMO system model with linear dispersion code
is introduced. A tight upper bound of the pair-wise error probability at high signal-
to-noise ratio is derived and veried by simulation results.
Statistics of signal-to-interference-noise ratio has been studied for the adaptive
system with full-rate linear dispersion code and linear minimum-mean-square-error
receiver. The associated probability density function of the signal-to-interference-
noise is derived, which will benet the future study, such as error-rate probability.
With these theoretical results as guidelines, an adaptive algorithm for the
selection-mode adaptation is proposed. Based on the proposed algorithm, we have
introduced three novel adaptation techniques for the adaptive system with full-rate
linear dispersion code and linear minimum-mean-square-error receiver.
The rst technique is an extension of commonly used adaptation technique for
SISO systems. We have identied the signal-to-noise ratio regions for which specic
constellations can be applied. This technique is called as adaptive constellation. The
second new technique for adaptation of full-rate linear dispersion code is promising.
The technique is called as adaptive space-time symbol rate. Finally, to further im-
prove the average transmission rate, we introduced a novel adaptive procedure which
45
takes advantages of both adaptation techniques. This technique is called as joint
adaptation.
Theoretical analysis and simulation results are provided to verify our new de-
sign. The contributions of this study are summarized as follows.
A tight upper bound of the pair-wise error probability at high signal-to-noise
ratio is derived for full-rate linear dispersion with maximum likelihood receiver
and veried by simulation results.
Probability density function of the signal-to-interference-noise ratio is derived
for full-rate linear dispersion with linear minimum-mean-square-error receiver,
which will benet the future study
The development of a novel adaptive full-rate linear dispersion with linear
minimum-mean-square-error receiver. An adaptive algorithm for the selection-
mode adaptation is proposed. Based on the proposed algorithm, two adaptation
techniques using constellation and space-time symbol rate are studied, respec-
tively. To improve the average transmission rate, a new adaptation design is
proposed. In the new design, constellation and space-time symbol rate are
considered jointly
In the following section, the future works will be discussed.
4.2 Remaining Works
Five tasks are identied as listed in Section 1.3. The rst two tasked has been
accomplished and the related research results have been presented in Chapter 2 and
3.
In the rst remaining task, we will study the beamforming technique. To per-
form beamforming, when perfect channel information is available at the transmitter,
one needs to perform singular value decomposition on the channel matrix H. This
is also called eigen-beamforming since it uses eigenvectors to nd the linear beam-
former that optimizes the performance. Inspired by existing beamforming schemes,
we will propose a new beamforming technique called minimum eigenvector beam-
forming or beamforming-nulling (BN). With this technique, the feedback bandwidth
for channel state information can be reduced and the loss of channel capacity as
compared to the optimal water-lling scheme can also be minimized.
46
Table 4.1: Schedule for the remaining tasks.
ID Task Name Schedule
1. New Beamforming Techniques 2007/3 - 2007/10
2. New Adaptation Strategy 2007/4 - 2007/7
3. Overall Concatenated System 2007/8 - 2007/12
4. Wrap-ups 2007/11 - 2007/12
5. Thesis Writing 2008/1 - 2008/4
The second remaining task is to propose the strategy for the new adaptive
system. The basic task of any adaptive strategy is how to inform channel state
information to the transmitter coordinating the receiver together and thus adapt to
the channel variations. Related to this issue, we will study dierent strategies that
can be used in adaptive MIMO wireless communication systems.
The third remaining task is to design the overall adaptation for a concate-
nated system. In this task, we will study the adaptation in concatenated MIMO
transmission systems. Instead of exhaustive error-rate simulation, the technique of
EXIT Chart will be used for joint adaptation between the coding rate, constellation
and ST symbol rate.
The remaining three tasks are scheduled in Table 4.1 and the associated Gantt
chart is also shown in Figure. 4.1.
ID Task Name
2007
Q1
1 New Beamforming Techniques
2 New Adaptation Strategy
3 Overall Concatenated System
2008
Q2 Q3 Q4 Q1 Q2
4
5 Thesis Writing
Wrap-ups
Figure 4.1: Gantt chart for the remaining tasks.
47
Bibliography
[1] 3rd Generation Partnership Project Release 4, 3GPP Technical Report vol.4.0.0,
Mar.2003.
[2] I. E. Telatar, Capacity of multi-antenna Gaussian channels, Eur. Trans. Tele-
com., vol 10, pp. 585-595, Nov. 1999.
[3] G. J. Foschini, M. J. Gans, On limits of wireless communications in a fading
environment when using multiple antennas, Wireless Personal Communications,
vol. 6, no. 3, pp. 311-335, 1998.
[4] D. Gesbert, M. Sha, D. S. Shiu, P. Smith and A. Naguib, From theory to
practice: An overview of MIMO space-time coded wireless systems, IEEE J.
Select. Areas Commun., Special Issue on MIMO Systems, pt. I, vol. 21, pp. 281-
302, Apr. 2003.
[5] V. Tarokh, N. Seshadri, and A. Calderbank, Space-time codes for high data rate
wireless communications: Performance criterion and code construction, IEEE
Trans. Inform. Theory, vol. 44, pp. 744-765, Mar. 1998.
[6] S. Alamouti, A simple transmitter diversity scheme for wireless communica-
tions, IEEE J. Select. Areas Commun., vol. 16, pp. 1451-1458, Oct. 1998.
[7] D. Gesbert, R. W. Heath,and S. Catreux,V. Erceg Adaptive Modulation and
MIMO Coding for Broadband Wireless Data Networks, in 2002 IEEE Commu-
nications Magazine vol. 40, pp. 108-115 , June 2002.
[8] JK. Cavers, Variable-rate transmission for Rayleigh fading channels, IEEE
Transactions on Communications, COM-20, pp.15-22, 1972.
[9] J. Mitola, The software radio architecture, IEEE Communication Magazine,
33(5), pp.26-38, 1995.
48
[10] L. Yang and L. Hanzo, Software-dened-radio-assisted adaptive boardband fre-
quency hopping multicarrier DS-CMDA, IEEE Communication Magazine, 40(3),
pp.174-183, 2002.
[11] Z. Luo, H. Gao,and Y. Liu,J. Gao Capacity Limits of Time-Varying MIMO
Channels, IEEE International Conference On Communications vol.2, Mar.2003.
[12] Z. Shen, R. W. Heath, Jr., J. G. Andrews, and B. L. Evans, Comparison of
Space-Time Water-lling and Spatial Water-lling for MIMO Fading Channels,
in Proc. IEEE Int Global Communications Conf. vol. 1, pp. 431 435, Nov. 29-Dec.
3, 2004, Dallas, TX, USA.
[13] Z. Zhou and B. Vucetic Design of adaptive modulation using imperfect CSI in
MIMO systems, 2004 Eelectronics Letters vol. 40 no. 17, Aug. 2004.
[14] X. Zhang and B. Ottersten, Power allocation and bit loading for spatial mul-
tiplexing in MIMO systems, IEEE Int. Conf.on Acoustics, Speech, and Signal
Processing, 2003. Proceedings (ICASSP 03) vol.5 pp. 54-56, Apr. 2003.
[15] S. Zhou and G. B. Giannakis, Optimal Transmitter Eigen-Beamforming and
Space-Time Block Coding Based on Channel Mean Feedback IEEE Transactions
on Signal Processing, vol. 50, no. 10, October 2002.
[16] P. Xia,and G. B. Giannakis, Multiantenna Adaptive Modulation With Beam-
forming Based on Bandwidth- Constrained Feedback IEEE Transactions on
Communications, vol.53, no.3, March 2005.
[17] Bishwarup Mondal and Robert W. Heath, Jr., Performance Analysis of Quan-
tized Beamforming MIMO Systems IEEE Transactions on Signal Processing ,
vol. 54, no. 12, Dec. 2006.
[18] Y. Liu and M. Fitz, space-time turbo codes, 13th Annual Allerton Conf. on
Commun. Control and Computing, Sep. 1999.
[19] D. Cui and A. M. Haimovich, Design and performance of turbo space-time
coded modulation, IEEE GLOBECOM00, vol. 3, pp1627-1631, Nov. 2000.
[20] D. Tujkovic, Recursive space-time trellis codes for turbo coded modulation,
Proc. of GlobeCom 2000, San Francisco.
49
[21] V. Tarokh, H. Jafarkhani, and A. R. Calderbank, Space-time block code from
orthogonal designs, IEEE Trans. Inform. Theory, vol. 45, pp. 1456-1467, July
1999.
[22] G. J. Foschini, Layered space-time architecture for wireless communication in
fading environments when using multiple antennas, Bell labs. Tech. J.,vol. 1, no.
2, pp. 41-59, 1996.
[23] H. El Gamal and A. R. Hammons Jr., A new approach to layered space-time
coding and signal processing, IEEE Trans. Inf. Theory, vol. 47, pp. 2321-2334.
Sep. 2001.
[24] B. Hassibi and B. Hochwald, High-rate codes that are linear in space and time,
IEEE Trans. Inform. Theory, vol. 48, pp. 1804-1824, July 2002.
[25] R. W. Heath and A. Paulraj, Linear Dispersion Codes for MIMO Systems
Based on Frame Theory, IEEE Trans. on Signal Processing, vol. 50, No. 10, pp.
2429-2441, October 2002.
[26] X. Ma and G. B. Giannakis, Full-Diversity Full-Rate Complex-Field Space-
Time Coding, IEEE Trans. Signal Processing, vol. 51, no. 11, pp. 2917-2930,
July 2003.
[27] L. Zheng and D. Tse, Diversity and multiplexing: A fundamental tradeo in
multiple antenna channels IEEE Trans. Inform. Theory, vol. 49, pp. 1073-96,
May 2003.
[28] Z. Wu and X. F. Wang, Design of coded space-time modulation, IEEE Interna-
tional Conference on Wireless Networks, Communications and Mobile Computing,
vol. 2, pp. 1059-1064, Jun. 13-16, 2005.
[29] G. D. Forney Jr., Concatenated Codes, M. I. T. Press, Cambridge, Mas-
sachusetts, 1966.
[30] Youngwook KO and Cihan Tepedelenlioglu, Space-time block coded rate-
adaptive modulation with uncertain SNR feedback IEEE Signals, Systems and
Computers, vol.1, pp 1032- 1036, Nov. 2003.
[31] Bengt Holter, Geir E. mien, Kjell J. Hole, and Henrik Holm, Limitations in
spectral eciency of a rate-adaptive MIMO system utilizing pilot-aided channel
prediction IEEE 2003. VTC, vol. 1, pp 282- 286 , April 2003.
50
[32] Guillem Femenias, SR ARQ for Adaptive Modulation Systems Combined With
Selection Transmit Diversity 2005 IEEE Transaction on Communications, vol.
53, no. 6, pp 998-1006, June 2005.
[33] Marie-Hlne Bourles-Hamon and Hesham El Gamal,On the design of adaptive
space-time codes 2004 IEEE Transaction on Communications, vol. 52, no. 10,
Oct. 2004.
[34] Andreas Mller, Joachim SpeidelAdaptive modulation for MIMO spatial mul-
tiplexing systems with zero-forcing receivers in semi-correlated Rayleigh fading
channels Proceeding of the 2006 international conference on Communications
and mobile computing pp. 665 - 670, 2006.
[35] Zhendong Zhou, B. Vucetic, M. Dohler, Yonghui Li MIMO systems with adap-
tive modulation IEEE Transactions on Vehicular Technology vol.54, pp.1828-
1842, Sept. 2005.
[36] Robb J. Muirhead, Aspects Of Multivariate Statistical Theory, New York: John
Wiley and Sons Inc. 1982.
[37] T. W. Anderson, An Introduction to Multivariate Statistical Analysis,2th ed.
New York: John Wiley and Sons Inc. 1984.
[38] Anant M. Kshirsagar, Multivariate Analysis,Volume 2. New York: Marcel
Dekker,Inc. 1972.
[39] I. S. Gradshteyn ,I. M. Ryzhik,A. Jerey,and D. Zwillinger, Table of Integrals,
Series, and Products,6th ed. San Diego, Calif. : Academic Press 2000.
[40] R. Lupas and S. Verdu, Linear multiuser detectors for synchronous code-division
multiple-access channels, IEEE Trans. inform. Theory, vol. 35, pp. 123-136, Jan.
1989.
[41] H. V. Poor and S. Verdu, Probability of error in MMSE multiuser detection,
IEEE Trans. inform. Theory, vol. 43, pp. 858-871, May 1997.
[42] G. D. Golden, G. J. Foschini, R. A. Valenzuela, and P. W. Wolniansky, Detec-
tion algorithm and initial laboratory results using V-BLAST space-time commu-
nication architecture, Electron. Lett., vol. 35, pp. 14-16, Jan. 1999.
[43] S. Loyka , V-BLAST outage probability: analytical analysis, 2002 Proc. IEEE
Vehicular Technology Conference (VTC), vol.4 pp.1997-2001, Sept. 2002.
51
[44] R. Bhnke, Karl-Dirk Kammeyer, SINR Analysis for V-BLAST with Ordered
MMSE-SIC Detection, International Wireless Communications and Mobile Com-
puting Conference, pp. 623 - 628, July 2006.
[45] Marvin K. Simon ,Mohamed-Slim Alouini, Digital Communication over Fading
Channels,2th ed. , Hoboken, New Jersey : Wiley and Sons, Inc. 2005.
[46] G. E. Roberts and H. Kaufman, Table Of Laplace Transforms,Volume 2.
Philadelphia: W. B. Saunders Company. 1966.
T. Rapaport, Wireless Communications: Principles and Practice, 2nd ed. Pren-
tice Hall, 2001.
[47] J. Proakis, Digital Communications, 4th ed. New York: McGraw-Hill, 2001.
[48] T. M. Cover and J. A. Thomas, Elements of Information Theory, New York:
Wiley, 1991.
[49] B. Vucetic and J. Yuan, Space-Time Coding, John Wiley & Sons Ltd, England,
2003.
[50] S. Lin, D. J. Costello, Error Control Coding: Fundamentals and Applications,
Prentice Hall, New Jersey, 1983.
52

You might also like