Robust ML Detection Algorithm For Mimo Receivers in Presence of Channel Estimation Error

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

The 17th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC06)

ROBUST ML DETECTION ALGORITHM FOR MIMO RECEIVERS IN PRESENCE


OF CHANNEL ESTIMATION ERROR
Ahmad A. Farhoodi and Mehrzad Biguesh
DSP laboratory, Dept. of Elec. and Electronics Engineering,
Engineering School, Shiraz University, Shiraz, Iran

A BSTRACT

Tx

I.

I NTRODUCTION

Increasing demand for high data rate transmission, especially


in wireless communications, has motivated considerable attention on the use of multiple-input multiple-output (MIMO) wireless communication systems [1][2]. In fact, the bandwidth efficiency, the capability of reaching very high capacity, and the
available processing powers have made MIMO systems one of
the most promising techniques to achieve a high data rate signaling and a reliable transmission for the future wireless systems [3]. The key advantage of this technique is its ability
to combat the destructive effect of multipath propagation (i.e.
channel fading) and it can be shown that when the transmission medium is enriched with scatterers the system capacity
drastically increases in compare to single-input single-output
systems [4][5].
In a MIMO system, the transmitter is equipped with a set of
t antenna elements for transmission and a set of r antenna elements for receiving the signals. The transmitting data stream,
after probable coding, is demultiplexed into t parallel streams
and are transmitted by the set of t antennas. In this system,
there exist a channel between each transmit-receive antenna
pairs. Therefore, there are r t channels between transmitter
and receiver1 and in a rich scattering environment, these channels can be considered statistically independent [4]. To detect
1 Hereafter, we call these r t channels, the channel matrix or simply the
channel.

c
1-4244-0330-8/06/$20.002006
IEEE

Rx
Decoding and
MUX

Input
data

channel

Coding and
DEMUX

Capacity of wireless communication systems can be increased


enormously using arrays of antenna at transmitter and receiver.
However, in these so called multi-input multi-output (MIMO)
communication systems, the performance of the signal detection is tightly related to the channel knowledge at the receiver
side. Practically, the channel information at the receiver is subject to error because of the outdated data, quantization error,
and imperfect channel estimation/measurement. It is known
that in the case of imperfect channel knowledge, performance
of symbol detection in MIMO communications may severely
degrades.
The optimum detection method in the sense of symbol error
rate in MIMO systems is the maximum likelihood (ML) detection. In this paper, we proposed and studied a couple of robust
detection methods that improve the performance of ML detector in MIMO systems in the presence of channel matrix error
at the receiver.

output
data

Figure 1: A simple MIMO communication system for a flat fading


channel.

the signal at receiver, it is necessary to know these channel matrix.


There are methods to estimate the channel matrix at the receiver. These methods are mainly either training based channel matrix estimators [6][7] or the methods that estimate the
channel matrix blindly [8][9][10]. In the presence of noise, the
estimated channel deviate from the true channel for any estimation method. Additionally, quantization and outdated data
(specially when the channel is changing very fast) are other
sources of channel estimation errors.
Several techniques have been proposed to detect the transmitted symbols in MIMO receivers such as maximum likelihood (ML) method; that is optimum in the sense of symbol error rate (SER). However, this method is based on perfect channel knowledge at the receiver. Practically, these inevitable erroneous information about the channel matrix increases the SER
and demolishes the system performance.
In this paper, we briefly study the effect of channel estimation error on the performance of transmit symbol detection.
Then we propose some methods to cope with the demolishing
effect of channel errors on the ML detector in MIMO systems.
II.

MIMO SYSTEM MODEL

We consider a flat fading MIMO communication system with


t transmit and r receive antennas as shown in Fig. 1. We also
assume that the environment is reach in scatterers and the antennas at transmitter and receiver are separated enough such
that there are r t independent channels between transmitter
and receiver.
The transmit data; after probable coding; is demultiplexed
into t parallel substreams and each t data are transmitted by t
transmit antennas at each time instant. At kth time instant, the

The 17th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC06)

IV.

received signal can be expressed in the following matrix form,

CHANNEL ESTIMATION ERROR

y(k) = Hd(k) + n(k)

(1)

where d(k) = [d1 (k) dt (k)]T is the vector of transmitted data, y(k) = [y1 (k) yr (k)]T and n(k) =
[n1 (k) nr (k)]T represent the complex vector of received
signal and noise, respectively, and ()T stands for transposition.
The noise vector is assumed to be zero mean with covariance
matrix n2 I and is independent of transmit signal. Also, H is
the r t channel matrix where its elements are circularly symmetric complex Gaussian random variables [11]. Note that, in
(1) it is implicitly assumed that the channel does not changes
for a sufficiently long period of time. Hereafter, we drop the
time index k in our formulations for the sake of simplicity.
III.

ESTIMATION ERROR

Knowing the channel matrix H, for our assumed received signal model (1), if the noise is white Gaussian, the ML estimate
ML of the transmit symbols is found as,
d
ML = arg min y Hd2
d
t

A. Modified ML algorithm considering colored noise


is Gaussian distributed with zero
The equivalent noise vector n
mean and is independent of channel matrix. Here, we use the
modified version of ML detector in the presence of channel
estimation error. Let us use the definition of ML detectors,
which minimizes the probability of error as follows,
= d|H,
ML =arg min P rob{d
y}
d
t

dC


H R1 (y Hd)

=arg maxt exp (y Hd)

n
dC

(6)
where Rn is defined as (5).
Using the decomposition

ML DETECTION AND THE EFFECT OF CHANNEL

1
H
Rn
=U U

Here,   stands for Euclidian norm and Ct is the set of all


possible t dimensional transmit vectors.
As we explained earlier, an estimate of the channel matrix
is available at the receiver where it is corrupted by estimation
can be expressed in
error. The error free channel matrix H
terms of presumed channel matrix H as [12],
=H+E
H

(3)

The error matrix E is a complex Gaussian random matrix with


zero mean and here we assume that it is uncorrelated with the
transmit data and H, i.e.
E{EH H} = 0

(4)
H

where E{} is the expectation function and () denotes the


Hermitian transpose.
Using (3), the received signal y can be expressed as,

ML = arg min U(y Hd)2


d
t
dC

B.

Robust ML algorithm

In this subsection, we propose a robust ML detection algorithm


by means of an analytical approach. For our goal, let us write
the norm of y Hd in (2) as follows,
y Hd2 = (y Hd)H (y Hd)
= yH y yH Hd dH HH y + dH HH Hd
(9)
In the case that H is a square matrix (i.e, when the number
of transmit and receive antennas are equal), knowing that the
channel matrix H is full rank with probability one, (9) can be
written as,
y Hd2

Here,
where,

is the equivalent noise vector which is zero-mean Gaussian and


its correlation matrix is,
H }
= E{
nn
= E{(n + Ed)(n + Ed)H }
= Rn + d2 RE

(5)

where ,d2 = E{|dm |2 } and RE = E{EEH }. Using (5), it is


easy to understand that how channel estimation error increases
the noise level and therefore degrades the performance of ML
detector.

(8)

It is important to note that if Rn is proportional to identity


matrix I then (8) is exactly the same as ML detection method
(2).

+ n = (H + E)d + n = Hd + Ed + n = Hd + n

y = Hd
= n + Ed
n

(7)

the ML detection (6) can be written as,

(2)

dC

Rn

D ETECTION ALGORITHM CONCERNING THE

LS ]H HH H[d d
LS ]
= [d d
2
LS )
= H(d d

LS = (HH H)1 HH y
d

(10)
(11)

is the least squares estimate of the transmitted signal (it is also


known as zero-forcing equalization [13]).
As a result, in a MIMO system with the same number of
transmit and receive antennas (that is, whenever t = r), the
ML detection (2) can be written in the following form,
LS )2
ML = arg min H(d d
d
t
dC

(12)

In the sequel, we use (12) to make the ML estimator robust


against channel estimation error.

The 17th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC06)

In the case that the channel matrix is not known perfectly,


it is logical to use the following criterion that minimizes the
symbol error rate,


= d|H + E, y}
ML = arg min EE P rob{d
(13)
d
t

dC

where EE {} is the expected value on the channel error. Note


that (13) minimizes the average probability of error with respect to channel estimation error.
Based on our previous discussion, (13) can be written as,
LS )}
LS ) H H(d
ML = arg min EE {(d d
d
d
t
H

dC

Lemma: For any n1 nonzero vector x and nn matrix A,


which is unitarily similar to a diagonal matrix, the following
inequality holds:
xH Ax x2 A

when the equality is satisfied if and only if A = xxH , where


is an arbitrary scalar.
Based on the above lemma, if A is a Hermitian matrix and
b is an arbitrary vector then,
max bH Ab = bH b .

A

(14)

LS )2 + d d
LS 2
J = H(d d

ML = arg min (d d
LS )H EE {H
LS )
H H}(d

d
d
t
Here we can write,




= EE (H + E)H (H + E)
HH
EE H


= EE HH H + HH E + EH H + EH E

LS )H H
= arg min max EE {(d d
LS )}
H H(d
d
d
t
dC REH

s.t. REH F 
(17)
Let us write the following approximation for the expected value
in (17),


LS )H H
LS ) =
H H(d
d
EE (d d

= arg min H (d d


LS )2
d
t

H =

H
It

(24)


(25)

and It is the t-dimensional unit matrix.


As seen, our proposed robust ML (RML) detector (24) is similar to the traditional ML detector (2) except that channel matrix H is replaced with H . It is interesting to note that for
= 0 both the traditional and the proposed RML detectors are
exactly the same.
C. A conjectural RML method
Similar as before, it is straight forward to show the following
equality in the case that H is a square matrix,
LS )
U(y Hd) = UH(d d

(26)

As a result, the modified ML detector (8) can be written as,


= arg min UH(d d
LS )2
d
t
dC

(18)

(27)

Base on the result in the previous subsection and the above


equality, our conjecture is that the following may also prove
useful for detection in the presence of channel estimation error,

Using (18), the following can be written,


max

(23)

Noting that x2 + y2 = [xT yT ]T 2 , (23) can be expressed in the following compact form,

where, H is defined as,

Regarding the above constraint, we propose that the estimator


minimizes the estimation error for the worst effect of channel
error. That is, we propose to detect the transmit symbols using
the following min-max optimization problem,

REH 

dC

dC

In order to make the detection method robust against channel


estimation error, let us assume that the Frobenius norm of REH
is bounded as,
REH F 
(16)

J=

= arg min {H(d d


LS )2 + (d d
LS )2 }
d
t

(15)



REH = EE EH E

LS )H EE {H
LS )
H H}(d

(d d
d

(22)

Putting back (22) into (17), the the proposed ML detector can
be written as,

dC

where,

(21)

Using the above result, (19) can be written as,

We write Eq. 14 in the following approximate form,

= HH H + REH

(20)

LS )}
LS )H H
H H(d
d
EE {(d d

= arg min H (d d


LS )2
d
t

LS )H (HH H + REH )(d d


LS )
max (d d

dC

REH 

LS )H REH (d d
LS )
LS )2+ max (d d
=H(d d
REH 

(19)
To find the maximum value of (19), we use the following
lemma [14],

where,


H =

UH
It

(28)


(29)

and U is as defined in (7).


We investigate the performance of conjectured RML method
(28) using computer simulations.

The 17th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC06)

10

10

10

10

10

10

SER

10

SER

10

10

10

10

10

ML detection method (2)


Modified ML method (8)
Conjectured RML method (28)
Proposed RML method (24)
ML method (perfect channel knowledge)

10

10

10

10

15

SNR (dB)

Figure 2: SER versus SNR for various detection methods ( =


0.2, r = 0).

V.

C OMPUTER SIMULATION RESULTS

For simulation, a MIMO system with t = 2 transmit antennas


and r = 2 receive antennas is assumed. QPSK signaling is used
for transmission of the symbols over the random channel. A
random channel error is assumed in each simulation run where
the correlation matrix of the error matrix E has the following
structure,


r
(30)
RE =
r
Four detection methods are studied in our simulations. These
methods are the traditional ML detection method (2), the modified ML method (8), our proposed RML method (24), and the
conjectured RML method (28). Additionally, the performance
of ML detection method (2) in the case of perfect channel
knowledge; i.e. the optimum detection with error-free channel
knowledge; is also shown for comparison.
Fig. 2 shows the symbol error rates as a function of SNR
for the case where = 0.2, r = 0. Here, = 2 is used in
simulations. Fig. 3 shows the symbol error rates as a function
of SNR for the case where = 2, = 0.2 and r = 0.08.
Evidently our proposed RML method significantly outperforms the other three methods in the case that channel knowledge is imperfect. As seen, the performance of conjectured
RML method is between the RML and traditional ML methods. We have to mention that the curves related to traditional
ML detector and the modified ML methods are coincided in the
shown simulation results.
VI.

ML detection method (2)


Modified ML method (8)
Conjectured RML method (28)
Proposed RML method (24)
ML method (perfect channel knowledge)

S UMMARY AND C ONCLUSIONS

In this paper, we briefly studied the effect of imperfect channel


knowledge at receiver on the performance of signal detection
in MIMO communication systems. We analytically proposed
some methods to enhance the performance of symbol detection
in the presence of channel error. Using computer simulations,

10

10

15

SNR (dB)

Figure 3: SER versus SNR for various detection methods ( =


0.2, r = 0.08).

we showed that our proposed RML method significantly improves the performance of the symbol detection in compare to
the traditional ML method when the number of transmit and
receive antennas are the same.
ACKNOWLEDGE
The authors would like to thanks Iran Telecommunication Research Center (ITRC) for its support on this project.
R EFERENCES
[1] A. J. Paulraj, D. A. Gore, R. U. Nabar, and H. Bolcskei,
An overview of MIMO communication- a key to Gigabit
wireless, in Proceeding of the IEEE, vol. 92, no. 2, pp.
198-218 , Feb. 2004.
[2] D. Gesbert, M. Shafi, D. Shiu, P. J. Smith, and A. Naguib,
From theory to practice: An overview of MIMO spacetime coded wireless systems, IEEE J. Selec. Areas Comm.,
vol. 21, no. 3, pp. 281-302, April 2003.
[3] E. Teletar, Capacity of multi-antenna Gaussian channels,
European Trans. Telecommun., vol. 6, pp. 585-595, Dec.
1999.
[4] G. J. Foschini and M. J. Gans, On limits of wireless communications in a fading environment when using multiple
antennas,Wireless Personal Communication, vol. 6, no. 3,
pp. 311-335, Mar. 1998.
[5] R. Heath, S. Sandhu, and A. Paulraj, Antenna selection for
spatial multiplexing systems with linear receivers, IEEE
Communications Letters, vol. 5, no. 4, pp. 142-144, Apr.
2001.
[6] B. Hassibi and B. M. Hockwald, How much training is
needed in multiple-antenna wireless links?, IEEE Trans.
Info. Theory, vol. 49, no.4, pp. 951-963, April 2003.

The 17th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC06)

[7] M. Biguesh and A. B. Gershman, Training based MIMO


channel estimation: A study of estimator tradeoffs and optimal training signals, IEEE Trans. Sig. Proc., vol. 54,
no. 3, pp. 884-893, March 2006.
[8] S. Shahbazpanahi, A. B. Gershman, and J. H. Manton,
Closed-form blind MIMO channel estimation for orthogonal space-time block codes, IEEE Trans. Sig. Proc., vol.
53, no. 12, pp. 4506-4517, Dec. 2005.
[9] J. K. Tugnait and B. Huang , Blind channel estimation and
equalization of multiple-input multiple-output channels,
in Proc. IEEE Int. Conf. on Personal Wireless Comm., pp.
231-235, Feb. 1999.
[10] S. Yatawatta and A. P. Petropulu, Blind channel estimation in MIMO OFDM systems, in Proc. IEEE Workshop
on Stat. Sig. Proc., pp. 363-366, Sept./Oct. 2003.

[11] T. Yoo and A. Goldsmith, Capacity of fading MIMO


channels with channel estimation error, in Proc. IEEE Int.
Conf. on Comm. (ICC), vol.2, pp. 808-813, June 2004.
[12] H. Artes, D. Seethaler, and F. Hlawatsch, Efficient detection algorithm for MIMO channels: A geometrical approach to approximate ML detection, IEEE Trans. Sig.
Proc., vol. 51, no. 11, pp. 2808-2820, Nov. 2003.
[13] G. H. Golub and C. F. van Loan, Matrix Computations,
Johns Hopkins Univ. Press, 3rd edition, 1996.
[14] M. Biguesh, S. Shahbazpanahi, and A. B. Gershman,
Robust downlink power control in wireless cellular systems, EURASIP Journal on Wireless Communications
and Networking, pp. 261-272, Feb. 2004.

You might also like