Avoiding Matrix Inversion in DD SAGE Channel

Estimation in MIMO-OFDM with M-QAM

Jari Ylioinas

, M. R. Raghavendra, and Markku Juntti

Centre for Wireless Communications
P.O. Box 4500 FI-90014 University of Oulu
Finland
jari.ylioinas and markku.juntti@ee.oulu.
AbstractA decision directed (DD) channel estimation is
considered for a multiple-input multiple-output (MIMO) or-
thogonal frequency division multiplexing (OFDM) system to
improve the spectral efciency. Unlike in the pilot based channel
estimation, the least-squares (LS) channel estimator operating
in the DD mode for MIMO-OFDM requires a matrix inversion.
The size of the matrix to be inverted depends on the number
of transmit antennas and the length of the channel impulse re-
sponse. The frequency domain (FD) space-alternating generalized
expectation-maximization (SAGE) channel estimator calculates
the LS estimate iteratively avoiding the matrix inversion with
a constant envelope modulation. The drawback with the FD-
SAGE channel estimator is the required matrix inversion with
a non-constant envelope modulation. The size of the matrix
to be inverted depends on the length of the channel impulse
response. However, it is considerably less complex than the LS
channel estimator in the DD mode. In this paper, a time domain
(TD) SAGE channel estimator is derived to avoid the matrix
inversion in DD channel estimation for MIMO-OFDM systems
when using non-constant envelope modulation. The derived TD-
SAGE channel estimator is shown to offer the same performance
as the FD-SAGE channel estimator with reduced complexity.
I. INTRODUCTION
Multiple-input multiple-output (MIMO) communication
(see [1], [2] and references therein) combined with orthogonal
frequency division multiplexing (OFDM) is the key enabling
technology for several current and future broadband wireless
access systems and standards. Accurate channel state infor-
mation (CSI) between each transmit and receive antenna is
required at the receiver to coherently detect the information
from the received signal in MIMO-OFDM systems.
A channel estimation based on only pilot symbols fails
to exploit the channel information contained in the detected
and decoded data symbols. In order to achieve better spectral
efciency with fast fading channel conditions, the decision
directed (DD) channel estimation algorithms (see, e.g., [3,
Chapt. 15]) which take advantage of the detected and de-
coded data symbols in channel estimation, are a promising
approach. With the DD least-squares (LS) channel estimation,
calculation of a matrix inverse is required. The size of the
matrix to be inverted depends on the number of transmit
This research has been supported in part by Tekes, the Finnish Funding
Agency for Technology and Innovation, Elektrobit, Nokia, Nokia Siemens
Networks, Texas Instruments, and Uninord.
antennas and the length of the channel impulse response.
With constant envelope constellation, the frequency domain
(FD) space-alternating expectation-maximization (SAGE) [4]
channel estimator avoids this matrix inversion due to dividing
the MIMO channel estimation problem into multiple SISO
channel estimation problems. However, the drawback of the
FD-SAGE channel estimator is that it requires a matrix inver-
sion having the number of elements per one dimension equal to
the length of the channel impulse response when non-constant
envelope constellations are used.
In this paper, we derive the DD time domain (TD) SAGE
channel estimator for a MIMO-OFDM system. The SISO
channel estimation is further divided into multiple single tap
estimation problems by the DD-TD-SAGE estimator. The per-
formance and the complexity of the DD-TD-SAGE estimator
is compared to those of the DD-LS and DD-FD-SAGE estima-
tors. As a benchmark, the performance of the preamble-based
LS estimation with minimum mean square error (MMSE)
post-processing is provided as well. The proposed estimator
gives practically the same performance as the DD-FD-SAGE
channel estimator without the need for matrix inversion with
constant envelope constellations, and, thus, resulting in lower
complexity, which is also characterized.
II. SYSTEM MODEL
A. OFDM Signal Model
Let us consider a MIMO-OFDM system with P subcarriers,
M
T
transmit, and M
R
receive antennas. The received signal is
the superposition of M
T
distorted transmitted signals. Conse-
quently, the received signal on the m
R
th receive antenna at
time n after performing a discrete Fourier transform (DFT)
can be expressed as
y
m
R
(n) = X(n)Fh
m
R
(n) + w
m
R
(n), (1)
where y
m
R
= [y
m
R
,1
, ..., y
m
R
,p
, ..., y
m
R
,P
]
T
C
P
, X =
[X
1
, ..., X
M
T
] C
PM
T
P
consists of the transmitted symbols,
X
m
T
C
PP
includes the symbols transmitted over P
subcarriers from the m
T
th transmit antenna on its diagonal,
F = I
M
T
F C
M
T
PM
T
L
is the truncated DFT ma-
trix, with [F]
u,s
=
1

P
e
j2us/P
, and u = 0, 1, ..., P
1; s = 0, 1, ..., L 1, h
m
R
= [h
T
1,m
R
, ..., h
T
M
T
,m
R
]
T

978-1-4244-2515-0/09/$25.00 2009 IEEE

C
M
T
L
is the time domain channel vector, with h
m
T
,m
R
=
[h
m
T
,m
R
,0
, ..., h
m
T
,m
R
,l
, ..., h
m
T
,m
R
,L1
]
T
C
L
, and h
m
T
,m
R
,l
is the lth multipath component between the m
T
th transmit
and the m
R
th receive antenna, L is the length of the channel
impulse response for all channels, and w
m
R
(^(0,
2
w
I
P
).
B. Channel Model
A wideband stochastic MIMO channel model [5], [6] is
adopted. It is assumed that the amplitude of h
m
T
,m
R
,l
is
Rayleigh distributed, and the fading gains are uncorrelated
over the delay domain or h
m
T
,m
R
,l
1
, h
m
T
,m
R
,l
2
) = 0 for
l
1
,= l
2
, where a, b) = E[ab

]/

E[[a[
2
]E[[b[
2
] denotes
the normalized correlation coefcient between the random
variables a and b. The spatial correlation coefcient between
the transmit antennas m
T
and m

T
is given by
t
m
T
,m

T
=
h
m
T
,m
R
,l
, h
m

T
,m
R
,l
). The spatial correlation function at the
transmitter is assumed to be independent of the receive antenna
index m
R
. The correlation coefcient between the receive
antennas is correspondingly denoted by
r
m
R
,m

R
. The overall
spatial correlation model is assumed to obey the Kronecker
product model C
tr
= C
t
C
r
IR
M
T
M
R
M
T
M
R
, where C
t
and C
r
are the transmit and receive correlation matrices with
elements
t
m
T
,m

T
and
r
m
R
,m

R
, respectively [6].
III. RECEIVER ALGORITHMS
A. LS Channel Estimation
The LS method is the best linear unbiased channel estimator
in Gaussian noise [7]. The LS channel estimate for (1) is
expressed as

h
m
R
(n) =

F
H
X
H
(n)X(n)F

1
F
H
X
H
(n)y
m
R
(n). (2)
In the pilot based channel estimation, the pilots can be
designed to diagonalize the matrix to be inverted. However, if
high spectral efciency is sought and the LS method is used
in the DD mode, the inversion of a M
T
L M
T
L matrix is
inevitable. The LS estimator as well as the estimators below
assume L to be known in the receiver.
1
B. FD-SAGE Channel Estimation
The FD-SAGE channel estimator [4] calculates iteratively
the LS solution as in (2) resulting in lower complexity with
imperceptible loss in the performance. With the iterative
processing, the size of the required matrix inversion is M
T
times smaller per dimension in the case of non-constant
envelope modulation and with a constant envelope modulation
the matrix inversion can be avoided.
In the FD-SAGE channel estimator, the received signal
y
m
R
is viewed as the incomplete data and the complete
data is dened as z
m
T
,m
R
, which is the component of the
received signal at the m
R
th receive antenna transmitted by
the m
T
th transmit antenna. The FD-SAGE channel estimator
is initialized by
z
(0)
m
T
,m
R
(n) = X
m
T
(n)F

h
(0)
m
T
,m
R
(n), (3)
1
Channel length estimators are available in literature (see e.g. [8] and
references therein) and it is out of the scope of this paper.
where m
T
= 1, 2, ..., M
T
and m
R
= 1, 2, ..., M
R
. The SAGE
algorithms, as expectation-maximization (EM) based methods
in general, require and are sensitive to an initial guess of
the parameters to be estimated. This is well addressed in the
literature [9]. The channel estimate of the previous OFDM
symbol can be used as the initial time domain channel estimate

h
(0)
m
T
,m
R
(n). In the beginning of the transmission, the initial
channel estimate can be obtained by using a preamble.
After the initialization, the estimates of the channel impulse
responses between the transmit antennas and the receive an-
tenna m
R
are rened by the following iterations (for notational
simplicity the time index n is omitted)
z
(i)
m
T
,m
R
= z
(i)
m
T
,m
R
+

y
m
R

M
T

T
=1
z
(i)
m

T
,m
R

(4)

h
(i+1)
m
T
,m
R
= (F
H
X
H
m
T
X
m
T
F)
1
F
H
X
H
m
T
z
(i)
m
T
,m
R
(5)
z
(i+1)
m
T
,m
R
= X
m
T
F

h
(i+1)
m
T
,m
R
(6)
z
(i+1)
m

T
,m
R
= z
(i)
m

T
,m
R
, (7)
where m
T
= 1 + (i mod M
T
), (i = 0, 1, 2, ...), and in (7)
1 m

T
M
T
but m

T
,= m
T
. With one iteration in the case
of the FD-SAGE channel estimator, we refer to the cycles
during which every link is updated once. The above algorithm
transforms the LS channel estimation problem of a system with
multiple transmit antennas as in (2) back to the multiple sep-
arate single transmit antenna LS channel estimation problems
performed serially. Consequently, the matrix inversion needed
is M
T
times smaller per dimension than the one required with
the DD-LS estimation. Furthermore, with constant envelope
modulation, (5) can be simplied to

h
(i+1)
m
T
,m
R
(n) = F
H
X
1
m
T
(n)z
(i)
m
T
,m
R
(n) (8)
leading to the trivial inversion of the diagonal matrix X
m
T
(n).
A more detailed description of the algorithm can be found in
[4] in which constant envelope constellation is assumed and
the drawback of the required inversion of a LL matrix with
non-constant envelope constellations was not treated.
C. TD-SAGE Channel Estimation
The matrix inversion in the FD-SAGE algorithm with non-
constant envelope modulations can be avoided by choosing
the incomplete and complete data spaces so that in the
maximization step only one tap of one link is rened at a time.
Then, the incomplete data is viewed as the time domain
received signal y
m
R
and the complete data z
m
T
,m
R
,l
, which is
the component of the time domain received signal at the m
R
th
receive antenna transmitted by the m
T
th transmit antenna and
propagated through the lth path, is dened by
z
m
T
,m
R
,l
= x
m
T
,l
h
m
T
,m
R
,l
+ w
m
R
, (9)
where m
T
= 1, 2, ..., M
T
, m
R
= 1, 2, ..., M
R
, l = 0, ..., L 1,
the column vector x
m
T
,l
= x
m
T
(L
CP
+ 1 l : P + L
CP
l),
x
m
T
= T
CP
F
H
f
diag(X
m
T
) is the time domain signal including
the CP from the m
T
th transmit antenna, T
CP
= [I
T
L
CP
I
T
P
]
T
,
TABLE I
COMPLEXITY OF THE ALGORITHMS
Algorithm # complex multiplications # complex divisions
LS M
R
(M
3
T
L
3
+M
2
T
P
2
L +M
2
T
L
2
(P + 1) +M
T
PL) M
R
(M
T
L)
2
FD-SAGE (simp.) (N
i
+ 1)M
T
M
R
P(2L + 2) +N
i
M
T
M
R
P -
FD-SAGE (gen.) N
i
M
T
M
R
(L
3
+PL
2
+L
2
+ 3PL +P) +M
T
M
R
(LP +P) N
i
M
T
M
R
L
2
TD-SAGE N
i
M
T
M
R
L(4P + 1) N
i
M
T
M
R
L
OFDM
demod.
OFDM
demod.
Soft
MIMO
detector
Turbo
decoder
Channel
estimator
Interleaver
DeInterleaver
Decisions
LD1
LA1
LE1
LA2
LD2 LE2
Fig. 1. Receiver structure.
I
L
CP
contains the last L
CP
rows from I
P
, L
CP
is the length of
the cyclic prex, diag() takes the diagonal of a matrix and
makes it a vector, and w
m
R
is the noise term of (1) in the time
domain. The time domain SAGE estimator is initialized by
z
(0)
m
T
,m
R
,l
= x
m
T
,l

h
(0)
m
T
,m
R
,l
. (10)
The channel estimate of the previous OFDM symbol is used
as the initial channel estimate similarly as with the FD-SAGE
channel estimator. After the initialization, the estimates of the
channel impulse responses between the transmit antennas and
the receive antenna m
R
are rened tap by tap through the
following iterations
z
(i)
m
T
,m
R
,l
= z
(i)
m
T
,m
R
,l
+

y
m
R

M
T

T
=1
L1

=0
z
(i)
m

T
,m
R
,l

(11)

h
(i+1)
m
T
,m
R
,l
(n) =
x
H
m
T
,l
z
(i)
m
T
,m
R
(n)
x
H
m
T
,l
x
m
T
,l
(12)
z
(i+1)
m
T
,m
R
,l
(n) = x
m
T
,l

h
(i+1)
m
T
,m
R
,l
(13)
z
(i+1)
m

T
,m
R
,l

(n) = z
(i)
m

T
,m
R
,l

, (14)
where m
T
= 1 + (i mod (M
T
+ L)), l = (i mod L), (i =
0, 1, 2, ...), and in (14) 1 m

T
M
T
but m

T
,= m
T
as well
as 0 l

L but l

,= l. With one iteration in the case of

the TD-SAGE estimator, we refer to the cycles during which
every link is updated once.
D. Receiver Structure
The receiver structure is shown in Fig. 1. The feedback
from the decoder to the channel estimator is required only
when the SAGE estimators are applied. In the core of the
receiver, iterative detection and decoding is performed by the
soft MIMO detector and the turbo decoder. The iterations
between the detector and the decoder are referred to as global
iterations. Let us assume that one frame includes multiple
OFDM symbols and the channel coding is over one OFDM
symbol. In each frame, the preamble symbol index set is
o
P
= 1, N
symb
/2+1, where N
symb
is the number of OFDM
symbols in the frame.
In the preamble-based LS estimation, the estimate given by
the rst preamble is used in the detection of the symbols with
indices 2...N
symb
/2 1 and the estimate given by the second
preamble is used in the detection of the symbols with indices
N
symb
/2+2...N
symb
. In the preamble based LS estimation with
the MMSE post-processing, the two LS estimates given by
the preambles are ltered by the MMSE lter which exploits
the time domain correlation of the channel taps to calculate
channel estimates also for data symbol positions. In the DD
SAGE channel estimation, the LS estimate obtained from
the preamble is used as an initial guess in the beginning of
the frame and after the second OFDM symbol the previous
estimate is used to initialize the SAGE algorithm. The SAGE
algorithm is performed after the global iterations.
IV. COMPLEXITY COMPARISONS
The complexity of the TD-SAGE channel estimator is com-
pared to those of the FD-SAGE channel estimator and the LS
algorithm in DD mode. For the FD-SAGE channel estimator,
the complexity is presented in both the simplied form for
constant envelope constellation and in the general form for
non-constant envelope constellations. The required number of
complex multiplications and divisions for the algorithms is
presented in symbolic form in Table I. In Table II, an example
of the number of required operations is presented assuming
M
T
= M
R
= 2, L = 10, P = 512, and the number of
iterations, N
i
, is set 3. We consider
1
x
as multiplication, when
x belongs to a set with nite alphabet, since these inverse
values can be precomputed and stored into a memory. One can
note that the DD-LS method requires clearly a larger number
of complex multiplications than the other algorithms. The FD-
SAGE channel estimator in the general form is more complex
than the FD-SAGE channel estimator in the simplied form
due to the matrix inversion. The TD-SAGE channel estimator
can be stated to be clearly less complex than the FD-SAGE
channel estimator in the general form.
TABLE II
NUMERICAL EXAMPLE OF THE COMPLEXITY
Algorithm # complex multiplications # complex divisions
LS 21418400 800
FD-SAGE (simp.) 186368 -
FD-SAGE (gen.) 840592 1200
TD-SAGE 245880 120
V. NUMERICAL EXAMPLES
Monte Carlo computer simulations were conducted to study
the performance of the channel estimation algorithms. The
entire channel bandwidth of 7.68 MHz was divided into
512 subcarriers. The symbol duration of 66.7 s and the
cyclic prex length of 16.68 s were assumed. The carrier
frequency was 2.4 GHz. The simulations were performed
with the typical urban (TU) channel model [10]. The key
parameters of the stochastic MIMO radio channel model are
summarized in [11, Table I]. The signal-to-noise ratio (SNR)
is dened as E
S
/N
0
, where E
S
is symbol energy received
by a receive antenna and N
0
I is the covariance matrix of the
noise. Only one iteration is performed between the soft MIMO
detector and the turbo decoder. Turbo coding of rate 1/2 was
assumed. The turbo coding scheme is a parallel concatenated
convolutional code (PCCC) with two 8-state encoders and
an internal interleaver followed by puncturing, multiplexing,
and bit-level interleaving. The constituent codes (CCs) are
rate 1/2 binary recursive systematic convolutional codes [12].
The turbo decoder consists of two component maximum a
posteriori (MAP) decoders [13] and 8 decoder iterations were
performed. Coding was performed over the transmit antennas
and one code word consisted of 16QAM or 64QAM symbols
transmitted over one OFDM symbol. The increasing radius list
sphere detector (IR-LSD) [14] was operating as a soft MIMO
detector.
Mean square error (MSE) performance of the TD-SAGE
channel estimator is compared to that of the FD-SAGE and LS
channel estimators in 22 and 44 antenna congurations in
Figs. 2 and 3, respectively. Also, the performance of the SAGE
estimator in the simplied form as in (8) is given to show
the loss due to the simplication with non-constant envelope
constellations. For the SAGE estimators, the MSE is given
after the rst and the third iterations. In these cases, the known
transmitted symbols are used in the estimation in order to
evaluate the channel estimation performance and the previous
channel estimate is used to initialize the SAGE algorithms.
64QAM and mobile speed of 120 km/h are assumed. One
can note that a loss of about 5 dB is caused in MSE in
the 2 2 antenna conguration if the FD-SAGE channel
estimator is used in the simplied form. In the 4 4 antenna
conguration, the loss of the simplied FD-SAGE channel
estimator is 6.5 dB. The TD-SAGE channel estimator gives
practically the same performance as the FD-SAGE channel
estimator in general form with lower complexity in both
antenna congurations. One can note from Fig. 3 that three
iterations for the TD-SAGE and FD-SAGE channel estimators
are not enough to achieve the LS performance beyond 30 dB
in the 4 4 antenna conguration.
The frame error rate (FER) performance results for the
SAGE estimators with 16QAM and 64QAM are presented
in Fig. 4. With frame error, we refer to an erroneous code
word. Here, the detected and decoded data symbols are used
in the channel estimation. Only every 25th OFDM symbol
was a preamble symbol meaning that in every 25th OFDM
symbol all the sub-carriers from all the transmit antennas
are used for pilot transmission. The channel estimate for the
previous OFDM symbol is used to initialize the SAGE channel
estimators. Mobile speed of 50 km/h was assumed. It can be
seen that the FD-SAGE channel estimator in the general form
and the TD-SAGE channel estimator exhibit approximately
the same performance. The FD-SAGE channel estimator in
0 5 10 15 20 25 30 35 40
10
5
10
4
10
3
10
2
10
1
10
0
E
S
/N
0
[dB]
M
S
E


LS
TDSAGE 1 iterat.
TDSAGE 3 iterat.
gen. FDSAGE 1 iterat.
gen. FDSAGE 3 iterat.
simp. FDSAGE 1 iterat.
simp. FDSAGE 3 iterat.
Fig. 2. MSE performance in TU channel with 64QAM (22).
0 5 10 15 20 25 30 35 40
10
5
10
4
10
3
10
2
10
1
10
0
10
1
E
S
/N
0
[dB]
M
S
E


LS
TDSAGE 1 iterat.
TDSAGE 3 iterat.
gen. FDSAGE 1 iterat.
gen. FDSAGE 3 iterat.
simp. FDSAGE 1 iterat.
simp. FDSAGE 3 iterat.
Fig. 3. MSE performance in TU channel with 64QAM (44).
the simplied form loses approximately 1 dB in 16QAM case
and about 2 dB in 64QAM case to the general FD-SAGE
and the TD-SAGE channel estimators. The TD-SAGE and
general form FD-SAGE channel estimators lose around 2 dB
in 16QAM and close to 4 dB in 64QAM when compared
to the performance with the ideal CSI in the receiver. The
preamble-based LS estimator with the time domain MMSE
post-processing [15] is presented as well. The preamble-based
channel estimation suffers from saturation in the performance
due to the low preamble density.
VI. CONCLUSIONS
Decision directed channel estimation was considered for
MIMO-OFDM systems to improve the spectral efciency by
decreasing the pilot overhead. The conventional LS estimator
operating in the DD mode requires inversion of a matrix
having the size per dimension dependent of the number of
transmit antennas and the length of the channel impulse
response. The matrix inversion can be avoided with the FD-
SAGE channel estimator with a constant envelope modulation.
16 18 20 22 24 26 28 30 32 34
10
2
10
1
10
0
E
S
/N
0
[dB]
F
E
R


Ideal CSI
TDSAGE
gen. FDSAGE
simp. FDSAGE
Preamble based
16QAM
64QAM
Fig. 4. FER performance in TU channel with 16QAM and 64QAM (22).
However, the drawback with the FD-SAGE channel estimator
is the required inversion of a matrix of size depending on the
length of the channel impulse response when non-constant
envelope constellations are used. The TD-SAGE estimator
was derived to lower the complexity by avoiding the matrix
inversion without performance degradation.
REFERENCES
[1] D. Gesbert, M. Sha, D. Shiu, P. J. Smith, and A. Naguib, From theory
to practice: An overview of MIMO space-time coded wireless systems,
IEEE J. Select. Areas Commun., vol. 21, no. 3, pp. 281302, 2003.
[2] A. J. Paulraj, D. A. Gore, R. U. Nabar, and H. Bolcskei, An overview
of MIMO communications A key to gigabit wireless, Proc. IEEE,
vol. 92, no. 2, pp. 198218, Feb. 2004.
[3] L. Hanzo, M. M unster, B. J. Choi, and T. Keller, OFDM and MC-
CDMA for Broadcasting Multi-User Communications, WLANs and
Broadcasting. New York: John Wiley & Sons, 2003.
[4] Y. Xie and C. N. Georghiades, Two EMtype channel estimation
algorithms for OFDM with transmitter diversity, IEEE Trans. Commun.,
vol. 51, no. 1, pp. 106115, Jan. 2003.
[5] K. I. Pedersen, J. B. Andersen, J. P. Kermoal, and P. Mogensen, A sto-
hastic multiple-input-mobile-output radio channel model for evaluation
of space-time coding algorithms, in Proc. IEEE Veh. Technol. Conf.,
Boston, USA, Sept. 2428 2000, pp. 893897.
[6] J. P. Kermoal, L. Schumacher, K. Pedersen, P. E. Mogensen, and F. Fred-
eriksen, A stochastic MIMO radio channel model with experimental
validation, IEEE J. Select. Areas Commun., vol. 20, no. 6, pp. 1211
1226, Aug. 2002.
[7] S. M. Kay, Fundamentals of Statistical Signal Processing: Estimation
Theory. Englewood Cliffs, NJ, USA: Prentice-Hall, 1993.
[8] V. D. Nguyen, H. P. Kuchenbecker, H. Haas, K. Kyamakya, and G. Gelle,
Channel impulse response length and noise variance estimation for
ofdm systems with adaptive guard interval, EURASIP J. Wireless
Comm. and Netw., vol. 2007, no. 1, 2007.
[9] G. J. McLachlan and T. Krishnan, The EM Algorithm and Extensions.
New York, USA: Wiley, 1997.
[10] 3rd Generation Partnership Project (3GPP); Technical Specication
Group Radio Access Network, Radio transmission and reception (3G
TS 45.005 version 5.4.0 (release 5)), 3rd Generation Partnership Project
(3GPP), Tech. Rep., 2002.
[11] J. Ylioinas and M. Juntti, Iterative joint detection, decoding, and
channel estimation in turbo coded MIMO-OFDM, IEEE Trans. Veh.
Technol.to appear.
[12] 3rd Generation Partnership Project (3GPP); Technical Specication
Group Radio Access Network, Multiplexing and channel coding (FDD)
(3G TS 25.212 version 7.2.0), 3rd Generation Partnership Project
(3GPP), Tech. Rep., 2006.
[13] L. R. Bahl, J. Cocke, F. Jelinek, and J. Raviv, Optimal decoding of
linear codes for minimizing symbol error rate, IEEE Trans. Inform.
Theory, vol. 20, no. 2, pp. 284287, Mar. 1974.
[14] M. Myllyl a, M. Juntti, and J. R. Cavallaro, A list sphere detector based
on Dijkstras algorithm for MIMO-OFDM system, in Proc. IEEE Int.
Symp. Pers., Indoor, Mobile Radio Commun., Athens, Greece, Sept. 3-7
2007, pp. 15.
[15] H. Miao and M. Juntti, Space-time channel estimation and performance
analysis for wireless MIMO-OFDM systems with spatial correlation,
IEEE Trans. on Vehicular Technology, vol. 54, no. 6, pp. 2003 2016,
Nov. 2005.