ADAPTIVE CHANNEL MATCHED EXTENDED ALAMOUTI SPACE-TIME

CODE EXPLOITING PARTIAL FEEDBACK

Biljana Badic, Markus Rupp, Hans Weinrichter
Institute of Communications and Radio Frequency Engineering
Vienna University of Technology
Gusshausstr.25, A-1040 Vienna, Austria
Phone: +4315880138922
Fax: +4315880138999
Email: bbadic@nt.tuwien.ac.at
AbstractSince the publication of Alamoutis famous (2 2)
space-time block code, various suboptimal space-time codes
for multi-input/multi-output fading channels for four transmit
antennas have been proposed not achieving full diversity at
full rate. In this paper we present a simple adaptive, channel
dependent, rate-one Extended Alamouti space-time block code
for four transmit antennas and one receive antenna using partial
feedback. The relevant channel state information is sent back
from the receiver to the transmitter quantized to one or two bits
per code block. In this way signal transmission with improved
diversity is achieved being near to the maximum diversity value
of d = 4. Such high diversity can be exploited with zero forcing
receiver as well as with maximum-likelihood (ML) receiver.
I. Introduction
Since the work of Alamouti [1], several orthogonal
and quasi-orthogonal space-time code designs have been
investigated. The simple Alamouti scheme using two transmit
antennas increases the diversity at the receiver from d = 1
to d = 2 in a at fading channel. It was shown in [2], that
orthogonal full-rate design offering full diversity for any
arbitrary complex symbol constellation is limited to the case
of two transmit antennas. Data-rate or decoding simplicity
must be sacriced if the number of antennas is increased.
To preserve the full rate at a small loss in complexity and
performance, quasi- orthogonal designs have been proposed
[3].
Most of the space-time block codes (STBC) are designed
under the assumption that the transmitter has no knowledge
about the channel. It has been shown that outage performance
with perfect channel state information (CSI) at the transmitter
and at the receiver is better compared to the case when
only the receiver has perfect knowledge of the channel.
For instance, with complete CSI at the transmitter data
can be transmitted on the eigenvector related to the largest
eigenvalue [4] providing full diversity combined with a
substantial transmit array gain. In some situations, low
coding complexity, high diversity and a better code rate is
obtained with only partial feedback to the transmitter [5].
It is possible, for instance, that the receiver only returns
quantized amplitude information about the channel, as it has
been proposed for certain power control schemes [6]. In the
scheme proposed in [5] the relevant CSI is sent back from
the receiver to the transmitter using one bit per code block.
Research on adapting the block code considering partial
feedback has been an intensive area of research, with several
design strategies having been proposed recently. In this paper
a simple way to combine extended Alamouti space-time block
codes over four transmit antennas is proposed to achieve
full diversity and a low bit error rate with a full rate code.
First, a feedback system returning only one bit (b = 1)
to the transmitter is presented. Depending on the value of
b the transmitter switches between two predened STBCs
and chooses that code which achieves higher diversity. In a
second approach the receiver sends back two bits per block to
the transmitter for switching between four space-time block
codes. With this simple scheme a ZF as well as a ML receiver
achieves full rate data transmission with optimum diversity
d = 4.
The paper is organized as follows. First, a short overwiev of
the well-known Alamouti scheme is given and the Extended
Alamouti space-time block code (EASTBC) is explained. In
Section III the generation of the feedback information is
explained and the probability of a certain random variable
causing channel interference is derived. This variable controlls
the feedback information for both one and two bit feedback
information per code block. Simulation results are presented
in Section IV. and conclusions are given in Section V.
II. System Design based on Extended Alamouti Schemes
A. Review of the 2 1 Alamouti Scheme
A very simple and effective coding scheme for two transmit
antennas and a single receive antenna was introduced by
Alamouti [1]. The sequence s
1
, s

2
is sent over the rst antenna
and s
2
, s

1
over the second antenna, where denotes complex
conjugation. Assuming a at-fading channel with coefcients
h
1
and h
2
, the received vector r is formed by stacking two
consecutive data samples r = [r
1
, r
2
]
T
in time, resulting in
r = Sh +v, (1)
International Conference on Cellular and Intelligent Communications (CIC), Seoul, Korea; pp. 350, Oct.
2003.

where h = [h
1
, h
2
]
T
is a channel vector and v is the noise
vector. Explicitly (1) reads
r
1
= s
1
h
1
+ s
2
h
2
+ v
1
r
2
= s

2
h
1
s

1
h
2
+ v
2
.
Here the symbol block S is dened as:
S =
_
s
1
s
2
s

2
s

1
_
. (2)
This is equivalent to:
r
1
= h
1
s
1
+ h
2
s
2
+ v
1
(3)
r

2
= h

2
s
1
+ h

1
s
2
+ v
2
. (4)
Vector equation (1) can be rewritten as
_
r
1
r

2
_
=
_
h
1
h
2
h

2
h

1
__
s
1
s
2
_
+
_
v
1
v

2
_
or in short notation:
y = Hs +v, (5)
where the vector y = [r
1
, r

2
]
T
has been introduced. The
resulting channel matrix H is orthogonal, i.e.
H
H
H = HH
H
= h
2
I
2
,
where the 22 identity matrix I
2
and the gain of the channel
h
2
= |h
1
|
2
+|h
2
|
2
has been introduced.
B. Extended Alamouti Scheme
In the case of four transmit and one receive antenna (Fig-
ure 1) the space-time code matrix can be dened as [8]:
Extendend
Alamouti Receiver
Output Input
noise
Channel
s
h
4
y
1
h
2
3
h
h
Fig. 1. Extended Alamouti Scheme
S =
_
_
_
_
s
1
s
2
s
3
s
4
s

2
s

1
s

4
s

3
s

3
s

4
s

1
s

2
s
4
s
3
s
2
s
1
_
_
_
_
. (6)
Following the previous equations the four-antenna case can be
described by
r
1
= s
1
h
1
+ s
2
h
2
+ s
3
h
3
+ s
4
h
4
r
2
= s

2
h
1
s

1
h
2
+ s

4
h
3
s

3
h
4
r
3
= s

3
h
1
+ s

4
h
2
s

1
h
3
s

2
h
4
r
4
= s
4
h
1
s
3
h
2
s
2
h
3
+ s
1
h
4
.
With the complex conjugation of the second and the third
equation cited above, the received vector can be expressed as
y
1
= r
1
, v
1
= v
1
y
2
= r

2
, v
2
= v

2
y
3
= r

3
, v
3
= v

3
y
4
= r
4
, v
4
= v
4
.
resulting in the following equivalent transmission scheme
y = Hs +v, (7)
where
H =
_
_
_
_
h
1
h
2
h
3
h
4
h

2
h

1
h

4
h

3
h

3
h

4
h

1
h

2
h
4
h
3
h
2
h
1
_
_
_
_
(8)
is the effective channel transmission matrix. In this way, a
virtual, specicly structured channel with four transmit and
four receive antennas is obtained. In the following it will be
shown that H is nearly orthogonal.
III. Feedback Approaches
A. EASTBC scheme with one channel information bit per code
block returned back to the transmitter
The feedback scheme of one bit per code block from the
receiver to the transmitter characterizing the channel will be
explained rst. The scheme is depicted in Figure 2.
The EASTBC consists of two (4 4) code blocks S
1
and S
2
Fig. 2. Scheme with One Bit Feedback
of length T = 4 time intervals. Four transmit and one receive
antennas and a channel transfer vector h = [h
1
, h
2
, h
3
, h
4
]
T
are considered. The channel transfer elements may fade in any
arbitrary way but are assumed to be constant during the code
block of length four. The signal transmission can be described
by
y = Sh +v, (9)
where y is the (4 1) vector of received signals from the
code-block S of length T = 4. S is either S
1
or S
2
and
depending on the channel parameter b dened below. v is the
(4 1) noise vector with complex Gaussian components with
zero mean and variance
2
v
. S
1
and S
2
are dened as:
S
1(b=1)
=
_

_
s
1
s
2
s
3
s
4
s

2
s

1
s

4
s

3
s

3
s

4
s

1
s

2
s
4
s
3
s
2
s
1
_

_
(10)
S
2
(b=1)
=
_

_
s
1
s
2
s
3
s
4
s

2
s

1
s

4
s

3
s

3
s

4
s

1
s

2
s
4
s
3
s
2
s
1
_

_
(11)
Obviously, S
1
and S
2
differ only in the sign of the symbols
in the rst column. During simulation the signal elements s
1
to
s
4
are taken from a simple QPSK signal constellation. Eqn.(9)
can be rewritten in the form of Eqn.(5) as
y = Hs +v,
with s = [s
1
, s
2
, s
3
, s
4
]
T
and the effective channel matrix
H, that is now equal to
H
1
=
_

_
h
1
h
2
h
3
h
4
h

2
h

1
h

4
h

3
h

3
h

4
h

1
h

2
h
4
h
3
h
2
h
1
_

_
, (12)
if S = S
1
, or
H
2
=
_

_
h
1
h
2
h
3
h
4
h

2
h

1
h

4
h

3
h

3
h

4
h

1
h

2
h
4
h
3
h
2
h
1
_

_
, (13)
if S = S
2
.
In both cases we obtain (see also [8])
G
i
= H
H
i
H
i
= H
i
H
H
i
= h
2

I
2
X
i
J
2
X
i
J
2
I
2

(14)
(i = 1, 2)
with
I
2
=
_
1 0
0 1
_
, J
2
=
_
0 1
1 0
_
,
h
2
= |h
1
|
2
+|h
2
|
2
+|h
3
|
2
+|h
4
|
2
and X
i
equal to
X
1
(b=1)
=
2Re(h
1
h

4
h
2
h

3
)
h
2
(15)
X
2
(b=1)
=
2Re(h
1
h

4
h
2
h

3
)
h
2
It is well known that G should approximate a scaled identity-
matrix as far as possible to achieve full diversity and optimum
BER performance. This means, X should be as small as
possible. As G indicates, our scheme inherently supports full
diversity d = 4, if X = 0.
Therefore, the strategy is to transmit that code S
1
or S
2
that minimizes |X|. In any case, a performance loss due
to the non vanishing value of X is expected causing some
interference between the signal elements s
1
and s
4
, or s
2
and s
3
, respectively. As it is assumed that the receiver has
full information of the channel state, knowing h
1
to h
4
, the
receiver can compute X
1
and X
2
. With this information the
receiver returns the feedback bit b informing the transmitter to
select that code block S
i
(i = 1, 2) which leads to the smaller
value of X
i
. With this information the transmitter switches
between the EASTBC S
1
and S
2
such that the resulting |X|
will be min(|X
1
|, |X
2
|). Obviously the control information
sent back to the transmitter only needs one feedback bit
information per code block. During simulation it is assumed
that the channel varies slowly such that the delay of the
feedback information can be neglected.
B. Derivation of the PDF of the interference parameter X
If the channel coefcients h
i
are i.i.d. complex Gaussian
variables, then the probability density function of X is given
in [8] as:
f
X
(x) =
_
3
4
(1 x
2
) ; |x| < 1,
0 ; else
(16)
To derive the probability density function of min(|X
1
|, |X
2
|)
in case of one feedback bit per code block, the PDF of two
random variables needs to be considered. For this purpose a
new random variable is dened as:
W =
_
X
1
;if |X
1
| < |X
2
|
X
2
;else
. (17)
Due to its symmetry only the one sided (positive) PDF is
considered:
f
W
(w) = f
X
1
(w) + f
X
2
(w) f
X
1
(w)F
X
2
(w)
F
X
1
(w)f
X
2
(w)
= 2f
X
1
(w)(1 F
X
1
(z)) (18)
where X
1
, X
2
are assumed to be two statistically independent
random variables. The nal solution for f
W
(w) is :
f
W
(w) =

3
2
(1 w
2
)

1
3
2
|w|(1
w
2
3
)

; |w| 1
0 ; else
(19)
The simulations results verifying this result are presented in
Fig. 3 further ahead.
C. Two bits fed back to the transmitter
In a similar way as discussed in the section A we can send
more than one bit back to the transmitter. Let us discuss the
case when we send two bits b
1
, b
2
as a feedback information
to the transmitter. Now, we dene four versions of S, namely
S
1
and S
2
dened in Eqn.(10) and Eqn.(11) and additionally
S
3
, S
4
dened as:
S
3
=
_

_
js
1
js
2
s
3
s
4
js

2
js

1
s

4
s

3
js

3
js

4
s

1
s

2
js
4
js
3
s
2
s
1
_

_
(20)
S
4
=
_

_
js
1
js
2
s
3
s
4
js

2
js

1
s

4
s

3
js

3
js

4
s

1
s

2
js
4
js
3
s
2
s
1
_

_
. (21)
With S
3
and S
4
the corresponding channel matrix H is equal
to:
H
3
=
_

_
jh
1
jh
2
h
3
h
4
jh

2
jh

1
h

4
h

3
h

3
h

4
jh

1
jh

2
h
4
h
3
jh
2
jh
1
_

_
(22)
and
H
4
=
_

_
jh
1
jh
2
h
3
h
4
jh

2
jh

1
h

4
h

3
h

3
h

4
jh

1
jh

2
h
4
h
3
jh
2
jh
1
_

_
(23)
respectively.
The resulting matrix G and channel gain h can be written as
in Eqn.(14). The channel dependent interference parameter X
in case of S
3
and S
4
results now in:
X
3
=
2Im(h
1
h

4
+ h
2
h

3
)
h
2
(24)
X
4
=
2Im(h
1
h

4
h
2
h

3
)
h
2
.
Using two feed back bits the transmitter can switch between
four space-time block codes S
i
, i = 1, 2, 3, 4, to provide still
higher diversity and smaller bit error rate than in case of
relying only on one bit feedback information. The transmitter
chooses that block-code S
i
which leads to an interference
parameter Z with minimum absolute value presented in (25)
further ahead.
D. Derivation of the PDF of Z
As explained in the last section the main idea of our adaptive
coding is to reduce the relevant interference parameter X, W
or Z in order to improve the orthogonality of the equivalent
channel matrix. Therefore we want to derive now the corre-
sponding probability densities of these random variables.
If we have n statistically independent random variables, the
density of the variable
Z =
_

_
X
1
;if |X
1
| = min(|X
1
|, |X
2
|, |X
3
|, |X
4
|)
X
2
;if |X
2
| = min(|X
1
|, |X
2
|, |X
3
|, |X
4
|)
X
3
;if |X
3
| = min(|X
1
|, |X
2
|, |X
3
|, |X
4
|)
X
4
;else
. (25)
is given by [9][p.186] :
f
Z
(z) = n[1 F
X
(z)]
n1
f
x
(z). (26)
With Eqn.(16) and n = 4 we get the PDF of Z:
f
Z
(z) =

3
2
(1 z
2
)

1
3
2
|z|(1
z
2
3
)

3
; |z| 1
0 ; else
(27)
The one sided PDF of the three RVs X, W and Z is
shown in Fig. 3. A comparison with Monte-Carlo simu-
lations also shown in Fig. 3 exhibits excellent agreement
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
1
2
3
4
5
6
x, w, z
P
D
F
|X| analytic
|W| analytic
|W| simulated
|Z| analytic
|Z| simulated
Fig. 3. One sided PDF of X, W, and Z
between simulation results and analytical formulas given in
Eqn.(16),(19) and (27). Obviously, the mean absolut values of
the resulting interference parameters W(E[|W|] = 0.2) and
Z(E[|Z|] = 0.1) are substantially reduced compared to the
original parameter X(E[|X|] = 0.3).
IV. BER Simulation Results
In our simulations, we have used QPSK constellation. A
Rayleigh fading channel was applied being constant during
the transmission of each code block. At the receiver side,
we applied zero forcing (ZF) and maximum likelihood (ML)
receiver types. The BER results were averaged over 2,048
QPSK information symbols and 10
4
realizations of i.i.d.
channel matrices.
Fig. 4 shows the resulting BER as a function of E
b
/N
0
for
ZF and Fig. 5 shows the results for ML receivers. The resulting
curves are compared with ideal two and four path diversity.
Obviously a substantial improvement of the BER can be
achieved by providing only a small partial feedback informa-
tion about the channel to the transmitter enabling this device
to switch between two or four predened code matrices.
V. Conclusion
In this work, the EASTBC scheme for four transmit an-
tennas is combined with limited feedback information sent to
the transmitter. We have shown that this simple transmission
scheme with one or two feedback bits per code block used to
adapt the transmission code to the channel improves diversity
and bit error rate over the whole SNR range compared to the
case of an open loop system. Even with only one bit per code
block fed back, the resulting system achieves a diversity which
is near to the maximum value of four.
15 10 5 0 5 10 15 20 25
10
5
10
4
10
3
10
2
10
1
10
0
ZF Receiver
E
b
/N
0
B
E
R
ideal 4path diversity
ideal 2path diversity

ZF with one bit feedback
ZF without feedback
ZF with two bits feedback
Fig. 4. BER for 4 1 extended Alamouti scheme with feedback applying
a ZF receiver.
15 10 5 0 5 10 15 20 25
10
5
10
4
10
3
10
2
10
1
10
0
E
b
/N
0
B
E
R
ML Receiver
ML with one bit feedback
ML without feedback
ideal 4path diversity
ideal 2path diversity

ML with two bits feedback
Fig. 5. BER for 4 1 extended Alamouti scheme with feedback applying
an ML receiver.
REFERENCES
[1] S.M Alamouti A Simple Diversity Technique for Wireless Communica-
tions, IEEE J. Sel. Ar. Comm., vol.16, no.8, pp. 1451-1458, Oct. 1998.
[2] V.Tarokh, H. Jafarkhani and A.R. Calderbank, Space-Time block codes
from orthogonal designs, IEEE Trans.Inf. Theory, vol.45, pp. 1456-1467,
July 1999.
[3] H. Jafarkhani, A quasi orthogonal space-time block code, IEEE Trans.
Comm., vol. 49, pp. 1-4, Jan. 2001.
[4] A. Narula, M. Lopez, M.Trott and G. Wornell, Efcient use of side
information in multiple antenna data transmission over fading channels,
IEEE J. Sel. Ar. Comm., vol. 16, pp.1423-1436, Oct. 1998.
[5] J.Akhtar, D.Gesbert Partial Feedback Based Orthogonal Block Coding,
Proceedings IEEE Vehicular Technology Conference, 2003, VTC 2003-
Spring. The 57th IEEE Semiannual, Vol.1, April 22-25, 2003.
[6] S. Bhashyam, A. Sabharwal and B. Aazhang, Feedback gain in multiple
antenna systems, IEEE Trans, Comm., vol.48, pp. 83-94, May 2002.
[7] O. Tirkkonen, A. Boariu, A. Hottinen. Minimal non-orthogonality rate 1
space-time block code for 3+ Tx antennas. In Proc. IEEE ISSSTA 2000,
vol. 2, pp. 429-432, Sep. 2000.
[8] C.F.Mecklenbr auker, M.Rupp On Extended Alamouti Schemes for
Space-Time Coding, Proc. WPMC02, Honolulu, Oct. 2002.
[9] A.Papoulis, Probability, Random Variables and Stochastic Processes,
McGraw-Hill, 3.Ed., 1991.