Anda di halaman 1dari 5

Copyright 2002 IEEE, Published in the Proceedings of 5th International Symposium on Wireless Personal

Multimedia Communications, Oct.,27-30, 2002, Hawaii, USA

On Extended Alamouti Schemes for Space-Time Coding


Markus Rupp
Christoph F. Mecklenbrauker
Vienna University of Technology ftw., Forschungszentrum f. Telekommunikation Wien
Gusshausstr. 25/389, A1040 Vienna, Austria Donau-City Str. 1, A1220 Vienna, Austria
mrupp@nt.tuwien.ac.at cfm@ftw.at

Abstract REVIEW OF ALAMOUTI SCHEME


The Alamouti space-time code for transmission diver- A very simple and effective scheme for two transmit antenna-
sity is extended recursively to 2n 2n antenna elements elements and a single receive antenna achieving diversity or-
as well as to unsymmetrical arrays of the size m n. der two was introduced by Alamouti [1, 6]. It sends the
The proposed schemes exhibit two significant advan- sequence {s1 , s2 } on the first antenna and {s2 , s1 } on
tages: they allow for higher diversity (of degree 2n for the other. Assuming a flat-fading channel with coefficients
the symmetrical case) and at the same time can be im- h1 , h2 , the received vector r is formed by stacking two con-
plemented as zero-forcing or MMSE type at low com- secutive data samples [r1 , r2 ]T in time r = Sh + v . Here,
plexity avoiding any matrix inversion due to the struc- the symbol block S and the channel vector h are introduced,
ture in the transmission matrices as well as low com-
plexity ML. s1 s2 h1
S= , h=
s2 s1 h2
Keywords This can be reformulated as
Transmit Diversity, MIMO transmission, STC, Low Com-
plexity Receivers r1 h1 h2 s1 v1
= + (1)
r2 h2 h1 s2 v2
INTRODUCTION or in short notation y = Hs + v, where the vector y =
Alamouti [1] introduced a very simple scheme allowing to [r1 , r2 ]T was introduced. The resulting channel matrix H
transmit at two antennas with the same data rate as on a single is orthogonal, i.e. HH H = HHH = h2 I2 , where I2 is the
antenna but increasing the diversity at the receiver from one 2 2 identity matrix and the gain of the channel is h2 =
to two in a flat-fading channel. While the scheme works |h1 |2 + |h2 |2 . The transmitted symbols can be computed by
for BPSK even with four and eight antennas, it was proven the ZF approach, i.e.
for QPSK that only the two antenna scheme offers the full
1 H
diversity gain [6, 2]. s = H y = s + [HH H]1 HH v . (2)
In this contribution, the Alamouti space-time code for trans- h2
mission diversity is extended recursively to M = 2m antenna The relevant noise variance is given by v2 tr([HH H]1 ) =
elements at the transmitter applying QPSK, the receiver need 2v2 /h2 . The nominator h2 indicates diversity order two for
only use a single antenna, but may use more. While it is true the reception of both symbols. Using unspecified complex-
that the resulting transmission matrix for flat-fading looses its valued modulation, such an improvement is possible only
orthogonality for m 2, it is shown that the loss in orthog- for the two antenna scheme. Higher schemes with 4, and
onality is not severe. Starting with a four antenna scheme in 8 antennas exist only in the case of binary transmission [2].
Section it will be demonstrated that linear receivers perform In UMTS, QPSK is utilized on CDMA preventing perfectly
close to the theoretical bound for four-path diversity offer- orthogonal schemes with code-rate 1. An orthogonal scheme
ing significant gain over the two-antenna case proposed by exists for complex-valued modulation with code-rate 43 and it
Alamouti. Even more interestingly, linear interference sup- is known that a higher code-rate cannot be achieved without
pression can be implemented with low complexity because sacrificing orthogonality [3].
the channel matrix exhibits a high degree of structure en-
abling a factorization in closed-form. Transmission schemes
FOUR ANTENNA SCHEME
with more than one receive antenna will be considered and
it will be shown that even in cases with N 6= 2n receive In the following, a four times diversity scheme for UMTS
antennas, preservation of orthogonality is possible. Variable is proposed1 . It comes with a small caveat which will be
bit-rate services and bursty packet arrivals are handled flexi- explained in the following.
bly in UMTS by dynamically changing the spreading factor
in conjunction with the transmit power, thus preserving an Proposition 1 Starting with the 2 2-Alamouti scheme
average Eb /No , but without changing the diversity order and the following recursive construction rule (similar to the
outage probability. 1 The scheme was already proposed in [4], however only for the
interest of outage capacity and was not considered for further
investigation.
construction of a Walsh-Hadamard code) is applied Performance of linear receivers
Linear receivers suffer of noise enhancement. In this section,
h1 h2 h3 h4
h2 the increased noise caused by the ZF and Minimum Mean
h1 h2 h1 h4 h3

. (3) Squared Error (MMSE) detectors is investigated. Both re-

h2 h1
h3 h4 h1 h2
ceivers can be described by the following detection principle
h4 h3 h2 h1
i.e. the complex scalars h1 and h2 appearing to the left s = (G + I4 )1 z , (9)
of the arrow are replaced by the 2 2 matrices where I4 is the 4 4 identity matrix. We set = 0 for
ZF and = v2 for MMSE. It turns out that both detection
h1 h2 h3 h4
H1 = and H2 = principles have essentially the same receiver complexity.
h2 h1 h4 h3
Lemma 1 Given the 44 Alamouti scheme as described
and then re-inserted into the Alamouti Space-Time chan-
in (3), the eigenvalues i of G in Eq.(6) are given by
nel matrix

H1 H2 1 = 2 = h2 (1 + X), 3 = 4 = h2 (1 X) . (10)
H2 H1 Proof: The Grammian G is diagonalized by VGVT with

where denotes complex conjugation without transpo- the orthogonal matrix
sition.
1 J2 J2
This scheme is implementable via the following symbol V= (11)
2 I2 I2
block S for transmitting the four symbols s = [s1 , . . . , s4 ]T :
.
s1 s2 s3 s4 Some favorable properties are worth mentioning. The eigen-
s2 s1 s4 s3
S= s3
, (4) vectors of G which are stacked in the columns of V do not

s4 s1 s2

depend
s4 s3 s2 s1 on the channel; they are constant. The scaled ma-
trix 2V is sparse, i.e., half of its elements vanish and the
and the element [S]ik is transmitted over the k-th antenna non-zero entries are 1.
in the i-th time step. Assuming a flat-fading channel with Lemma 2 If the channel coefficients hi are i.i.d. com-
coefficient vector h = [h1 , . . . , h4 ]T , the received vector r is plex Gaussian variates then the scaled eigenvalue ui =
formed by stacking four consecutive data samples [r1 , . . . , r4 ]T i /2 is Beta(2,2)-distributed for all i = 1, . . . , 4, i.e.
in time, r = Sh + v . Converting the received vector r by the probability density of the eigenvalues is given by
complex conjugation f4 () = 34 (2 ) for 0 < < 2 and zero elsewhere.
yi = ri , vi = vi , for i = 1, 4 Proof: see the Appendix.
(5)
yi = ri , vi = vi , for i = 2, 3
Let 1 be the following random variable which depends
results in the equivalent transmission scheme y = Hs + v, on the channel gain if > 0,
in which now H appears as the channel transmission matrix. (
is a complex Gaussian vector with i.i.d. elements then so
If v h2 + 1, for ZF,
is v. The resulting matrix H is not unitary anymore, instead = = 2 (12)
h2 1 + hv2 , for MMSE,

I2 XJ2
G = HH H = HHH = h2 (6) For evaluating the BER of the linear receiver for general
XJ2 I2 6= 0,

0 1 tr (G + I4 )1 G(G + I4 )1 =
where J2 = . The gain of the channel is h2 =
1 0
4 1 2 + X 2
|h1 |2 + |h2 |2 + |h3 |2 + |h4 |2 and the channel dependent 2( 1) (13)
,
real-valued random variable X is defined by h2 2 X 2 ( 2 X 2 )2

X = 2 Re(h1 h4 h2 h3 )/h2 . (7) needs to be evaluated which is obtained via



1 I2 XJ2
Applying the matched filter HH , results in the reception of [G + I4 ]1 = 2 2 . (14)
h ( X 2 ) XJ2 I2
the following vector:
When replacing the arguments of the complementary error
s1 + Xs4
s2 Xs3 function with (13), two interpretations can be discussed. In
z = HH y = h2
s3 Xs2 + H v (8)
H h2 the required form for four times diversity is recognized
together with an additional term starting with 1/( 2 X 2 ).
s4 + Xs1
This term can be interpreted as an increase in noise. Alterna-
in which the pair {s1 , s4 } is decoupled from {s2 , s3 } allow- tively, one can interpret the whole expression as defining the
ing for a low complexity solution based on the newly formed true diversity order without noise increase. Both interpreta-
received vector z. tions can be used to describe the schemes performance. The
first interpretation is favoured in this deliverable and leads to 2X Re {[z1 h2 (s1 + Xs4 )][z4 h2 (s4 + Xs1 )]} ,
the following result. 2b = 2b (s2 , s3 ) =
|z2 h2 (s2 Xs3 )|2 + |z3 h2 (s3 Xs2 )|2
Lemma 3 Given the 44 Alamouti scheme as described
in (3), a four times diversity is obtained at the expense +2X Re {[z2 h2 (s2 Xs3 )][z3 h2 (s3 Xs2 )]} .
of a noise enhancement of Note that the two metrics 2a , 2b are positive definite qua-

1 2 + X 2 datics when |X| < 1. Therefore, s1 , s4 can be detected
E 2 2( 1) = regardless of s2 , s3 and vice versa.
X2 ( 2 X 2 )2
3 Simulation Results
2 2 + 2e2 Ei(1, 2) 2 2 + 2 (15)
2 Figure 1 displays the simulated behavior of the uncoded bit
where Ei(n, x) denotes the exponential integral defined error rate of the linear MMSE receiver and vanishing fading
R xt correlation between the four transmit paths. The BER results
for Re (x) > 0 by Ei(n, x) = 1 e tn dt .
were averaged over 16000 symbols and 3200 selections of
Proof: In the case of i.i.d. complex-valued Gaussian dis-
channel matrices H for each simulated Eb /N0 . For compar-
tributed variables hi , the noise enhancement can be evalu-
ison, the BER from the ZF receiver and the cases of ideal
ated in closed form. As shown in the Appendix, the noise
two and fourpath diversity are also shown. The values
is increased by (15). In case of a ZF receiver, the value is
marked by circles () labeled Theory including n.e. of
3/2 which corresponds to 1.76 dB, a value for which the four
1.76 dB are the same as for fourpath diversity, but shifted
times diversity scheme gives much better results as long as
by the noise enhancement (n.e.) of 1.76 dB. Compared to the
Eb /No is larger than about 3 dB.
ZF receiver performance, there is only little improvement for
MMSE.
Therefore, the noise enhancement is maximum for = 0 and
does not exceed 1.76 dB. Further ahead noise enhancement 10
0

will be discussed in a more general form. The behavior of


(15) vs. 1/ (which equals E b /No for the MMSE) is shown 10
1

in Figure 1 indicated by little crosses labelled .


2
10

Maximum-Likelihood Receiver
uncoded bit error rate

The Maximum-Likelihood (ML) receiver selects the s which 10


3

minimizes
1 (s) = kyHsk2 = sH Gs2 Re(yH Hs)+kyk2 (16)
4
10

for all permissible symbol vectors s from the transmitter 10


5
ZF simulated
MMSE simulated

alphabet. We have assumed spatially white interference plus Perfect 4x diversity


Theory including n.e. of 1.76dB
Perfect 2x diversity
noise. In the following we specialize to QPSK which is 10
6

10 5 0 5 10 15 20 25
Eb/N0 in dB
relevant for UMTS. The transmitter alphabet of the Four
Antenna Scheme using QPSK consists of 44 = 256 symbol Figure 1. BER for Four Antenna Scheme with linear
vectors s. MMSE and ZF receiver and no correlation between an-
Alternatively, we can apply the matched filter HH to y and tennas.
implement the ML estimator on its output z given in Eq.(8).
However, we need to take into account that the additive noise For practical considerations, it is of interest to investigate
plus interference is spatially correlated after filtering: As- the performance when the L paths are correlated, as can be
suming v to be zero-mean and spatially white with 4 4 expected in a typical transmission environment. Figure 2 dis-
covariance matrix 2 I4 , gives a w = HH v with covariance plays the situation when the antenna elements are correlated
matrix 2 G. The advantage of this approach is that this by a factor of % {0.5, 0.75, 0.95}. The figure shows that
partly decouples the symbols. The pair {s1 , s4 } is decou- a fading correlation above 0.5 results in some losses. Only
pled from {s2 , s3 } allowing for an ML receiver which needs with very strong correlation (0.95) a degradation towards the
to search over 42 = 16 vector symbols only twice, two-times diversity performance was noticed.
The results for the BER of the ML receiver are given in Fig-
2 (s) = (z Gs)H G1 (z Gs) . (17) ure 3. The BER results were averaged over 12800 QPSK
The metric 2 can be re-written as the sum of two indepen- symbols for each channel matrix H and 8000 selections of
dent terms 2a + 2b which depend on mutually exclusive channel matrices H for each simulated Eb /N0 . For compar-
sets of variables each. ison, the BER from the ZF receiver and the cases of ideal
two and fourpath diversity are also shown. The magnified
2a = 2a (s1 , s4 ) = portion of this plot shows the vicinity around a BER of 10%
|z1 h2 (s1 + Xs4 )|2 + |z4 h2 (s4 + Xs1 )|2 which is where typical packet data services in UMTS FDD
will operate. In this range of Eb /N0 , we see that the ZF This results in a space-time code with
rank matrix A(S, S).
receiver performs rather weakly, i.e. worse than the two- diversity order 4 in flat Rayleigh fading. The corresponding
antenna Alamouti scheme by approximately 0.5 dB. The ML simulation result is shown in Figure 4.
receiver for the Four Antenna Scheme, however, outperforms
the two-antenna Alamouti scheme by 0.1 to 0.5 dB depend- 0
10
Four Antenna Scheme for QPSK with Symbolrotation (4Tx, 1Rx): BER

ing on the chosen point of operation.


0 1
10 10
1
10

uncoded bit error rate


1
10
2
10

1 0 1 2 3 4
2
10
uncoded bit error rate

3
10
3
10
ideal 2path diversity
Four Antenna Scheme (ZF)
Four Antenna Scheme (MMSE)
4 Four Antenna Scheme (ML)
10 4 ideal 4path diversity
10
4path diversity with n.e. 1.76 dB

10 5 0 5 10 15 20 25 30
ZF simulated =0.5 Eb/N0 in dB
10
5
ZF simulated =0.75
ZF simulated =0.95
Perfect 4x diversity Figure 4. BER for Four Antenna Scheme with ML re-
Theory including n.e. of 1.76dB

10
6
Perfect 2x diversity
ceiver, rotated QPSK constellation, no correlation be-
10 5 0 5 10 15 20 25
Eb/N0 in dB
tween antenna elements.
Figure 2. BER for Four Antenna Scheme with Zero-
Forcing receiver, no correlation between adjacent an-
tenna elements is % {0.5, 0.75, 0.95}. THE UNSYMMETRICAL CASE
An interesting approach to increase the diversity when the
number of receiver antennas is more than one, is given by the
10
0
Four Antenna Scheme for QPSK (4Tx, 1Rx): BER
following. Assume that a block matrix form of the channel
matrix H is given by

10
1 H = [H1 H2 ]. (18)
1
10

Then, the scheme can be Alamouted by performing the


uncoded bit error rate

2
following operation:

10

1 0 1 2 3 4
H1 H2
H2 H1
F= H2
(19)
H1
3
10

ideal 2path diversity


Four Antenna Scheme (ZF)
H1 H2
Four Antenna Scheme (ML)

The corresponding term FH F =


10
4 ideal 4path diversity
4path diversity with n.e. 1.76 dB

10 5 0 5 10
Eb/N0 in dB
15 20 25 30

H1 H1 + H2 H2 0
Figure 3. BER for Four Antenna Scheme with ML re- 2 . (20)
0 H1 H1 + H2 H2
ceiver for QPSK, No correlation between antenna ele-
ments. Example: a two transmit-two receive antenna system is con-
sidered. We set
Let us consider the ML receiver for the Four Antenna Scheme
h1 h3
proposed in Sec. for two symbol alphabets in further detail. H1 = , H2 =
h2 h4
For the QPSK symbol alphabet {1, j, 1, j}, there exist
44 (44 1) = 65280 error events S S where we have The matrix FH F becomes

counted only those pairs S, S with S 6= S. It turns out
FH F = 2(|h1 |2 + |h2 |2 + |h3 |2 + |h4 |2 )I2
that exactly 2080 error events are characterized by a rank-
deficient matrix A(S, S) of rank r = 2 instead of 4. The Thus, the full 4 times diversity can be explored, without
corresponding simulation result is shown in Figure 3. a matrix inverse to compute! Note that in this case, the
For a slightly modified QPSK symbol alphabet where the sequence to transmit at the two antennas reads
constellation is rotated by ej/4 for the symbols s2 and s4 ,
s1 s2 s3 s4 s3 s4 s1 s2
i.e.
s3 s4 s1 s2 s1 s2 s3 s4
sk {ej(k3)/4 , ej(k1)/4 , ej(k+1)/4 , ej(k+3)/4 } ,
Note that during 8 time periods only four symbols are trans-
it turns out that all error events are characterized by a full- mitted, i.e., this general scheme has the drawback that it can
only offer half the symbol rate! Many other examples will For the linear receiver with > 0,
be provided in the full paper. " #
1 1+X 1X
E 2 + 2
Consider now a 4 by 2 transmission scheme. The matrices 2 1 + 2 + X 1 + 2 X
h h
are identified to 2
we substitute = h and evaluate the double integral
h1 h2 h3 h4
H1 = , H2 = Z Z1
h5 h6 h7 h8 1+x
fX (x)f () dx d ,
H
We now obtain a matrix F F consisting of two block ma- (1 + + x)2
=0 x=1
trices of size 2 2 on the diagonal. Thus, the scheme is
128 3 4
still rather simple, since only a 2 2matrix is to be inverted where f () = 3 e is the 28 -density re-scaled to
although a fourpath diversity is achieved. Note that now the E [] = 1.
data rate is at full speed!
The previously discussed four antenna system can be ob- CONCLUSION
tained when setting In this article we provided a deep insight into the possibility
of extending the Alamouti scheme for more than two an-
H1 = h1 h2 , H2 = h3 h4
tennas. With some small modifications, the scheme allows
A comparison of the trace term shows that for the 4 2 an- high diversity even for UMTS like scenarios. Extending to
tenna system 3 dB is gained compared to the 4 1 antenna higher than four antennas is straight forward. By structuring
system. the transmitted data, it is possible to utilize much of the ad-
vantages of low complexity receivers without sacrificing too
much in diversity gain. More details of this will be shown
APPENDIX in[5].
Starting from the definition of X in (7) it is observed that the
squares in the denominator can be appended, ACKNOWLEDGMENT
2 2 The authors thank Ralf Muller (ftw.) for valuable comments
|h1 + h4 | + |h2 h3 |
X +1= . and improving the manuscript.
|h1 |2 + |h2 |2 + |h3 |2 + |h4 |2
In the case of i.i.d. complex-valued Gaussian distributed REFERENCES
variables hi , the two variates h1 + h4 and h2 + h3 become
[1] Alamouti, S.M., A simple transmitter diversity sceme
complex Gaussian distributed and independent of each other.
for wireless communications. IEEE J. Sel. Ar. Comm.
They depend, however, on the variates in the nominator.
16(8):14511458, Oct. 1998.
Now, a linear orthogonal coordinate transformation
[2] Hochwald, B., Marzetta, T., and Papadias, C., A
u = (h1 + h4 )/ 2, v = (h2 h3 )/ 2, transmitter diversity scheme for wideband CDMA sys-
u0 = (h1 h4 )/ 2, v 0 = (h2 + h3 )/ 2 . tems based on spacetime spreading. IEEE J. Sel. Ar.
is defined and in the new variables, Comm. 19(1):4860, Jan. 2001.
[3] Calderbank, A. and Diggavi, S., Tutorial on space-
X +1 |u|2 + |v|2 time coding and signal processing for high data rate wire-
= . (21)
2 |u|2 + |v|2 + |u0 |2 + |v 0 |2 less communications. In Proc. Int. Z urich Seminar,
Generally, if X12 and X22 are independent random vari- IZS02, Zurich, Switzerland, Feb. 2002.
ables following chi-square distributions with 1 and 2 de- [4] Papadias, C. and Foschini, G., A space-time coding
X12 approach for systems employing four transmit antennas.
grees of freedom respectively, then X 2 +X 2 is said to fol-
1 2 In Proc. IEEE ICASSP (2001), Vol. 4, pp. 24812484.
low a Beta(p,q) distribution with 1 = 2p and 2 = 2q [5] Rupp, M. and Mecklenbrauker, C.F., On ex-
degrees of freedom and the probability density is given by tended alamouti schemes for space-time coding. In
1
B(p,q)
p1
(1 )q1 , with p = 21 , q = 22 . This matches Proc. Wireless Personal Multimedia Communica-
our case (21) for 1 = 2 = 4 and the probability density tions (WPMC02) Honolulu, Hawaii, Oct. 2730,
specializes to 6(1 ). Transforming this back to X gives 2002.
the probability density [6] Tarokh, V., Seshadri, N., and Calderbank, A.,
3 2
4 (1 x ), for |x| < 1, Space-time codes for high data rate wireless commu-
fX (x) = (22)
0, elsewhere. nication: Performance criterion and code construction.
IEEE Trans. Inf. Theor. 44(2):744765, Mar. 1998.
For the ZF receiver, it follows that the noise is increased by
a factor of
Z 2
1 3
E = 1 fX ( 1) d = .
1 X2 0 2