SPACE TIME CODING FOR MIMO SYSTEMS

Fernando H. Gregorio

Helsinki University of Technology

Signal Processing Laboratory,
POB 3000, FIN-02015 HUT, Finland
E-mail:[email protected]

ABSTRACT 2. DIVERSITY

With space-time codes (STC) the same information is trans- Depending on surrounding environment, a transmitted ra-
mitted in appropriate manner simultaneously from different dio signal propagates through several different paths. This
transmit antennas to obtain transmit diversity. The main phenomenon is often referred as multipath propagation [3].
idea of transmit diversity is that if a message is lost in a The signal received by the receiver antenna consists of the
channel with probability p and we can transmit replicas of superposition of various multipaths. If there is Non-Line
the message over n independent such channels, the loss of Sight components between the transmitter and receiver ,
probability becomes p n . Using diversity, more reliability is the attenuation coefficients corresponding to different paths
given to the symbols which allows employing higher order are assumed to be independent and identically distributed.
constellation resulting in higher throughput. In which case the central limit theorem applies and the re-
sulting path can be modeled as a complex Gaussian random
variable. In this case, the channel is said to be Rayleigh.
1. INTRODUCTION Signal power in a wireless system fluctuates. When this
signal power drops significantly, the channel is said to be
in fade. Diversity is used in wireless channels to combat
Multiple-input multiple-output (MIMO) communication tech-
the fading. Receive diversity and transmit diversity mitigate
nology has received significant recent attention due to the
fading and significantly improve link quality. The receive
rapid development of high-speed broadband wireless com-
antennas see independently faded versions of the same sig-
munication systems employing multiple transmit and receive
nals. The receiver combines these signals so that the resul-
antennas. Information theoretic results show that MIMO
tant signal exhibits considerably reduced amplitude fading.
systems can offer significant capacity gains over traditional
single-input single-output channels. In most scattering environments, antenna diversity is a
practical, effective and, hence, a widely applied technique
This increase in capacity is enabled by the fact that in for reducing the effect of multipath fading . The classical
rich scattering wireless environments, the signals from each approach is to use multiple antennas at the receiver and per-
individual transmitter appear highly uncorrelated at each of form combining or selection and switching in order to im-
the receive antennas. When conveyed through uncorrelated prove the quality of the received signal. The major problem
channels between the transmitter and the receiver, the sig- with using the receive diversity approach is the cost, size,
nals corresponding to each of the individual transmit anten- and power of the remote units. The use of multiple anten-
nas have attained different spatial signatures. The receiver nas and radio frequency (RF) chains makes the remote units
can exploit these differences in spatial signatures to separate larger and more expensive. As a result, diversity techniques
the signals originated from different transmit antennas. have almost exclusively been applied to base stations to im-
In Space-Time Coding (STC) systems , the same infor- prove their reception quality. A base station often serves
mation symbol stream is transmitted from different transmit hundreds to thousands of remote units. It is therefore more
antennas in appropriate manner to obtain transmit diversity. economical to add equipment to base stations rather than the
This paper is organized as follows. In Section 2 diver- remote units. For this reason, transmit diversity schemes are
sity concepts are introduced. In Section 3,4 and 5 space- very attractive.
time coding techniques are discussed. Error probability is In recent years it has been realized that many of the ben-
introduced in section 6. Differential STBC (DSTBC) is pre- efits as well as substantial amount of the performance gain
sented in section 7. STC for OFDM are included in the sec- of receive diversity can be reproduced by using multiple an-
tion 8. Finally conclusions and references. tennas at the transmitter to achieve transmit diversity.
 The use of transmit diversity at the base stations in a cel- single antenna case, the error probability of detecting
lular o wireless local area network has atracted an special in- s decay as SN R−1 . The diversity order of the system
terest; this is so primarily because a performance increase is is the slope of the BER curved plotted versus the av-
possible without adding extra antennas, power consumption erage SNR. In our example diversity gain equal to 2
or significant complexity to the mobile. Also, the cost of was reached.
the extra transmit antenna at the base station can be shared
among all users. • Two transmit antennas and one receive antenna At a
given time instant, the symbol s pre-weighted with
3. SPACE TIME CODING w1 and w2 is transmitted. The received sample can
be written
An effective and practical way to approaching the capac-
ity of MIMO wireless channels is to employ Space-Time y = h1 w1 s + h2 w2 s + n (3)
Coding (STC). STC is a coding technique designed to be
used with multiple transmit antennas. Coding is performed where n is the noise sample. The SNR in y is given
in both spatial and temporal domains to introduce corre- by
lation between signals transmitted for various antennas at
various time periods [2]. Space-Time coding can achieve
transmit diversity and power gain over spatially uncoded |h1 w1 + h2 w2 |2
SN R = E[|s|2 ] (4)
systems without sacrificing the bandwidth. There several σ2
approaches in coding structures, Space-Time Block Coding
(STBC), Space Time Trellis Coding (STTC), Differential If w1 and w2 are fixed , the SNR has the same sta-
Space-Time Block Coding (DSTBC) that will be presented tistical distribution as |h1 |2 ( or |h2 |2 ). Therefore,
in the next sections. if the weights are not allowed to depend on h 1 and
h2 it is impossible to achieve diversity of order two.
3.1. STC - An introductory example However, if we assume that the transmitter knows the
channel, and w 1 and w2 are chosen to be functions of
• One transmit antenna and two receive antennas h1 and h2 , it is possible to achieve an error probabil-
Considering a system with two receive antennas and ity that behaves SN R −2 .
one transmit antenna. If the fading is frequency flat, We have seen that without channel knowledge at the
the twor received signals can be written as transmitter, diversity can not be achieved. However, if
two time intervals for the transmission is allowed, we
y 1 = h1 s + n1 can achieve diversity of order two easily. At Time=t,
y 2 = h2 s + n2 (1) antenna 1 is used and at Time=t+1 antenna 2 is used.
In this case, the received samples signal at different
where h1 and h2 are the channel gains, s is the trans-
time instants are
mitted signal and n 1 ,n2 are mutually uncorrelated
noise terms. By the following linear combination, it y1 = h1 s + n1
is possible to recover s
y2 = h2 s + n2 (5)
ŝ = w1 ∗ y1 + w2 ∗ y2
= (w1∗ h1 + w2∗ h2 ) + w1∗ n1 + w2∗ n2 (2) Equation 5 is of the same form than equation 1, so
that the error associated with this method is equal to
where w1 and w2 are the weights. The SNR in ŝ is that for the previous case ( 1 × 2 system). In this case
given by diversity gain equal to 2 is achieved but a sacrifice in
information rate is necessary.

|w1∗ h1 + w2∗ h2 |2 Space Time coding is concerned with the harder and in-
SN R = E[|s|2 ] teresting topic, How maximize the transmitted information
(|w1 |2 + |w2 |2 )σ 2
rate at the same time that minimize the error probability.
where σ 2 is the power noise. We can choose w 1 and 4. ALAMOUTI CODE
w2 proportional to h 1 and h2 that maximizes the SNR.
The resulting SNR is proportional to (|h 1 |2 + |h2 |2 ) Figure 1 shows the baseband representation of the Alamouti
If the fading is Rayleigh, then (|h 1 |2 + |h2 |2 ) is χ dis- two branch transmit diversity scheme [1]. The scheme uses
tributed, and we can show that the error probability of two transmit antennas and one receive antenna and may be
detecting s decay as SN R a−2 in high SNR values. In defined by the following three functions
• The encoding and transmission sequence of informa-
tion
At a given symbol period, two signals are simultane-
ously transmitted from the two antennas. The signal
transmitted from antenna zero is denoted by S 0 and
from antenna one by S 1 . During the next symbol pe-
riod signal S1∗ is transmitted from antenna zero, and
signal S0∗ is transmitted from antenna one. The en-
coding, however, may also be done in space and fre-
quency. Instead of two adjacent symbol periods, two
adjacent carriers may be used (spacefrequency cod-
ing). The 2 × 2 space time code is written in matrix
form as

Fig. 1. Alamouti transceiver structure


s0 s1
S= (6)
−s∗1 s∗0

The channel at time t may be modeled by a complex

multiplicative distortion h 0 (t) for transmit antenna
zero and h0 (t) for transmit antenna one. Assuming
that fading is constant across two consecutive sym-
bols, we can write

h0 (t) = h0 (t + T ) = α0 ejθ0
h1 (t) = h1 (t + T ) = α1 ejθ1 (7)

The received signal can be written

r0 = r(t) = h0 s0 + h1 s1 + n0
r1 = r(t + T ) = −h0 s∗1 + h1 s∗0 + n1 (8) Fig. 2. BER Alamouti code compared with MRCC system

where r0 and r1 are the received signals at time t and

The resulting combined signals in 11 are equivalent
t + T.
to that obtained from two-branch MRRC. The only
• The combining scheme at the receiver difference is phase rotations on the noise components
which do not degrade the effective SNR.
The combiner builds the following two combined sig-
nals that are sent to the maximum likelihood detector
Therefore, the resulting diversity order from the Alam-
outi transmit diversity scheme with one receiver is
ŝ0 = h∗0 r0 + h1 r1∗ equal to that of two-branch MRRC.
ŝ1 = h∗1 r0 − h0 r1∗ (9)
Figure 2 shows the performance of Alamouti code com-
The decision statistics can be expressed as
pared with a MRCC system in a Rayleigh fading channel
using BPSK modulation. From the figure, we can conclude
that the performance of Alamouti code with two transmit-
s0 = (α20 + α21 )s0 + h∗0 n0 + h1 n∗1 (10) ters and a single receiver is 3 dB worse than two-branch
s1 = (α20 + α21 )s1 − h0 n∗1 + h∗1 n0 (11) MRRC. The 3-dB penalty is incurred because is assumed
that each transmit antenna radiates half the energy in order
• The decision rule for maximum likelihood detection to ensure the same total radiated power as with one transmit
These combined signals are then sent to the maximum antenna. If each transmit antenna was to radiate the same
likelihood detector. energy as the single transmit antenna for MRRC, the per-
formance would be identical
 5. GENERALIZED STBC 6. ERROR PROBABILITY ON SLOW FADING
CHANNELS
The Alamouti scheme works only with two transmit anten-
nas. This scheme was later generalized in to an arbitrary On slow fading channels, the fading coefficients within each
number of transmit antennas. Similarly to the Alamouti frame are constant. So we can ignore the subscript of the
code, the general STBC is defined by a code matrix with or- fading coefficients
thogonal columns. Just like in the Alamouti scheme, a sim-
ple linear receiver is also obtained due to the orthogonality h1j,i = h2j,i = . . . = hL j,i = hj,i
of the columns of the code matrix. In general, an STBC is i = 1, 2, . . . , nT , i = j, 2, . . . , nR (14)
defined by a (p × n T ) matrix G. The entries of the matrix G
are linear (possibly complex) combinations of the variables Let us define a codeword difference matrix B(X, X̂) as
x1 ; x2 ; . . . ; xk (representing symbols). The columns of the  
matrix represent antennas and the rows time slots [2]. x11 − x̂11 x12 − x̂12 ... x1L − x̂1L
Therefore, p time slots are needed to transmit k sym-  x21 − x̂21 x22 − x̂22 ... x2L − x̂2L 
 
bols, resulting in a code rate R = k/p. B=  .. .. .. .. (15)
 . . . . 
It is of special interest code matrices achieving the max-
xnT nT
1 − x̂1 xnT nT
2 − x̂2 . . . xnT nT
L − x̂L
imum transmission rate permitted by the STC theory, i.e, R
= 1 symbol/channel use. For a fixed n T , among the code (16)
matrices that achieve the maximum rate, we will be inter-
ested in those with minimum values of p or equivalently, We can construct an n T × xnT codeword distance matrix
minimum number of time slots needed to transmit a block. A(X, X̂) defined as
These code matrices are referred as delay optimal and they
are interesting because they minimize the memory require- A(X, X̂) = B((X, X̂).B H (X, X̂)
ments at the transmitter and at the receiver [5].
Exist a unitary matrix V and a real diagonal matrix ∆ such
The construction of STBC using the generalized com-
plex orthogonal for a rate equal to 1/2 give the following that
matrix:   V A(X, X̂)V H = ∆
s1 s2 s3
 −s2 s1 s4 
  where the rows of V are the eigenvectors of A(X, X̂),
 −s3 s4 s1 
  forming a complete orthogonal basis of an N-dimensional
 −s −s3 s2 
G3 =  ∗4 ∗ ∗  (12) vector space. The diagonal elements of ∆ are the eigenval-
 s1 s2 s3 
 −s∗ s∗ s∗  ues λ1 > 0, i = 0, 1, . . . , nT of A(X, X̂).
 2 1 4 
 −s∗ s∗ s∗  The modified Euclidean distance between the two space-
3 4 1
times codeword matrices X and X̂ can be written as
−s∗4 −s∗3 s∗2
nR

G3 code is designed for 3 transmit antennas with a code d2h (X, X̂) = hj A(X, X̂)hH
j
rate R = 1/2 and G4 is designed for 4 antennas with the j=1
same code rate.   nR
nT
s1 s2 s3 s4

= λi |βj,i |2 (17)
 −s2 s1 s4 s3 
  j=1 i=1
 −s3 s4 s1 −s2 
 
 −s −s3 s2 s1  where βj,i = hj .vi and . denotes the inner product. In the
G4 =  ∗4 ∗ ∗ ∗  (13)
 s1 s s s  case of Rayleigh fading and high SNR, the upper bound of
 −s∗ s2∗ s3∗ s4∗ 
 2 1 4 3  the error probability is [2]
 −s∗ s∗ s∗ −s∗ 
3 4 1 2
 −nR
−s∗4 −s∗3 s∗2 s∗1 r
 Es −rnR
P (X, X̂) ≤ λi ( ) (18)
Compared with Alamouti code, the STBC G 3 and G4
i=1
4N0
have two disadvantages: The bandwidth efficiency is re-
duced by a factor of two. The number of time slots across where r denotes the rank of the matrix A(X, X̂) and λi are
which the channel is required to have a constant fading en-
the nonzero eigenvalues of matrix A(X, X̂). The exponent
velope is increased by a factor of four. Figure 3 shows the
of the SNR term, rnR is called diversity order and the prod-
BER performance of G 3 ,G4 and G2 (Alamouti). 16 QAM
uct of the eigenvalues is called coding gain.
modulation is used for G 3 and G4 and 4 DPSK modulation
When a space time code is designed, the following cri-
for Alamouti in order to have the same data rate (2 bits/sec)
teria can be considered:
[6]. BER results for 1 bit/sec is shown in figure 4.
 Fig. 3. BER G3 ,G4 and G2 codes - 2 bits/sec Fig. 4. BER G3 ,G4 and G2 codes - 1 bits/sec

• Maximize the minimum rank r of the matrix A(X, X̂) The transmitter then sends s 2t+1 and s2t+2 , respectively,
over all pairs of distinct codewords. from transmit antennas one and two at time 2t + 1 and
 −s∗2t+2 and s∗2t+1 from antennas one and two at time 2t+ 2.
• Maximizes the minimum product ri=1 λi of matrix This process is inductively repeated until the end of the
A(X, X̂) along the pairs of distinct codewords with frame. Block diagram of the differential encoder and de-
the minimum rank. coder is shown in Figure 5.
Example
√ Consider √a BPSK constellation of two signal
7. DIFFERENTIAL STBC points 1/ 2 and −1/ 2. The coefficient vector set is given
by
When the channel changes slowly compared to the symbol V = [(1, 0), (0, 1), (−1, 0), (0, −1)] (19)
rate, the transmitter can send pilot sequences that enable At each encoding operation, a block of 2m = 2 bits arrives
the receiver to estimate the channel accurately. However, in at the encoder and is mapped into V . The mapping function
some situations, such as high mobility channels or channel can be computed using
fading conditions changing rapidly, it may be difficult to es-
timate the channel with good precision. For such situations, M (00) = (1, 0)
it is useful to develop space time codes that do not requires M (10) = (0, 1)
channel estimates either at the receiver or at the transmitter
M (01) = (0, −1)
[7].
M (11) = (−1, 0)
7.1. Encoding Algorithm √
Assuming
√ that at time 2t − 1, x2t−1 = −1/ 2 and x2t =
The transmitter begins the transmission with sending arbi- −1/ 2 are sent from antennas √ one and two, respectively,
√
∗
∗
trary symbols s 1 and s2 at time 1 and symbols −s 2 and and at time 2t, −x 2t = 1/ 2 and x∗2t−1 = −1/ 2 are sent
s∗1 at time 2 unknown to the receiver. These two transmis- from antennas one and two. If the two information bits at
sions do not convey any information. A mapping function the encoder input at time 2t + 1 are 11, according to the
M is defined The transmitter subsequently encodes the rest mapping function M (11) = (−1, 0), the coefficients used
of the data in an inductive manner. Suppose that s 2t−1 and to compute the transmitted signals for the next two trans-
s2t are sent, respectively, from transmit antennas one and missions are R1 = −1 and R2 = 0. Thus, we have
∗ ∗
two at time 2t − 1, and that −s 2t and s2t−1 , are sent re- √ √ √ √
x2t+1 , x2t+2 = −1(−1/ 2, −1/ 2) + 0(+1/ 2, −1/ 2)
spectively, from antennas one and two at time 2t . At time √ √
2t + 1, a block of 2b bits B 2t+1 arrives at the encoder. = −1(+1/ 2, +1/ 2) (20)
The transmitter uses the mapping function M and computes √ √
M(B2t+1 ) = A(B2t+1 ) + B(B2t+1 ) Then it computes At time 2t + 1, x2t+1 = +1/ 2 and x2t+2 = −1/√2
are sent from antennas one and two and −x ∗2t+2 = −1/ 2
√
(s2t+1 s2t+2 ) = A(B2t−1 )(s2t−1 s2t ) + B(B2t+1 )(−s∗2t s2t−1 ) and x∗2t+1 = +1/ 2 at time 2t + 2.
 In [2], details related with decoding process can be found. Differential encoder
From [1]
The DSTBC detection scheme is 3 dB worse than that of the St,t+1
Z-2 I
transmit diversity scheme of employs coherent detection at St+2,t+3
high SNR [7]. However, it is an excellent option in high Mapper
mobility environments like cellular systems. Incoming
At+2,t+3
pairs of
bits
8. OFDM AND ST CODES Compute the matrix A that is
t+2, t+3 closest to Yin terms Euclidean
xt,t+1 distance
Space-time (ST) coding has been proved effective in com- Z-2 I
Compute
bating fading, and enhancing data rates. Exploiting the pres- closest Inverse
Estimate of
matrix to Mapper
ence of spatial diversity offered by multiple transmit and/or xt+2,t+3 A(t+2,t+3)
original pairs
of data bits
receive antennas, ST coding relies on simultaneous coding yt+2,t+3
Estimate of
At+2,t+3 transmitted at
across space and time to achieve diversity gain without nec- t+2 and t+3

essarily sacrificing bandwidth. Two typical examples of ST

codes are ST trellis codes and ST block codes .
Fig. 5. Differential Encoder - Decoder
Multipath diversity becomes available when frequency
selectivity is present, which is the typical situation for broad-
The technique can be generalized to extract spatial di-
band wireless channels. Multiantenna transmissions over
versity over a large number of antennas by using STBC
frequency-selective fading channels can potentially provide
techniques. In this case, we need a block size T ≥ M T
a maximum diversity gain that is multiplicative in the num-
and the channel must be identical over the T tones.
ber of transmit antennas, receive antennas, and the channel
length. Inspired by this result, a number of coding schemes
have been proposed recently to exploit multipath diversity. 9. CONCLUSIONS
Because they offer low-complexity equalization decoding
and facilitate the support of multirate services, multicarrier As conclusion of this paper we can mention:
transmissions are typically adopted by those schemes. • Alamouti code is the best option when 2 Transmis-
Diversity techniques designed for single carrier (SC) mod- sion antennas is considered.
ulation are easily extended to OFDM modulation with the
time index for SC modulation replaced by the tone index • Low complexity receiver is a good characteristic for
in OFDM. For example, considering the Alamouti scheme STBC.
which requires that the channel remains constant over con-
secutive symbols periods. In the OFDM context, this trans- • DSTBC can be considered in high mobility channels.
lates to the channel remaining constant over consecutive
tones, i.e, H(k) = H(k + 1) [4]. Consider two data sym-
bols, s1 and s2 , to be transmitted over antennas 1 and 2 10. REFERENCES
respectively on tone k, and s ∗2 and s∗1 are transmitted over
antennas 1 and 2 respectively on tone k + 1 within the same [1] Simon Haykin and Michael Moher, Modern Wireless Commu-
nications.
OFDM symbol.
The receiver detects the transmitted symbols from the [2] B. Vucetic and J. Yuan, Space-Time Coding, John Wiley,
received signal on the two tones using the Alamouti detec- 2003.
tion techniques. As in SC modulation, the effective channel
is orthogonalized irrespective of the channel realization and [3] P. Stoica and E. Larson, Space-Time Block Coding for Wire-
less Communications, Cambridge University Press, 2003
the vector detection problem collapses into scalar detection
problems with the effective input-output relation for sym- [4] A. Paulraj, R. Nabar, D. , Introduction to Space-Time Wire-
bols is given by less Communications, Cambridge University Press, 2003

Es 2
[5] D. Gespert, et. Al. ,”From theory to practice: An overview of
yi = H[k]f si + n1 , i = 1, 2 (21) MIMO Space-Time Coded Wireless Systems”, IEEE JSAC,
2
Vol. 21, April 2003
2
where ni is a noise component with variance H[k] f N0 . [6] Tarokh, V.; Jafarkhani, H., ”A differential detection scheme
Assuming that the 2M R elements of H[k] undergo inde- for transmit diversity”’, IEEE JSAC ,Vol. 18 ,July 2000 .
pendent fading, the Alamouti scheme extracts 2M R order
diversity, just as SC modulation. [7] Tarokh, V.; Jafarkhani, H.; Calderbank, A.R.; ”Space-time
The use of consecutive tones is not strictly necessary, block coding for wireless communications: performance re-
sults”, IEEE JSAC ,Vol. 17 ,March 1999
any pair of tones can be used as long as the associated chan-
nel are equal.