IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 59, NO.

3, MARCH 2011 823

Space-Time Block Coded Spatial Modulation

Ertuğrul Başar, Student Member, IEEE, Ümit Aygölü, Member, IEEE, Erdal Panayırcı, Fellow, IEEE,
and H. Vincent Poor, Fellow, IEEE

Abstract—A novel multiple-input multiple-output (MIMO) systems, a high level of inter-channel interference (ICI) occurs
transmission scheme, called space-time block coded spatial modu- at the receiver since all antennas transmit their own data
lation (STBC-SM), is proposed. It combines spatial modulation streams at the same time. This further increases the complexity
(SM) and space-time block coding (STBC) to take advantage of
the benefits of both while avoiding their drawbacks. In the STBC- of an optimal decoder exponentially, while low-complexity
SM scheme, the transmitted information symbols are expanded suboptimum linear decoders, such as the minimum mean
not only to the space and time domains but also to the spatial square error (MMSE) decoder, degrade the error performance
(antenna) domain which corresponds to the on/off status of the of the system significantly. On the other hand, STBCs offer
transmit antennas available at the space domain, and therefore an excellent way to exploit the potential of MIMO systems
both core STBC and antenna indices carry information. A
general technique is presented for the design of the STBC-SM because of their implementation simplicity as well as their low
scheme for any number of transmit antennas. Besides the high decoding complexity [3], [4]. A special class of STBCs, called
spectral efficiency advantage provided by the antenna domain, orthogonal STBCs (OSTBCs), have attracted attention due
the proposed scheme is also optimized by deriving its diversity to their single-symbol maximum likelihood (ML) receivers
and coding gains to exploit the diversity advantage of STBC. with linear decoding complexity. However it has been shown
A low-complexity maximum likelihood (ML) decoder is given
for the new scheme which profits from the orthogonality of that the symbol rate of an OSTBC is upper bounded by 3/4
the core STBC. The performance advantages of the STBC-SM symbols per channel use (pcu) for more than two transmit
over simple SM and over V-BLAST are shown by simulation antennas [5]. Several high rate STBCs have been proposed
results for various spectral efficiencies and are supported by the in the past decade (see [6]-[8] and references therein), but
derivation of a closed form expression for the union bound on their ML decoding complexity grows exponentially with the
the bit error probability.
constellation size, which makes their implementation difficult
Index Terms—Maximum likelihood decoding, MIMO systems, and expensive for future wireless communication systems.
space-time block codes/coding, spatial modulation. Recently, a novel concept known as spatial modulation (SM)
has been introduced by Mesleh et al. in [9] and [10] to
I. I NTRODUCTION remove the ICI completely between the transmit antennas of

T HE use of multiple antennas at both transmitter and a MIMO link. The basic idea of SM is an extension of two
receiver has been shown to be an effective way to im- dimensional signal constellations (such as 𝑀 -ary phase shift
prove capacity and reliability over those achievable with single keying (𝑀 -PSK) and 𝑀 -ary quadrature amplitude modulation
antenna wireless systems [1]. Consequently, multiple-input (𝑀 -QAM), where 𝑀 is the constellation size) to a third di-
multiple-output (MIMO) transmission techniques have been mension, which is the spatial (antenna) dimension. Therefore,
comprehensively studied over the past decade by numerous the information is conveyed not only by the amplitude/phase
researchers, and two general MIMO transmission strategies, modulation (APM) techniques, but also by the antenna indices.
a space-time block coding1 (STBC) and spatial multiplexing, An optimal ML decoder for the SM scheme, which makes an
have been proposed. The increasing demand for high data rates exhaustive search over the aforementioned three dimensional
and, consequently, high spectral efficiencies has led to the de- space has been presented in [11]. It has been shown in [11] that
velopment of spatial multiplexing systems such as V-BLAST the error performance of the SM scheme [9] can be improved
(Vertical-Bell Lab Layered Space-Time) [2]. In V-BLAST approximately in the amount of 4 dB by the use of the
optimal detector under conventional channel assumptions and
Paper approved by H. Leib, the Editor for Communication and Information that SM provides better error performance than V-BLAST and
Theory of the IEEE Communications Society. Manuscript received March 15,
2010; revised August 12, 2010. maximal ratio combining (MRC). More recently, Jeganathan
This paper was presented in part at the IEEE International Symposium et al. have introduced a so-called space shift keying (SSK)
on Personal, Indoor and Mobile Radio Communications (PIMRC), Istanbul, modulation scheme for MIMO channels in [12]. In SSK
Turkey, September 2010.
E. Başar and Ü. Aygölü are with Istanbul Technical University, Faculty
modulation, APM is eliminated and only antenna indices are
of Electrical and Electronics Engineering, 34469, Maslak, Istanbul, Turkey used to transmit information, to obtain further simplification in
(e-mail: {basarer, aygolu}@itu.edu.tr). system design and reduction in decoding complexity. However,
E. Panayırcı is with Kadir Has University, Department of Electronics
Engineering, 34083, Cibali, Istanbul, Turkey (e-mail: [email protected]).
SSK modulation does not provide any performance advantage
H. V. Poor is with the Department of Electrical Engineering, Princeton compared to SM. In both of the SM and SSK modulation
University, Princeton, NJ, 08544, USA (e-mail: [email protected]). systems, only one transmit antenna is active during each
This work was supported in part by the U. S. National Science Foundation
under Grant CNS-09-05398.
transmission interval, and therefore ICI is totally eliminated.
Digital Object Identifier 10.1109/TCOMM.2011.121410.100149 SSK modulation has been generalized in [13], where different
1 The abbreviation "STBC(s)" stands for space-time block coding/code(s)
combinations of the transmit antenna indices are used to
depending on the context.
0090-6778/11$25.00 ⃝
c 2011 IEEE
824 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 59, NO. 3, MARCH 2011

convey information for further design flexibility. Both the integer less than or equal to 𝑥, that is an integer power of 2.
SM and SSK modulation systems have been concerned with 𝛾 denotes a complex signal constellation of size 𝑀 .
exploiting the multiplexing gain of multiple transmit antennas,
but the potential for transmit diversity of MIMO systems is not II. S PACE -T IME B LOCK C ODED S PATIAL M ODULATION
exploited by these two systems. This leads to the introduction (STBC-SM)
here of Space-Time Block Coded Spatial Modulation (STBC- In the STBC-SM scheme, both STBC symbols and the
SM), designed to take advantage of both SM and STBC. indices of the transmit antennas from which these symbols are
The main contributions of this paper can be summarized as transmitted, carry information. We choose Alamouti’s STBC,
follows: which transmits one symbol pcu, as the core STBC due to
∙ A new MIMO transmission scheme, called STBC-SM, its advantages in terms of spectral efficiency and simplified
is proposed, in which information is conveyed with an ML detection. In Alamouti’s STBC, two complex information
STBC matrix that is transmitted from combinations of the symbols (𝑥1 and 𝑥2 ) drawn from an 𝑀 -PSK or 𝑀 -QAM
transmit antennas of the corresponding MIMO system. constellation are transmitted from two transmit antennas in
The Alamouti code [3] is chosen as the target STBC two symbol intervals in an orthogonal manner by the codeword
( )
to exploit. As a source of information, we consider not ( ) 𝑥1 𝑥2
X = x1 x2 = (1)
only the two complex information symbols embedded in −𝑥∗2 𝑥∗1
Alamouti’s STBC, but also the indices (positions) of the where columns and rows correspond to the transmit antennas
two transmit antennas employed for the transmission of
and the symbol intervals, respectively. For the STBC-SM
the Alamouti STBC.
scheme we extend the matrix in (1) to the antenna domain.
∙ A general technique is presented for constructing the Let us introduce the concept of STBC-SM via the following
STBC-SM scheme for any number of transmit antennas.
simple example.
Since our scheme relies on STBC, by considering the Example (STBC-SM with four transmit antennas, BPSK modu-
general STBC performance criteria proposed by Tarokh et lation): Consider a MIMO system with four transmit antennas
al. [14], diversity and coding gain analyses are performed
which transmits the Alamouti STBC using one of the follow-
for the STBC-SM scheme to benefit the second order ing four codewords:
transmit diversity advantage of the Alamouti code. {( ) ( )}
∙ A low complexity ML decoder is derived for the proposed 𝑥1 𝑥2 0 0 0 0 𝑥1 𝑥2
𝜒1 = {X11 , X12 } = ,
STBC-SM system, to decide on the transmitted symbols −𝑥∗ 𝑥∗ 0 0 0 0 −𝑥∗2 𝑥∗1
{( 2 1 ) ( )}
as well as on the indices of the two transmit antennas 0 𝑥1 𝑥2 0 𝑥2 0 0 𝑥1
𝜒2 = {X21 , X22 } = , ∗ 𝑒𝑗𝜃
that are used in the STBC transmission. 0 −𝑥∗2 𝑥∗1 0 𝑥1 0 0 −𝑥∗2
∙ It is shown by computer simulations that the proposed (2)
STBC-SM scheme has significant performance advan-
where 𝜒𝑖 , 𝑖 = 1, 2 are called the STBC-SM codebooks each
tages over the SM with an optimal decoder, due to its
containing two STBC-SM codewords X𝑖𝑗 , 𝑗 = 1, 2 which
diversity advantage. A closed form expression for the
do not interfere
∪ to each other. The resulting STBC-SM code
union bound on the bit error probability of the STBC-
is 𝜒 = 2𝑖=1 𝜒𝑖 . A non-interfering codeword group having
SM scheme is also derived to support our results. The
𝑎 elements is defined as a group of codewords satisfying
derived upper bound is shown to become very tight with
X𝑖𝑗 X𝐻 𝑖𝑘 = 02×2 , 𝑗, 𝑘 = 1, 2, . . . , 𝑎, 𝑗 ∕= 𝑘; that is they have
increasing signal-to-noise (SNR) ratio.
no overlapping columns. In (2), 𝜃 is a rotation angle to be
The organization of the paper is as follows. In Section II, we optimized for a given modulation format to ensure maximum
introduce our STBC-SM transmission scheme via an example diversity and coding gain at the expense of expansion of the
with four transmit antennas, give a general STBC-SM design signal constellation. However, if 𝜃 is not considered, over-
technique for 𝑛𝑇 transmit antennas, and formulate the optimal lapping columns of codeword pairs from different codebooks
STBC-SM ML detector. In Section III, the performance analy- would reduce the transmit diversity order to one. Assume
sis of the STBC-SM system is presented. Simulation results now that we have four information bits (𝑢1 , 𝑢2 , 𝑢3 , 𝑢4 ) to be
and performance comparisons are presented in Section IV. transmitted in two consecutive symbol intervals by the STBC-
Finally, Section V includes the main conclusions of the paper. SM technique. The mapping rule for 2 bits/s/Hz transmission
Notation: Bold lowercase and capital letters are used for is given by Table I for the codebooks of (2) and for binary
∗ 𝐻
column vectors and matrices, respectively. (.) and (.) de- phase-shift keying (BPSK) modulation, where a realization of
note complex conjugation and Hermitian transposition, respec- any codeword is called a transmission matrix. In Table I, the
tively. For a complex variable 𝑥, ℜ {𝑥} denotes the real part first two information bits (𝑢1 , 𝑢2 ) are used to determine the
of 𝑥. 0𝑚×𝑛 denotes the 𝑚 × 𝑛 matrix with all-zero elements. antenna-pair position ℓ while the last two (𝑢3 , 𝑢4 ) determine
∥⋅∥, tr (⋅) and det (⋅) stand for the Frobenius norm, trace and the BPSK symbol pair. If we generalize this system to 𝑀 -
determinant of a matrix, respectively. The probability of an ary signaling, we have four different codewords each having
event is denoted by 𝑃 (⋅) and 𝐸 {⋅} represents∪expectation. 𝑀 2 different realizations. Consequently, the spectral efficiency
𝑛
𝑛
) of sets 𝐴1 through 𝐴𝑛 is written as 𝑖=1 𝐴𝑖 . We
The (union of the STBC-SM scheme for four transmit antennas becomes
use 𝑘 , ⌊𝑥⌋, and ⌈𝑥⌉ for the binomial coefficient, the largest 𝑚 = (1/2) log2 4𝑀 2 = 1 + log2 𝑀 bits/s/Hz, where the
integer less than or equal to 𝑥, and the smallest integer larger factor 1/2 normalizes for the two channel uses spanned by the
than or equal to 𝑥, respectively. We use ⌊𝑥⌋2𝑝 for the largest matrices in (2). For STBCs using larger numbers of symbol
BAŞAR et al.: SPACE-TIME BLOCK CODED SPATIAL MODULATION 825

TABLE I
STBC-SM MAPPING RULE FOR 2 BITS / S /H Z TRANSMISSION USING the following, we give an algorithm to design the STBC-SM
BPSK, FOUR TRANSMIT ANTENNAS AND A LAMOUTI ’ S STBC scheme:
Input Transmission Input Transmission 1) Given the total number of transmit antennas 𝑛𝑇 , calcu-
Bits
(
Matrices
)
Bits
(
Matrices
) late the number of possible antenna combinations for the
1 1 0 0 0 1 1 0 𝑗𝜃 transmission of Alamouti’s STBC, i.e.,
0000 1000 𝑒 ⌊( the)⌋total number
(−1 1 0 0) (0 −1 1 0) of STBC-SM codewords from 𝑐 = 𝑛2𝑇 2𝑝 , where 𝑝
1 −1 0 0 0 1 −1 0 𝑗𝜃
0001 1001 𝑒 is a positive integer.
ℓ=0 ( 1 1 0 0 ) ℓ=2 ( 0 1 1 0 ) 2) Calculate the number of codewords in each codebook
−1 1 0 0 0 −1 1 0 𝑗𝜃
0010 1010 𝑒 𝜒𝑖 , 𝑖 = 1, 2, . . . , 𝑛 − 1 from 𝑎 = ⌊𝑛𝑇 /2⌋ and the total
( −1 −1 0 0) ( −1 −1 0)
0
−1 −1 0 0 0 −1 −1 0 𝑗𝜃 number of codebooks from 𝑛 = ⌈𝑐/𝑎⌉. Note that the
0011 1011 𝑒
𝜒1 ( 1 −1 0 0
) 𝜒2 ( 1 −1 0)
0 last codebook 𝜒𝑛 does not need to have 𝑎 codewords,
0100
0 0 1 1
1100
1 0 0 1
𝑒𝑗𝜃 i.e, its cardinality is 𝑎′ = 𝑐 − 𝑎(𝑛 − 1).
( 0 0 −1 1) (1 0 0 −1) 3) Start with the construction of 𝜒1 which contains 𝑎 non-
0 0 1 −1 −1 0 0 1 𝑗𝜃
0101 1101 𝑒 interfering codewords as
ℓ=1 ( 0 0 1 1 ) ℓ=3 ( 1 0 0 1 ) {( )
0 0 −1 1 1 0 0 −1 𝑗𝜃 𝜒1 = X 02×(𝑛𝑇 −2)
0110 1110 𝑒
( )
(0 0 −1 −1) (−1 0 0 −1) 02×2 X 02×(𝑛𝑇 −4)
0 0 −1 −1 −1 0 0 −1 𝑗𝜃 ( )
0111 1111 𝑒
0 0 1 −1 −1 0 0 1 02×4 X 02×(𝑛𝑇 −6)
..
.
( )}
intervals such as the quasi-orthogonal STBC [15] for four 02×2(𝑎−1) X 02×(𝑛𝑇 −2𝑎) (6)
transmit antennas which employs four symbol intervals, the where X is defined in (1).
spectral efficiency will be degraded substantially due to this 4) Using a similar approach, construct 𝜒𝑖 for 2 ≤ 𝑖 ≤ 𝑛
normalization term since the number of bits carried by the by considering the following two important facts:
antenna modulation (log2 𝑐), (where 𝑐 is the total number of ∙ Every codebook must contain non-interfering code-
antenna combinations) is normalized by the number of channel words chosen from pairwise combinations of 𝑛𝑇
uses of the corresponding STBC. available transmit antennas.
∙ Each codebook must be composed of codewords
A. STBC-SM System Design and Optimization with antenna combinations that were never used in
the construction of a previous codebook.
In this subsection, we generalize the STBC-SM scheme
for MIMO systems using Alamouti’s STBC to 𝑛𝑇 transmit 5) Determine the rotation angles 𝜃𝑖 for each 𝜒𝑖 , 2 ≤
antennas by giving a general design technique. An important 𝑖 ≤ 𝑛, that maximize 𝛿min (𝜒) in (5) for a given
design parameter for quasi-static Rayleigh fading channels is signal constellation and antenna configuration; that is
the minimum coding gain distance (CGD) [15] between two 𝜽 𝑜𝑝𝑡 = arg max 𝛿min (𝜒), where 𝜽 = (𝜃2 , 𝜃3 , . . . , 𝜃𝑛 ).
𝜽
STBC-SM codewords X𝑖𝑗 and X̂𝑖𝑗 , where X𝑖𝑗 is transmitted As long as the STBC-SM codewords are generated by
and X̂𝑖𝑗 is erroneously detected, is defined as the algorithm described above, the choice of other antenna
𝐻 combinations is also possible but this would not improve
𝛿min (X𝑖𝑗 , X̂𝑖𝑗 ) = min det(X𝑖𝑗 − X̂𝑖𝑗 )(X𝑖𝑗 − X̂𝑖𝑗 ) . the overall system performance for uncorrelated channels.
X𝑖𝑗 ,X̂𝑖𝑗
(3) Since we have 𝑐 antenna combinations, the resulting spectral
The minimum CGD between two codebooks 𝜒𝑖 and 𝜒𝑗 is efficiency of the STBC-SM scheme can be calculated as
defined as 1
𝑚= log 𝑐 + log2 𝑀 [bits/s/Hz]. (7)
2 2
𝛿min (𝜒𝑖 , 𝜒𝑗 ) = min 𝛿min (X𝑖𝑘 , X𝑗𝑙 ) (4)
𝑘,𝑙 The block diagram of the STBC-SM transmitter is shown in
and the minimum CGD of an STBC-SM code is defined by Fig. 1. (During each two consecutive symbol intervals, 2𝑚 ) bits
𝑢 = 𝑢1 , 𝑢2 , . . . , 𝑢log2 𝑐 , 𝑢log2 𝑐+1 , . . . , 𝑢log2 𝑐+2log2 𝑀 enter
𝛿min (𝜒) = min 𝛿min (𝜒𝑖 , 𝜒𝑗 ) . (5) the STBC-SM transmitter, where the first log2 𝑐 bits determine
𝑖,𝑗,𝑖∕=𝑗
the antenna-pair position ℓ = 𝑢1 2log2 𝑐−1 + 𝑢2 2log2 𝑐−2 + ⋅ ⋅ ⋅ +
Note that, 𝛿min (𝜒) corresponds to the determinant criterion 𝑢log2 𝑐 20 that is associated with the corresponding antenna
given in [14] since the minimum CGD between non-interfering pair, while the last 2log2 𝑀 bits determine the symbol pair
codewords of the same codebook is always greater than or (𝑥1 , 𝑥2 ) ∈ 𝛾 2 . If we compare the spectral efficiency (7) of the
equal to the right hand side of (5). STBC-SM scheme with that of Alamouti’s scheme (log2 𝑀
Unlike in the SM scheme, the number of transmit antennas bits/s/Hz), we observe an increment of 1/2log2 𝑐 bits/s/Hz
in the STBC-SM scheme need not be an integer power of 2, provided by the antenna modulation. We consider two different
since the pairwise combinations are chosen from 𝑛𝑇 available cases for the optimization of the STBC-SM scheme.
transmit antennas for STBC transmission. This provides de- Case 1 - 𝑛𝑇 ≤ 4: We have, in this case, two codebooks 𝜒1
sign flexibility. However, the total number of codeword com- and 𝜒2 and only one non-zero angle, say 𝜃, to be optimized.
binations considered should be an integer power of 2. In It can be seen that 𝛿min (𝜒1 , 𝜒2 ) is equal to the minimum
826 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 59, NO. 3, MARCH 2011

u1 14
BPSK, f2(T )

u2 1 QPSK, f 4(T )
Antenna-Pair A 12 16-QAM, f 16(T )
Selection
# 2
64-QAM, f 64(T )

ulog2 c 10
STBC-SM
ulog2 c 1 Mapper #
8
ulog2 c  2
Symbol-Pair x1 , x2
# Selection 6
nT
ulog2 c  2log2 M
4

Fig. 1. Block diagram of the STBC-SM transmitter.

2

CGD between any two interfering codewords from 𝜒1 and 0

𝜒2 . Without loss of generality, assume that the interfering 0 1/12 1/6 1/4
T /S (rad)
1/3 5/12 1/2

codewords are chosen as

( ) Fig. 2. Variation of 𝛿min (𝜒) given in (9) for BPSK, QPSK, 16-QAM and
X1𝑘 = x1 x2 02×(𝑛𝑇 −2) 64-QAM (𝑓2 (𝜃), 𝑓4 (𝜃), 𝑓16 (𝜃) and 𝑓64 (𝜃)).
( )
X2𝑙 = 02×1 x̂1 x̂2 02×(𝑛𝑇 −3) 𝑒𝑗𝜃 (8)

where X1𝑘 ∈ 𝜒1 is transmitted and X̂1𝑘 = X2𝑙 ∈ 𝜒2 𝜃3 < ⋅ ⋅ ⋅ < 𝜃𝑛 < 𝑝𝜋/2, where 𝑝 = 2 for BPSK and 𝑝 = 1 for
is erroneously detected. We calculate the minimum CGD QPSK. For BPSK and QPSK signaling, choosing
between X1𝑘 and X̂1𝑘 from (3) as {
(𝑘−1)𝜋
𝑛 , for BPSK
𝜃𝑘 = (𝑘−1)𝜋 (10)
𝛿min (X1𝑘 , X̂1𝑘 )
( ) 2𝑛 , for QPSK
𝑥1 𝑥2 − 𝑒𝑗𝜃 𝑥 ˆ1 −𝑒𝑗𝜃 𝑥 ˆ2 01×(𝑛𝑇 −3)
= mindet for 1 ≤ 𝑘 ≤ 𝑛 guarantees the maximization of the minimum
X1𝑘 ,X̂1𝑘 −𝑥∗2 𝑥∗1 + 𝑒𝑗𝜃 𝑥 ˆ∗2 −𝑒𝑗𝜃 𝑥ˆ∗1 01×(𝑛𝑇 −3)
⎛ ⎞ CGD for the STBC-SM scheme. This can be explained as
𝑥∗1 −𝑥2 follows. For any 𝑛, we have to maximize 𝛿min (𝜒) as
⎜𝑥∗2 − 𝑒−𝑗𝜃 𝑥 ˆ∗1 𝑥1 + 𝑒−𝑗𝜃 𝑥 ˆ2 ⎟
×⎜⎝ −𝑒−𝑗𝜃 𝑥ˆ∗2
⎟
−𝑒 𝑥ˆ1 ⎠
−𝑗𝜃 max 𝛿min (𝜒) = max min 𝛿min (𝜒𝑖 , 𝜒𝑗 )
𝑖,𝑗,𝑖∕=𝑗
0(𝑛𝑇 −3)×1 0(𝑛𝑇 −3)×1
{( { ∗ }) ( { }) = max min 𝑓𝑀 (𝜃𝑗 − 𝜃𝑖 ) (11)
−𝑗𝜃
= min 𝜅 − 2ℜ 𝑥 ˆ1 𝑥2 𝑒 𝜅 + 2ℜ 𝑥1 𝑥 ˆ∗2 𝑒𝑗𝜃 𝑖,𝑗,𝑖∕=𝑗
X1𝑘 ,X̂1𝑘
{ }} where 𝜃𝑗 > 𝜃𝑖 , for 𝑗 > 𝑖 and the minimum CGD between
2 2 2 2
−∣𝑥1 ∣ ∣ˆ𝑥1 ∣ − ∣𝑥2 ∣ ∣ˆ
𝑥2 ∣ + 2ℜ 𝑥1 𝑥 ˆ1 𝑥∗2 𝑥
ˆ∗2 𝑒𝑗2𝜃 (9) codebooks 𝜒𝑖 and 𝜒𝑗 is directly determined by the difference
between their rotation angles. This can be easily verified from
∑2 ( 2 2
)
where 𝜅 = 𝑖=1 ∣𝑥𝑖 ∣ + ∣ˆ 𝑥𝑖 ∣ . Although maximization of (9) by choosing the two interfering codewords as X𝑖𝑘 ∈ 𝜒𝑖
and X̂𝑖𝑘 = X𝑗𝑙 ∈ 𝜒𝑗 with the rotation angles 𝜃𝑖 and
𝛿min (X1𝑘 , X̂1𝑘 ) with respect to 𝜃 is analytically possible for
𝜃𝑗 , respectively. Then, to maximize 𝛿min (𝜒), it is sufficient
BPSK and quadrature phase-shift keying (QPSK) constella-
to maximize the minimum CGD between the consecutive
tions, it becomes unmanageable for 16-QAM and 64-QAM
codebooks 𝜒𝑖 and 𝜒𝑖+1 , 𝑖 = 1, 2, . . . , 𝑛 − 1. For QPSK
which are essential modulation formats for the next generation
signaling, this is accomplished by dividing the interval [0, 𝜋/2]
wireless standards such as LTE-advanced and WiMAX. We
into 𝑛 equal sub-intervals and choosing, for 𝑖 = 1, 2, . . . , 𝑛−1,
compute 𝛿min (X1𝑘 , X̂1𝑘 ) as a function of 𝜃 ∈ [0, 𝜋/2] for
BPSK, QPSK, 16-QAM and 64-QAM signal constellations 𝜋
𝜃𝑖+1 − 𝜃𝑖 = . (12)
via computer search and plot them in Fig. 2. These curves 2𝑛
are denoted by 𝑓𝑀 (𝜃) for 𝑀 = 2, 4, 16 and 64, respectively. The resulting maximum 𝛿min (𝜒) can be evaluated from (11)
𝜃 values maximizing these functions can be determined from as
Fig. 2 as follows:
⎧ max 𝛿min (𝜒) = min {𝑓4 (𝜃2 ) , 𝑓4 (𝜃3 ) , . . . , 𝑓4 (𝜃𝑛 )}
(𝜋 )

 max 𝑓2 (𝜃) = 12, if 𝜃 = 1.57 rad

 𝜃 = 𝑓4 (𝜃2 ) = 𝑓4 . (13)

⎨max 𝑓4 (𝜃) = 11.45, if 𝜃 = 0.61 rad 2𝑛
max 𝛿min (𝜒) = 𝜃 Similar results are obtained for BPSK signaling except
𝜃 
 max 𝑓16 (𝜃) = 9.05, if 𝜃 = 0.75 rad that 𝜋/2𝑛 is replaced by 𝜋/𝑛 in (12) and (13). We obtain

 𝜃

⎩max 𝑓 (𝜃) = 8.23, if 𝜃 = 0.54 rad. the corresponding maximum 𝛿min (𝜒) as 𝑓2 (𝜃2 ) = 𝑓2 (𝜋/𝑛).
64
𝜃
On the other hand, for 16-QAM and 64-QAM signaling, the
Case 2 - 𝑛𝑇 > 4: In this case, the number of codebooks, 𝑛, selection of {𝜃𝑘 }’s in integer multiples of 𝜋/2𝑛 would not
is greater than 2. Let the corresponding rotation angles to be guarantee to maximize the minimum CGD for the STBC-SM
optimized be denoted in ascending order by 𝜃1 = 0 < 𝜃2 < scheme since the behavior of the functions 𝑓16 (𝜃) and 𝑓64 (𝜃)
BAŞAR et al.: SPACE-TIME BLOCK CODED SPATIAL MODULATION 827

TABLE II
BASIC PARAMETERS OF THE STBC-SM SYSTEM FOR DIFFERENT NUMBER 3𝜋/8 for QPSK and 16-QAM. Similarly, max 𝛿min (𝜒) is
OF TRANSMIT ANTENNAS calculated for BPSK, QPSK and 16-QAM constellations as
{
𝛿min (𝜒) 𝑓2 (𝜋/4) = 4.69, for BPSK
𝑛𝑇 𝑐 𝑎 𝑛 𝑚 [bits/s/Hz] max 𝛿min (𝜒) =
𝑀 =2 𝑀 =4 𝑀 = 16 𝜽 𝑓4/16 (𝜋/8) = 4.87, for QPSK&16-QAM.
3 2 1 2 12 11.45 9.05 0.5 + log2 𝑀
4 4 2 2 12 11.45 9.05 1 + log2 𝑀
According to the design algorithm, the codebooks can be
5 8 2 4 4.69 4.87 4.87 1.5 + log2 𝑀
constructed as follows:
{( ) ( )
6 8 3 3 8.00 8.57 8.31 1.5 + log2 𝑀 𝜒1 = x1 x2 0 0 0 0 0 0 , 0 0 x1 x2 0 0 0 0 ,
7 16 3 6 2.14 2.18 2.18 2 + log2 𝑀 ( ) ( )}
0 0 0 0 x1 x2 0 0 , 0 0 0 0 0 0 x1 x2
8 16 4 4 4.69 4.87 4.87 2 + log2 𝑀 {( ) ( )
𝜒2 = 0 x1 x2 0 0 0 0 0 , 0 0 0 x1 x2 0 0 0 ,
( ) ( )}
0 0 0 0 0 x1 x2 0 , x2 0 0 0 0 0 0 x1 𝑒𝑗𝜃2
{( ) ( )
is very non-linear, having several zeros in [0, 𝜋/2]. However, 𝜒3 = x1 0 x2 0 0 0 0 0 , 0 x1 0 x2 0 0 0 0 ,
our extensive computer search has indicated that for 16-QAM ( ) ( )}
0 0 0 0 x1 0 x2 0 , 0 0 0 0 0 x1 0 x2 𝑒𝑗𝜃3
with 𝑛 ≤ 6, the rotation angles chosen as 𝜃𝑘 = (𝑘 − 1)𝜋/2𝑛 {( ) ( )
for 1 ≤ 𝑘 ≤ 𝑛 are still optimum. But for 16-QAM signaling 𝜒4 = x1 0 0 0 x2 0 0 0 , 0 x1 0 0 0 x2 0 0 ,
( ) ( )} 𝑗𝜃4
with 𝑛 > 6 as well as for 64-QAM signaling with 𝑛 > 2, the 0 0 x1 0 0 0 x2 0 , 0 0 0 x1 0 0 0 x2 𝑒 .
optimal {𝜃𝑘 }’s must be determined by an exhaustive computer
search. B. Optimal ML Decoder for the STBC-SM System
In Table II, we summarize the basic parameters of the
In this subsection, we formulate the ML decoder for the
STBC-SM system for 3 ≤ 𝑛𝑇 ≤ 8. We observe that increasing
STBC-SM scheme. The system with 𝑛𝑇 transmit and 𝑛𝑅
the number of transmit antennas results in an increasing
receive antennas is considered in the presence of a quasi-static
number of antenna combinations and, consequently, increasing
Rayleigh flat fading MIMO channel. The received 2 × 𝑛𝑅
spectral efficiency achieved by the STBC-SM scheme. How-
signal matrix Y can be expressed as
ever, this requires a larger number of angles to be optimized √
and causes some reduction in the minimum CGD. On the 𝜌
Y= X𝜒 H + N (14)
other hand, when the same number of combinations can be 𝜇
supported by different numbers of transmit antennas, a higher where X𝜒 ∈ 𝜒 is the 2 × 𝑛𝑇 STBC-SM transmission matrix,
number of transmit antennas requires fewer angles to be transmitted over two channel uses and 𝜇 is a normalization
optimized resulting in higher minimum CGD (for an example, factor to ensure that 𝜌 is the average SNR at each receive
the cases 𝑐 = 8, 𝑛𝑇 = 5 and 6 in Table II). antenna. H and N denote the 𝑛𝑇 × 𝑛𝑅 channel matrix and
We now give two examples for the codebook generation 2 × 𝑛𝑅 noise matrix, respectively. The entries of H and N are
process of the STBC-SM design algorithm, presented above. assumed to be independent and identically distributed (i.i.d.)
Design Example 1: From Table II, for 𝑛𝑇 = 6, we have 𝑐 = complex Gaussian random variables with zero means and unit
8, 𝑎 = 𝑛 = 3 and the optimized angles are 𝜃2 = 𝜋/3, 𝜃3 = variances. We assume that H remains constant during the
2𝜋/3 for BPSK and 𝜃2 = 𝜋/6, 𝜃3 = 𝜋/3 for QPSK and transmission of a codeword and takes independent values from
16-QAM. The maximum of 𝛿min (𝜒) is calculated for BPSK, one codeword to another. We further assume that H is known
QPSK and 16-QAM constellations as at the receiver, but not at the transmitter.
⎧
 Assuming 𝑛𝑇 transmit antennas are employed, the STBC-
⎨𝑓2 (𝜋/3) = 8.00, for BPSK
max 𝛿min (𝜒) = 𝑓4 (𝜋/6) = 8.57, for QPSK SM code has 𝑐 codewords, from which 𝑐𝑀 2 different trans-
𝜽 
⎩ mission matrices can be constructed. An ML decoder must
𝑓16 (𝜋/6) = 8.31, for 16-QAM. make an exhaustive search over all possible 𝑐𝑀 2 transmission
According to the design algorithm, the codebooks can be matrices, and decides in favor of the matrix that minimizes the
constructed as below, following metric:
{( ) ( ) ( )}  √ 2
𝜒1 = x1 x2 0 0 0 0 , 0 0 x1 x2 0 0 , 0 0 0 0 x1 x2  𝜌 
X̂𝜒 = arg min Y − X𝜒 H
{( ) ( ) ( )} X𝜒 ∈𝜒 𝜇  . (15)
𝜒2 = 0 x1 x2 0 0 0 , 0 0 0 x1 x2 0 , x2 0 0 0 0 x1 𝑒𝑗𝜃2
{( ) ( )} The minimization in (15) can be simplified due to the
𝜒3 = x1 0 x2 0 0 0 , 0 x1 0 x2 0 0 𝑒𝑗𝜃3
orthogonality of Alamouti’s STBC as follows. The decoder
(where
) 0 denotes the 2 × 1 all-zero vector. Since there are can extract the embedded information symbol vector from
6
2 = 15 possible antenna combinations, 7 of them are (14), and obtain the following equivalent channel model:
discarded to obtain 8 codewords. Note that the choice of √ [ ]
𝜌 𝑥
other combinations does not affect 𝛿min (𝜒). In other words, y= ℋ𝜒 1 + n (16)
𝜇 𝑥2
the codebooks given above represent only one of the possible
realizations of the STBC-SM scheme for six transmit antennas. where ℋ𝜒 is the 2𝑛𝑅 ×2 equivalent channel matrix [16] of the
Design Example 2: From Table II, for 𝑛𝑇 = 8, we have Alamouti coded SM scheme, which has 𝑐 different realizations
𝑐 = 16, 𝑎 = 𝑛 = 4 and optimized angles are 𝜃2 = 𝜋/4, 𝜃3 = according to the STBC-SM codewords. In (16), y and n
𝜋/2, 𝜃4 = 3𝜋/4 for BPSK and 𝜃2 = 𝜋/8, 𝜃3 = 𝜋/4, 𝜃4 = represent the 2𝑛𝑅 × 1 equivalent received signal and noise
828 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 59, NO. 3, MARCH 2011

m1,0
vectors, respectively. Due to the orthogonality of Alamouti’s
m0
STBC, the columns of ℋ𝜒 are orthogonal to each other for 0
+
all cases and, consequently, no ICI occurs in our scheme as in
the case of SM. Consider the STBC-SM transmission model m2,0
as described in Table I for four transmit antennas. Since there m1,1
Minimum
are 𝑐 = 4 STBC-SM codewords, as seen from Table II, we m1 Metric
have four different realizations for ℋ𝜒 , which are given for 1
+
y Select Demapper û
𝑛𝑅 receive antennas as m2,1
⎡ ⎤ ⎡ ⎤ Aˆ , xˆ1,Aˆ , xˆ2,Aˆ
ℎ1,1 ℎ1,2 ℎ1,3 ℎ1,4 #
⎢ ℎ∗1,2 −ℎ∗1,1 ⎥ ⎢ ℎ∗1,4 −ℎ∗1,3 ⎥
⎢ ⎥ ⎢ ⎥ m1,c 1
⎢ ℎ2,1 ℎ2,2 ⎥ ⎢ ℎ2,3 ℎ2,4 ⎥
⎢ ∗ ⎥ ⎢ ⎥
⎢ ∗ ⎥ ⎢ ∗ ∗ ⎥ mc 1
ℋ0 = ⎢ ℎ2,2 −ℎ2,1 ⎥ , ℋ1 = ⎢ ℎ2,4 −ℎ2,3 ⎥ , +
⎢ .. .. ⎥ ⎢ .. .. ⎥ c1
⎢ . . ⎥ ⎢ . . ⎥
⎢ ⎥ ⎢ ⎥ m2,c 1
⎣ℎ𝑛𝑅 ,1 ℎ𝑛𝑅 ,2 ⎦ ⎣ℎ𝑛𝑅 ,3 ℎ𝑛𝑅 ,4 ⎦
ℎ∗𝑛𝑅 ,2 −ℎ∗𝑛𝑅 ,1 ℎ∗𝑛 ,4 −ℎ∗𝑛𝑅 ,3 Fig. 3. Block diagram of the STBC-SM ML receiver.
⎡ ⎤ ⎡ 𝑅 ⎤
ℎ1,2 𝜑 ℎ1,3 𝜑 ℎ1,4 𝜑 ℎ1,1 𝜑
⎢ ℎ∗1,3 𝜑∗ −ℎ∗1,2 𝜑∗ ⎥ ⎢ ℎ∗1,1 𝜑∗ −ℎ∗1,4 𝜑∗ ⎥
⎢ ⎥ ⎢ ⎥ step of the decoding process is the demapping operation based
⎢ ℎ2,2 𝜑 ℎ2,3 𝜑 ⎥ ⎢ ℎ2,4 𝜑 ℎ2,1 𝜑 ⎥
⎢ ∗ ∗ ⎥ ⎢ ⎥ on the look-up table
( used at the transmitter, to recover the)
⎢ ∗ ∗ ⎥ ⎢ ∗ ∗ ∗ ∗ ⎥
ℋ2 = ⎢ ℎ2,3 𝜑 −ℎ2,2 𝜑 ⎥ , ℋ3 = ⎢ ℎ2,1 𝜑 −ℎ2,4 𝜑 ⎥ input bits 𝑢ˆ = 𝑢ˆ1 , . . . , 𝑢
ˆlog2 𝑐 , 𝑢ˆlog2 𝑐+1 , . . . , 𝑢
ˆlog2 𝑐+2log2 𝑀
⎢ .. .. ⎥ ⎢ .. .. ⎥
⎢ . . ⎥ ⎢ . . ⎥ from the determined spatial position (combination) ℓ̂ and the
⎢ ⎥ ⎢ ⎥
⎣ ℎ𝑛𝑅 ,2 𝜑 ℎ𝑛𝑅 ,3 𝜑 ⎦ ⎣ ℎ𝑛𝑅 ,4 𝜑 ℎ𝑛𝑅 ,1 𝜑 ⎦ information symbols 𝑥 ˆ1 and 𝑥 ˆ2 . The block diagram of the ML
ℎ∗𝑛𝑅 ,3 𝜑∗ −ℎ∗𝑛𝑅 ,2 𝜑∗ ℎ∗𝑛𝑅 ,1 𝜑∗ −ℎ∗𝑛𝑅 ,4 𝜑∗ decoder described above is given in Fig. 3.
(17)
III. P ERFORMANCE A NALYSIS OF THE STBC-SM S YSTEM
where ℎ𝑖,𝑗 is the channel fading coefficient between transmit In this section, we analyze the error performance of the
antenna 𝑗 and receive antenna 𝑖 and 𝜑 = 𝑒𝑗𝜃 . Generally, we STBC-SM system, in which 2𝑚 bits are transmitted during
have 𝑐 equivalent channel matrices ℋℓ , 0 ≤ ℓ ≤ 𝑐 − 1, and for two consecutive symbol intervals using one of the 𝑐𝑀 2 =
the ℓth combination, the receiver determines the ML estimates 22𝑚 different STBC-SM transmission matrices, denoted by
of 𝑥1 and 𝑥2 using the decomposition as follows [17], resulting X1 , X2 , . . . , X22𝑚 here for convenience. An upper bound on
from the orthogonality of hℓ,1 and hℓ,2 : the average bit error probability (BEP) is given by the well-
 √ 2 known union bound [18]:
 
ˆ1,ℓ = arg min y − 𝜇𝜌 hℓ,1 𝑥1 
𝑥
𝑥1 ∈𝛾
 √ 2 (18) 1 ∑22𝑚 ∑22𝑚 𝑃 (X𝑖 → X𝑗 )𝑛𝑖,𝑗
  𝑃𝑏 ≤ 2𝑚 (20)
ˆ2,ℓ = arg min y − 𝜇𝜌 hℓ,2 𝑥2 
𝑥 2 𝑖=1 𝑗=1 2𝑚
𝑥2 ∈𝛾
[ ] where 𝑃 (X𝑖 → X𝑗 ) is the pairwise error probability (PEP)
where ℋℓ = hℓ,1 hℓ,2 , 0 ≤ ℓ ≤ 𝑐− 1, and hℓ,𝑗 , 𝑗 = 1, 2, is a of deciding STBC-SM matrix X𝑗 given that the STBC-SM
2𝑛𝑅 × 1 column vector. The associated minimum ML metrics matrix X𝑖 is transmitted, and 𝑛𝑖,𝑗 is the number of bits in error
𝑚1,ℓ and 𝑚2,ℓ for 𝑥1 and 𝑥2 are between the matrices
{ ( X𝑖)}and X𝑗 . Under the normalization
 √ 2 𝜇 = 1 and 𝐸 tr X𝐻 𝜒 X𝜒 = 2 in (14), the conditional PEP
 
𝑚1,ℓ = min y − 𝜇𝜌 hℓ,1 𝑥1  of the STBC-SM system is calculated as
𝑥1 ∈𝛾
 √ 2 (19) (√ )
  𝜌
𝑚2,ℓ = min y − 𝜇𝜌 hℓ,2 𝑥2  𝑃 (X𝑖 → X𝑗 ∣H) = 𝑄 ∥(X𝑗 − X𝑖 ) H∥ (21)
𝑥2 ∈𝛾 2
respectively. Since 𝑚1,ℓ and 𝑚2,ℓ are calculated by the ML √ ∫∞ 2
where 𝑄(𝑥) = (1/ 2𝜋) 𝑥 𝑒−𝑦 /2 𝑑𝑦. Averaging (21) over
decoder for the ℓth combination, their summation 𝑚ℓ = the channel matrix H and using the moment generating func-
𝑚1,ℓ + 𝑚2,ℓ , 0 ≤ ℓ ≤ 𝑐 − 1 gives the total ML metric for tion (MGF) approach [18], the unconditional PEP is obtained
the ℓth combination. Finally, the receiver makes a decision by as
choosing the minimum antenna combination metric as ℓ̂ =
arg min 𝑚ℓ for which (ˆ 𝑥1 , 𝑥ˆ2 ) = (ˆ
𝑥1,ℓ̂ , 𝑥
ˆ2,ℓ̂ ). As a result, the ∫𝜋/2( )𝑛𝑅 ( )𝑛𝑅
1 1 1
ℓ 𝑃 (X𝑖 → X𝑗 ) = 𝑑𝜙
total number of ML metric calculations in (15) is reduced from 𝜋 𝜌𝜆𝑖,𝑗,1
1 + 4sin 2𝜙
𝜌𝜆𝑖,𝑗,2
1 + 4sin2𝜙
𝑐𝑀 2 to 2𝑐𝑀 , yielding a linear decoding complexity as is also 0
(22)
true for the SM scheme, whose optimal decoder requires 𝑀 𝑛𝑇
where 𝜆𝑖,𝑗,1 and 𝜆𝑖,𝑗,2 are the eigenvalues of the distance
metric calculations. Obviously, since 𝑐 ≥ 𝑛𝑇 for 𝑛𝑇 ≥ 4,
matrix (X𝑖 − X𝑗 )(X𝑖 − X𝑗 )𝐻 . If 𝜆𝑖,𝑗,1 = 𝜆𝑖,𝑗,2 = 𝜆𝑖,𝑗 , (22)
there will be a linear increase in ML decoding complexity
simplifies to
with STBC-SM as compared to the SM scheme. However, as
we will show in the next section, this insignificant increase in ∫𝜋/2( )2𝑛𝑅
1 1
decoding complexity is rewarded with significant performance 𝑃 (X𝑖 → X𝑗 ) = 𝜌𝜆
𝑑𝜙 (23)
improvement provided by the STBC-SM over SM. The last 𝜋 1 + 𝑖,𝑗 2
0 4sin 𝜙
BAŞAR et al.: SPACE-TIME BLOCK CODED SPATIAL MODULATION 829

which is the PEP of the conventional Alamouti STBC [15].

Closed form expressions can be obtained for the integrals in
(22) and (23) using the general formulas given in Section 5
and Appendix A of [18].
In case of 𝑐 = 𝑎𝑛, for 𝑛𝑇 = 3 and for an even number
of transmit antennas when 𝑛𝑇 ≥ 4, it is observed that all
transmission matrices have the uniform error property due to
the symmetry of STBC-SM codebooks, i.e., have the same
PEP as that of X1 . Thus, we obtain a BEP upper bound for
STBC-SM as follows:
∑22𝑚 𝑃 (X1 → X𝑗 )𝑛1,𝑗
𝑃𝑏 ≤ . (24)
𝑗=2 2𝑚
Applying the natural mapping to transmission matrices,
𝑛1,𝑗 can be directly calculated as 𝑛1,𝑗 = 𝑤 [(𝑗 − 1)2 ], where
𝑤[𝑥] and (𝑥)2 are the Hamming weight and the binary
representation of 𝑥, respectively. Consequently, from (24), we Fig. 4. BER performance of STBC-SM scheme for BPSK and QPSK
obtain the union bound on the BEP as compared with theoretical upper bounds.

𝑃𝑏 ≤
22𝑚
∑ ∫𝜋/2( )𝑛𝑅 ( )𝑛𝑅
𝑤 [(𝑗 − 1)2 ] 1 1
𝜌𝜆 𝜌𝜆
𝑑𝜙,
𝑗=2
2𝑚𝜋 1 + 1,𝑗,1
2 1 + 1,𝑗,2
2
0 4sin 𝜙 4sin 𝜙
(25)

which will be evaluated in the next section for different system

parameters.

IV. S IMULATION R ESULTS AND C OMPARISONS

In this section, we present simulation results for the STBC-
SM system with different numbers of transmit antennas and
make comparisons with SM, V-BLAST, rate-3/4 OSTBC for
four transmit antennas [15], Alamouti’s STBC, the Golden
Code [19] and double space-time transmit diversity (DSTTD)
scheme [20]. The bit error rate (BER) performance of these Fig. 5. BER performance at 3 bits/s/Hz for STBC-SM, SM, V-BLAST,
systems was evaluated by Monte Carlo simulations for various OSTBC and Alamouti’s STBC schemes.
spectral efficiencies as a function of the average SNR per
receive antenna (𝜌) and in all cases we assumed four receive
antennas. All performance comparisons are made for a BER A. Comparisons with SM, V-BLAST, rate-3/4 OSTBC and
value of 10−5 . The SM system uses the optimal decoder Alamouti’s STBC
derived in [11]. The V-BLAST system uses MMSE detec- In Fig. 5, the BER curves of STBC-SM with 𝑛𝑇 = 4 and
tion with ordered successive interference cancellation (SIC) QPSK, SM with 𝑛𝑇 = 4 and BPSK, V-BLAST with 𝑛𝑇 = 3
decoding where the layer with the highest post detection SNR and BPSK, OSTBC with 16-QAM and Alamouti’s STBC with
is detected first, then nulled and the process is repeated for all 8-QAM are evaluated for 3 bits/s/Hz transmission. We observe
layers, iteratively [21]. We employ ML decoders for both the that STBC-SM provides SNR gains of 3.8 dB, 5.1 dB, 2.8
Golden code and the DSTTD scheme. dB and 3.4 dB over SM, V-BLAST, OSTBC and Alamouti’s
We first present the BER performance curves of the STBC- STBC, respectively.
SM scheme with three and four transmit antennas for BPSK In Fig. 6, we employ two different STBC-SM schemes
and QPSK constellations in Fig. 4. As a reference, the BEP with 𝑛𝑇 = 8 and QPSK, and 𝑛𝑇 = 4 and 8-QAM (for
upper bound curves of the STBC-SM scheme are also eval- the case 𝑛𝑇 ≤ 4, the optimum rotation angle for rectangular
uated from (25) and depicted in the same figure. From Fig. 8-QAM is found from (9) as equal to 0.96 rad for which
4 it follows that the derived upper bound becomes very tight 𝛿min (𝜒) = 11.45) for 4 bits/s/Hz, and make comparisons
with increasing SNR values for all cases and can be used as with the following schemes: SM with 𝑛𝑇 = 8 and BPSK,
a helpful tool to estimate the error performance behavior of V-BLAST with 𝑛𝑇 = 2 and QPSK, OSTBC with 32-QAM,
the STBC-SM scheme with different setups. Also note that and Alamouti’s STBC with 16-QAM. It is seen that STBC-
the BER curves in Fig. 4 are shifted to the right while their SM with 𝑛𝑇 = 8 and QPSK provides SNR gains of 3.5
slope remains unchanged and equal to 2𝑛𝑅 , with increasing dB, 5 dB, 4.7 dB and 4.4 dB over, SM, V-BLAST, OSTBC
spectral efficiency. and Alamouti’s STBC, respectively. On the other hand, we
830 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 59, NO. 3, MARCH 2011

Fig. 6. BER performance at 4 bits/s/Hz for STBC-SM, SM, V-BLAST, Fig. 7. BER performance at 5 bits/s/Hz for STBC-SM, SM, V-BLAST,
OSTBC and Alamouti’s STBC schemes. OSTBC and Alamouti’s STBC schemes.

observe 3 dB SNR gap between two STBC-SM schemes in

favor of the one that uses a smaller constellation and relies
more heaviy on the use of the spatial domain to achieve 4
bits/s/Hz. This gap is also verified by the difference between
normalized minimum CGD values of these two schemes.
We conclude from this result that one can optimize the
error performance without expanding the signal constellation
but expanding the spatial constellation to improve spectral
efficiency. However the number of required metric calculations
for ML decoding of the first STBC-SM scheme is equal to
128 while the other one’s is equal to 64, which provides
an interesting trade-off between complexity and performance.
Based on these examples, we conclude that for a given spectral
efficiency, as the modulation order 𝑀 increases, the number
of transmit antennas 𝑛𝑇 should decrease, and consequently
the SNR level needed for a fixed BER will increase while Fig. 8. BER performance at 6 bits/s/Hz for STBC-SM, SM, V-BLAST,
the overall decoding complexity will be reduced. On the other OSTBC and Alamouti’s STBC schemes.
hand, as the modulation order 𝑀 decreases, the number of
transmit antennas 𝑛𝑇 should increase, and as a result the SNR
SM scheme. We also observe that the BER performance of
level needed for a fixed BER will decrease while the overall
Alamouti’s scheme can be greatly improved (approximately 3-
decoding complexity increases.
5 dB depending on the transmission rate) with the use of the
In Figs. 7 and 8, we extend our simulation studies to 5
spatial domain. Note that although having a lower diversity
and 6 bits/s/Hz transmission schemes, respectively. Since it
order, STBC-SM outperforms rate-3/4 OSTBC, since this
is not possible to obtain 5 bits/s/Hz with V-BLAST, we depict
OSTBC uses higher constellations to reach the same spectral
the BER curve of V-BLAST for 6 bits/s/Hz in both figures.
efficiency as STBC-SM. Finally, it is interesting to note that
As seen from Fig. 7, STBC-SM with 𝑛𝑇 = 4 and 16-QAM
in some cases, SM and V-BLAST systems are outperformed
provides SNR gains of 3 dB, 4 dB, 3 dB and 2.8 dB over
by Alamouti’s STBC for high SNR values even at a BER of
SM with 𝑛𝑇 = 4 and 8-QAM, V-BLAST with 𝑛𝑇 = 3 and
10−5 .
QPSK, OSTBC with 64-QAM and Alamouti’s STBC with 32-
QAM, respectively. For 6 bits/s/Hz transmission we consider
STBC-SM with 𝑛𝑇 = 8 and 16-QAM, SM with 𝑛𝑇 = 8 B. Comparisons with the Golden code and DSTTD scheme
and 8-QAM, OSTBC with 256-QAM and Alamouti’s STBC In Fig. 9, we compare the BER performance of the STBC-
with 64-QAM. We observe that the new scheme provides 3.4 SM scheme with the Golden code and DSTTD scheme which
dB, 3.7 dB, 8.6 dB and 5.4 dB SNR gains compared to SM, are rate-2 (transmitting four symbols in two time intervals)
V-BLAST, OSTBC and Alamouti’s STBC, respectively. STBCs for two and four transmit antennas, respectively, at
By considering the BER curves in Figs. 5-8, we conclude 4 and 6 bits/s/Hz. Although both systems have a brute-force
that the BER performance gap between the STBC-SM and SM ML decoding complexity that is proportional to the fourth
or V-BLAST systems increases for high SNR values due to power of the constellation size, by using low complexity
the second order transmit diversity advantage of the STBC- ML decoders recently proposed in the literature, their worst
BAŞAR et al.: SPACE-TIME BLOCK CODED SPATIAL MODULATION 831

Fig. 9. BER performance for STBC-SM, the Golden code and DSTTD Fig. 10. BER performance at 3 bits/s/Hz for STBC-SM, SM, and Alamouti’s
schemes at 4 and 6 bits/s/Hz spectral efficiencies. STBC schemes for SC channel with 𝑟 = 0, 0.5 and 0.9.

case ML decoding complexity can be reduced to 2𝑀 3 from V. C ONCLUSIONS

𝑀 4 for general 𝑀 -QAM constellations, which we consider In this paper, we have introduced a novel high-rate, low
in our comparisons. MMSE decoding is widely used for complexity MIMO transmission scheme, called STBC-SM,
the DSTTD scheme, however, we use an ML decoder to as an alternative to existing techniques such as SM and V-
compare the pure performances of the considered schemes. BLAST. The proposed new transmission scheme employs
From Fig. 9, we observe that STBC-SM offers SNR gains both APM techniques and antenna indices to convey in-
of 0.75 dB and 1.6 dB over the DSTTD scheme and the formation and exploits the transmit diversity potential of
Golden code, respectively, at 4 bits/s/Hz, while having the MIMO channels. A general technique has been presented
same ML decoding complexity, which is equal to 128. On the for the construction of the STBC-SM scheme for any num-
other hand, STBC-SM offers SNR gains of 0.4 dB and 1.5 dB ber of transmit antennas in which the STBC-SM system
over the DSTTD scheme and the Golden code, respectively, was optimized by deriving its diversity and coding gains to
at 6 bits/s/Hz, with 50% lower decoding complexity, which is reach optimum performance. It has been shown via computer
equal to 512. simulations and also supported by a theoretical upper bound
analysis that the STBC-SM offers significant improvements
in BER performance compared to SM and V-BLAST systems
C. STBC-SM Under Correlated Channel Conditions (approximately 3-5 dB depending on the spectral efficiency)
with an acceptable linear increase in decoding complexity.
Inadequate antenna spacing and the presence of local scat- From a practical implementation point of view, the RF (radio
terers lead to spatial correlation (SC) between transmit and frequency) front-end of the system should be able to switch
receive antennas of a MIMO link, which can be modeled between different transmit antennas similar to the classical
1/2
by a modified channel matrix [22] H𝑐𝑜𝑟𝑟 = R𝑡 HR1/2 𝑟 SM scheme. On the other hand, unlike V-BLAST in which all
where R𝑡 = [𝑟𝑖𝑗 ]𝑛𝑇 ×𝑛𝑇 and R𝑟 = [𝑟𝑖𝑗 ]𝑛𝑅 ×𝑛𝑅 are the SC antennas are employed to transmit simultaneously, the number
matrices at the transmitter and the receiver, respectively. In of required RF chains is only two in our scheme, and the
our simulations, we assume that these matrices are obtained synchronization of all transmit antennas would not be required.
from the exponential correlation matrix model [23], i.e., their We conclude that the STBC-SM scheme can be useful for
∗
components are calculated as 𝑟𝑖𝑗 = 𝑟𝑗𝑖 = 𝑟𝑗−𝑖 for 𝑖 ≤ 𝑗 where high-rate, low complexity, emerging wireless communication
𝑟 is the correlation coefficient of the neighboring transmit and systems such as LTE and WiMAX. Our future work will be
receive antennas’ branches. This model provides a simple and focused on the integration of trellis coding into the proposed
efficient tool to evaluate the BER performance of our scheme STBC-SM scheme.
under SC channel conditions. In Fig. 10, the BER curves for
the STBC-SM with 𝑛𝑇 = 4 and QPSK, the SM with 𝑛𝑇 = 4
R EFERENCES
and BPSK, and the Alamouti’s STBC with 8-QAM are shown
for 3 bits/s/Hz spectral efficiency with 𝑟 = 0, 0.5 and 0.9. [1] E. Telatar, “Capacity of multi-antenna Gaussian channels," European
As seen from Fig. 10, the BER performance of all schemes Trans. Telecommun., vol. 10, no. 6, pp. 558-595, Nov./Dec. 1999.
[2] P. Wolniansky, G. Foschini, G. Golden, and R. Valenzuela, “V-BLAST:
is degraded substantially by these correlations. However, we an architecture for realizing very high data rates over the rich-scattering
observe that while the degradation of Alamouti’s STBC and wireless channel," in Proc. International Symp. Signals, Syst., Electron.
our scheme are comparable, the degradation for SM is higher. (ISSSE’98), Pisa, Italy, pp. 295-300, Sep. 1998.
[3] S. M. Alamouti, “A simple transmit diversity technique for wireless
Consequently, we conclude that our scheme is more robust communications," IEEE J. Sel. Areas Commun., vol. 16, no. 8, pp. 1451-
against spatial correlation than pure SM. 1458, Oct. 1998.
832 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 59, NO. 3, MARCH 2011

[4] V. Tarokh, H. Jafarkhani, and A. R. Calderbank, “Space-time block Ümit Aygölü (M’90) received his B.S., M.S. and
codes from orthogonal designs," IEEE Trans. Inf. Theory, vol. 45, no. Ph.D. degrees, all in electrical engineering, from
5, pp. 1456-1467, July 1999. Istanbul Technical University, Istanbul, Turkey, in
[5] X.-B. Liang, “Orthogonal designs with maximal rates," IEEE Trans. Inf. 1978, 1984 and 1989, respectively. He was a Re-
Theory, vol. 49, no. 10, pp. 2468-2503, Oct. 2003. search Assistant from 1980 to 1986 and a Lecturer
[6] E. Biglieri, Y. Hong, and E. Viterbo, “On fast-decodable space-time from 1986 to 1989 at Yildiz Technical University,
block codes," IEEE Trans. Inf. Theory, vol. 55, no. 2, pp. 524-530, Feb. Istanbul, Turkey. In 1989, he became an Assistant
2009. Professor at Istanbul Technical University, where he
[7] E. Başar and Ü. Aygölü, “High-rate full-diversity space-time block codes became an Associate Professor and Professor, in
for three and four transmit antennas," IET Commun., vol. 3, no. 8, pp. 1992 and 1999, respectively. His current research
1371-1378, Aug. 2009. interests include the theory and applications of com-
[8] E. Başar and Ü. Aygölü, “Full-rate full-diversity STBCs for three and bined channel coding and modulation techniques, MIMO systems, space-time
four transmit antennas," Electron. Lett., vol. 44, no. 18, pp. 1076-1077, coding and cooperative communication.
Aug. 2008.
[9] R. Mesleh, H. Haas, C. W. Ahn, and S. Yun, “Spatial modulation–a new Erdal Panayırcı (S’73, M’80, SM’91, F’03) re-
low complexity spectral efficiency enhancing technique," in Proc. Conf. ceived the Diploma Engineering degree in Electrical
Commun. Netw. China, Beijing, China, pp. 1-5, Oct. 2006. Engineering from Istanbul Technical University, Is-
[10] R. Mesleh, H. Haas, S. Sinanovic, C. W. Ahn, and S. Yun, “Spatial tanbul, Turkey and the Ph.D. degree in Electrical
modulation," IEEE Trans. Veh. Technol., vol. 57, no. 4, pp. 2228-2241, Engineering and System Science from Michigan
July 2008. State University, USA. Until 1998 he has been with
[11] J. Jeganathan, A. Ghrayeb, and L. Szczecinski, “Spatial modulation: the Faculty of Electrical and Electronics Engineering
optimal detection and performance analysis," IEEE Commun. Lett., vol. at the Istanbul Technical University, where he was
12, no. 8, pp. 545-547, Aug. 2008. a Professor and Head of the Telecommunications
[12] J. Jeganathan, A. Ghrayeb, L. Szczecinski, and A. Ceron, “Space- Chair. Currently, he is Professor of Electrical En-
shift keying modulation for MIMO channels," IEEE Trans. Wireless gineering and Head of the Electronics Engineering
Commun., vol. 8, no. 7, pp. 3692-3703, July 2009. Department at Kadir Has University, Istanbul, Turkey. Dr. Panayırcı’s recent
[13] J. Jeganathan, A. Ghrayeb, and L. Szczecinski, “Generalized space shift research interests include communication theory, synchronization, advanced
keying modulation for MIMO channels," in Proc. IEEE Symp. Pers. signal processing techniques and their applications to wireless communica-
Indoor Mobile Radio Commun. (PIMRC), Cannes, France, Sep. 2008. tions, coded modulation and interference cancelation with array processing.
[14] V. Tarokh, N. Seshadri, and A. R. Calderbank, “Space-time codes for He published extensively in leading scientific journals and international
high data rate wireless communications: performance criterion and code conferences. He has co-authored the book Principles of Integrated Maritime
construction," IEEE Trans. Inf. Theory, vol. 44, no. 2, pp. 744-765, Mar. Surveillance Systems (Boston, Kluwer Academic Publishers, 2000).
1998. Dr. Panayırcı spent the academic year 2008-2009, in the Department
[15] H. Jafarkhani, Space-Time Coding, Theory and Practice. Cambridge of Electrical Engineering, Princeton University, New Jersey, USA. He has
University Press, 2005. been the principal coordinator of a 6th and 7th Frame European project
[16] B. Hassibi and B. M. Hochwald, “High-rate codes that are linear in space called NEWCOM (Network of Excellent on Wireless Communications) and
and time," IEEE Trans. Inf. Theory, vol. 48, no. 7, pp. 1804-1824, July WIMAGIC Strep project representing Kadir Has University. Dr. Panayırcı was
2002. an Editor for IEEE T RANSACTIONS ON C OMMUNICATIONS in the areas of
[17] X. Guo and X.-G. Xia, “On full diversity space-time block codes Synchronization and Equalizations in 1995-1999. He served as a Member of
with partial interference cancellation group decoding," IEEE Trans. Inf. IEEE Fellow Committee in 2005-2008. He was the Technical Program Chair
Theory, vol. 55, no. 10, pp. 4366-4385, Oct. 2009. of ICC-2006 and PIMRC-2010 both held in Istanbul, Turkey. Presently he is
[18] M. S. Alaouni and M. K. Simon, Digital Communications over Fading head of the Turkish Scientific Commission on Signals and Systems of URSI
Channels, 2nd edition. John & Wiley, 2005. (International Union of Radio Science).
[19] J.-C. Belfiore, G. Rekaya, and E. Viterbo, “The Golden code: a 2 × 2
full-rate space-time code with non-vanishing determinants," IEEE Trans. H. Vincent Poor (S’72, M’77, SM’82, F’87) re-
Inf. Theory, vol. 51, no. 4, pp. 1432-1436, Apr. 2005. ceived the Ph.D. degree in EECS from Princeton
[20] E. N. Onggosanusi, A. G. Dabak, and T. M. Schmidl, “High rate University in 1977. From 1977 until 1990, he was
space-time block coded scheme: performance and improvement in on the faculty of the University of Illinois at Urbana-
correlated fading channels," in Proc. IEEE Wireless Commun. Netw. Champaign. Since 1990 he has been on the faculty
Conf., Orlando, FL, pp. 194-199, Mar. 2002. at Princeton, where he is the Michael Henry Strater
[21] R. Böhnke, D. Wübben, V. Kühn, and K. D. Kammeyer, “Reduced University Professor, and Dean of the School of
complexity MMSE detection for BLAST architectures," in Proc. IEEE Engineering and Applied Science. Dr. Poor’s re-
Global Commun. Conf., San Francisco, CA, USA, Dec. 2003. search interests are in wireless networks and related
[22] A. Paulraj, R. Nabar, and D. Gore, Introduction to Space-Time Wireless fields. Among his publications in these areas are the
Communications. Cambridge University Press, 2003. books MIMO Wireless Communications (Cambridge
[23] S. L. Loyka, “Channel capacity of MIMO architecture using the expo- University Press, 2007) and Information Theoretic Security (Now Publishers,
nential correlation matrix," IEEE Commun. Lett., vol. 5, pp. 369-371, 2009).
Sep. 2001. Dr. Poor is a member of the National Academy of Engineering, a Fellow
of the American Academy of Arts and Sciences, and an International Fellow
Ertuğrul Başar (S’09) was born in Istanbul, Turkey, of the Royal Academy of Engineering of the U.K. He is also a Fellow of
in 1985. He received the B.S. degree from Istanbul the Institute of Mathematical Statistics, the Optical Society of America, and
University, Istanbul, Turkey, in 2007, and the M.S. other organizations. In 1990, he served as President of the IEEE Information
degree from the Istanbul Technical University, Is- Theory Society, and in 2004-07 he served as the Editor-in-Chief of the IEEE
tanbul, Turkey, in 2009. He is currently a research T RANSACTIONS ON I NFORMATION T HEORY. He was the recipient of the
assistant at Istanbul Technical University while pur- 2005 IEEE Education Medal. Recent recognition of his work includes the
suing his Ph.D. degree at the same university. His 2009 Edwin Howard Armstrong Award of the IEEE Communications Society,
primary research interests include MIMO systems, the 2010 IET Ambrose Fleming Medal for Achievement in Communications,
space-time coding and cooperative diversity. and the 2011 IEEE Eric E. Sumner Award.