See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/260672812

Spatial Modulation and Space-Time Shift Keying: Optimal Performance at a

Reduced Detection Complexity

Article in IEEE Transactions on Communications · January 2013

DOI: 10.1109/TCOMM.2012.100312.120251

CITATIONS READS

52 60

4 authors, including:

Chao Xu Soon Xin Ng

University of Southampton University of Southampton
46 PUBLICATIONS 487 CITATIONS 325 PUBLICATIONS 3,080 CITATIONS

SEE PROFILE SEE PROFILE

L. Hanzo
University of Southampton
2,403 PUBLICATIONS 33,097 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Distributed Source Coding View project

Radio Over Fibre Communications View project

All content following this page was uploaded by Chao Xu on 11 June 2019.

The user has requested enhancement of the downloaded file.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. 1

Spatial Modulation and Space-Time Shift Keying:

Optimal Performance at a Reduced Detection
Complexity
Chao Xu, Student Member, IEEE, Shinya Sugiura, Senior Member, IEEE, Soon Xin Ng, Senior Member, IEEE,
and Lajos Hanzo Fellow, IEEE

Abstract—In this paper, we propose a comprehensive reduced- was demonstrated in [8] that a low-complexity SM detector
complexity detector both for hard-decision-aided as well as for may be invoked for STSK detection.
the soft-decision-assisted Spatial Modulation (SM)/Space-Time Although the antenna activation index and the LPSK/QAM
Shift Keying (STSK). More explicitly, the detection of the SM
scheme, which activates a single one out of M antennas to symbol are encoded independently in SM schemes, these two
transmit a single LPSK/QAM symbol, may be carried out by signals fade together. Hence, the attempt of detecting the two
detecting the antenna activation index m and the LPSK/QAM terms completely independently results in a significant perfor-
symbol sl separately, so that the detection complexity may be mance loss [7], except when the Channel State Information
reduced from the order of O(M · L) to the lower bound of (CSI) is known at the transmitter [9]. As a remedy, Space-
O(M + log2 L). However, the QAM aided STSK hard detection
proposed in [1] results in a performance loss. Furthermore, the Shift Keying (SSK) was proposed in [10], where simply the
Max-Log-MAP algorithm proposed for soft STSK detection in [2] antenna activation index conveys the source information.
only takes into account the maximum a posteriori probabilities, Recently, the reduced-complexity hard-decision PSK aided
which also imposed a performance degradation. Therefore, in SM detection was proposed in the context of Differential
this paper, we propose a novel solution for hard-decision-aided STSK (DSTSK) [11], where the optimal performance was
SM/STSK detection, which retains its optimal performance,
despite its reduced detection complexity, when either LPSK
retained by taking into account the correlation between the
or LQAM is employed. Furthermore, we propose the reduced- antenna activation index and the LPSK symbol. Reduced-
complexity Approx-Log-MAP algorithm conceived for the soft- complexity hard-decision QAM aided STSK detection was
decision-aided SM/STSK detector, in order to replace the sub- proposed in [1], but a performance loss was imposed. Fur-
optimal Max-Log-MAP algorithm. thermore, the reduced-complexity Max-Log-MAP algorithm
conceived for soft STSK detection was proposed in [2].
However, the Max-Log-MAP algorithm only considers the
I. I NTRODUCTION maximum a posteriori probabilities, which results in a sub-
optimal performance. Against this background, the novel con-
M ULTIPLE-Input Multiple-Output (MIMO) schemes are
capable of providing wireless communication systems
either with an increased capacity as in V-BLAST [3] and/or
tributions of this paper are as follows:
(1) Both PSK as well as QAM based reduced-complexity
with an improved diversity gain [4]. However, full-search- SM/STSK hard-decision-aided detection is proposed.
based Maximum Likelihood (ML) MIMO detection may im- (2) For soft-decision-aided detection, a reduced-complexity
pose an excessive complexity in turbo detected schemes [5], Approx-Log-MAP algorithm is conceived for SM/STSK
[6]. As a remedy, Spatial Modulation (SM) was proposed detection.
in [7], where a single one out of M transmit antennas is (3) Both the hard and the soft-decision-aided SM/STSK de-
activated to transmit a single LPSK/QAM symbol, so that a tectors proposed are generalized for different PSK/QAM
single-antenna-based detector may be invoked at the receiver. constellations, which retain their optimal unimpaired
Furthermore, in order to benefit from a diversity gain, Space- detection capabilities, despite their reduced complexity.
Time Shift Keying (STSK) was proposed in [8], where one out The remainder of this paper is organized as follows. The
of Q dispersion matrices was activated to disperse a single hard-decision aided SM detector is proposed in Section II,
LPSK/QAM symbol to multiple antennas and time-slots. It while the soft-decision-aided SM detector is conceived in
Section III. The STSK scheme, which may invoke the SM
Copyright (c) 2012 IEEE. Personal use of this material is permitted. detector is reviewed in Section IV. Our performance results
However, permission to use this material for any other purposes must be
obtained from the IEEE by sending a request to [email protected]. are provided in Section V, while our conclusions are offered
C. Xu, S. X. Ng and L. Hanzo are with the School of Electronics and in Section VI.
Computer Science, University of Southampton, Southampton, SO17 1BJ, U.K. The following notations are used throughout the pa-
(e-mail: {cx1g08,sxn,lh}@ecs.soton.ac.uk).
S. Sugiura is currently with the Toyota Central R&D Labs, Inc., Aichi per. SM(M ,N )-LPSK/QAM as well as V-BLAST(M ,N )-
480-1192, Japan (e-mail:[email protected]). LPSK/QAM denote the SM scheme and the V-BLAST scheme
The financial support of the RC-UK under the auspices of the India-UK equipped with M transmit antennas and N receive antennas.
Advanced Technology Centre (IU-ATC) and that of the EPSRC under the
China-UK science bridge as well as that of the EU’s Concerto project is Furthermore, a STSK scheme is denoted by the acronym of
gratefully acknowledged. STSK(M ,N ,T ,Q)-LPSK/QAM, where T and Q represent the
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. 2

number of symbol periods per transmission block and the total B. Reduced-Complexity Hard-Decision-Aided SM Detection
number of dispersion matrices employed, respectively. Both the matrix-by-matrix based detection of Eq. (3) and the
vector-by-vector based detection of Eq. (7) have a complexity
II. H ARD -D ECISION -A IDED SM D ETECTION order of O(M · L). In this section, we proceed further by
A. Conventional Hard-Decision-Aided SM Detection detecting the antenna activation index m and the LPSK/QAM
symbol index l separately, so that the detection complexity
For a SM scheme, the transmit vector is given by [7]: may be further reduced to the lower bound of O(M + log2 L).
Si = [0 First of all, we further extend the vector-by-vector based
| ·{z · · 0}],
· · 0}, sl , 0| ·{z (1)
detection metric of Eq. (7) as:
m−1 M−m
kZn − µm Si k2 =kZn k2 + µ2m |sl |2 − 2µm Re(Znm )Re(sl )
where (log2 L) bits are assigned to modulate an LPSK/QAM
symbol, while (log2 M ) bits are assigned to activate a single − 2µm Im(Znm )Im(sl ),
one out of a total of M transmit antennas. (8)
The signal received by the N receive antennas may be where {Znm }M m=1 denotes the m-th element in the decorre-
modelled as: lating vector Zn . As a result, the LPSK/QAM aided SM
Yn = Sn Hn + Vn , (2) detection of Eq. (7) may be simplifed to:
where Yn ∈ C1×N and Vn ∈ C1×N refer to the received {m̂, l̂} = arg max Re(Zenm )Re(sl ) + Im(Zenm )Im(sl )
m∈m̄,l∈l̄ (9)
signal vector, and the Additive White Gaussian Noise (AWGN)
vector, which has a zero mean and a variance of N0 , re- − µ2m |sl |2 .
spectively, while Hn ∈ CM×N models the Rayleigh fading where we have {Z e m = 2µm Z m }M , while m̄ and l̄
n n m=1
channel. store the antenna activation indices and LPSK/QAM symbol
Based on Eq. (2), the conventional MIMO detector, which indices, respectively. The constant of kZn k2 seen in Eq. (8)
operates on a matrix-by-matrix basis, may be expressed as: is discarded.
In order to detect m and l separately, we have to drop the
Ŝn = arg min kYn − Si Hn k2 , (3)
Si ∈S LPSK/QAM index l in Eq. (9), when detecting the antenna
where S stores all the SM codewords. Let us further extend activation index m. Let us consider QPSK aided SM detection
the decision variable in Eq. (3) as: as an example, which has a PSK constellation set of {± √12 ±
 j √12 }1 . For a specific antenna index m, the maximum metric
kYn − Si Hn k2 = tr (Yn − Si Hn )(Yn − Si Hn )H
n o over all QPSK constellations
(
is given by:
)
2 2 2 ∗ m H
= kYn k + µm |sl | − 2Re sl Yn (Hn ) , Re(Ze ) Im(Z
m em )
n n 2
dm = max ± √ ± √ − µm
(4) l∈l̄ 2 2
(10)
where the variable {µm }M m=1 is given by (µm = kHn k),
m
Re(Z em ) Im(Zenm )
n 2
while {Hm M = √ + √ − µm ,
n }m=1 denotes the m-th row in Hn . Eq. (4) leads 2 2
to a decorrelating variable of:

H which is evaluated by a single equation instead of comparing
Zn = Yn Hn , (5) all the (L = 4) QPSK constellations. As a result, the optimum
antenna activation index m̂ may be found by searching for
M×N
where each row in the normalized fading matrix H ∈ C the maximum metric over all the M candidates {dm }M m=1 ,
n m oM n

is given by Hn = Hm n /µ m . It is well known that the regardless of which particular QPSK symbol was transmitted,
m=1
decorrelating detector of V-BLAST imposes a performance which may be expressed as:
loss. However, due to the fact that only a single transmit m̂ = arg max dm , (11)
m∈m̄
antenna was activated in our SM scheme, Eq. (4) now becomes
log2 I
equivalent to the vector-by-vector based detection metric of: and then the corresponding (log2 M ) bits {b̂k }k=log 2 L+1
2 2 2 2 H assigned to activate m̂ may be obtained accordingly, where
kZn −µm Si k = kZn k +µm |sl | −2Re{µm Zn (Si ) }, (6)
(I = M · L) denotes the total number of SM codewords. Hav-
H m H log L
where we have µm Zn (Si ) = sl Yn (Hn ) according to ing determined the optimum m̂, the (log2 L) bits {b̂k }k=12
∗

Eq. (5), while both kYn k2 in Eq. (4) and kZn k2 in Eq. (6) assigned to modulate the QPSK symbol may be detected as:

are constants. Hence minimizing Eq. (4) and Eq. (6) are 1, if Im(Z em̂ ) < 0
b̂1 = n ,
equivalent. 0, otherwise
In conclusion, the vector-by-vector based SM detection may  (12)
1, if Re( Zem̂ ) < 0
be formulated as: b̂2 = n .
0, otherwise
2
Ŝn = arg min kZn − µm Si k , (7)
Si ∈S 1 We deliberately rotated all the constellations of LPSK (L ≥ 4) in [12] anti-
π
clockwise by a phase of L , so that there are exactly L/4 constellation points
where µm may be found according to the tentative antenna in each quadrant. This feature will be beneficial for reducing the complexity
activation index m. of the soft PSK aided SM/STSK detection.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. 3

It can be seen that Eqs. (11) and (12) reduces the QPSK summarized in Table II2 .
aided SM detection complexity from the order of O(M · 4) We note that the low-complexity SM detection proposed
to O(M + 2) by detecting the antenna index and the QPSK in [7] always achieves the complexity lower bound of
symbol separately. O(M + log2 L), because it detects the antenna index as
Similarly, when Square 16QAM was employed, the max- m̂ = arg maxm∈m̄ |Znm |2 , which may result in an erroneous
imum metrics {dm }M m=1 seen in Eq. (11) may be obtained decision and hence the LPSK/QAM demodulator may be mis-
by testing the real and the imaginary parts of the LQAM led into detecting the wrong symbol. By contrast, our proposed
constellation separately, which may be expressed as: SM detection characterized in this section retains the same
  detection capability as the full-search-based MIMO detection
1 em ) − 1 µ2 of Eq. (3), because the proposed antenna index detection
dRe1
m = max ± √ Re(Z n
10 10 m takes into account which particular LPSK/QAM scheme was
1 employed. As a result, only the family of 1PSK/BPSK/QPSK
= √ Re(Z em ) − 1 µ2 ,
10 
n
10 m aided SM detection arrangements may achieve the complexity
 lower bounds of O(M ), O(M +1) and O(M +2), respectively,
Re2 3 e m 9 2
dm = max ± √ Re(Zn ) − µm where the special case of hard-decision 1PSK/BPSK aided
10 10
SM detection is summarized in the Appendix. The high-order
3 enm ) − 9 µ2m , Square LQAM aided SM√detection presented in Table I has
= √ Re(Z
10  10 a complexity order of O( L · M + 4), where a total number
 (13) √
1 e 1 2 of L comparisons have been made for estimating {dm }M m=1 ,
dIm1
m = max ± √ Im(Z n
m
) − µ
10 10 m while the streamlined Square LQAM demodulator detects the
1 log2 L bits by simply testing the two variables of Re(Z em̂ )
= √ Im(Z em ) − 1 µ2 , e m̂
n

10 
n
10 m and Im(Zn ), as well as the two magnitude indices î and
 ĵ. Similarly, the generalized LPSK/QAM aided SM detection
3 em ) − 9 µ2
dIm2
m = max ± √ Im(Z n complexity order of Table II is given by O( L4 · M + 3), which
10 10 m
is higher than that of Table I.
3 em ) − 9 µ2 , When the number of bits per Square LQAM symbol is
= √ Im(Z n
10 10 m an odd number, the Algorithm 1 shown in Table I may be
readily√ modified, where the real positive PAM magnitudes
where each one of them only has to be evaluated once.
have 2L/2 candidates,
p while the imaginary positive PAM
Furthermore, for a specific antenna index m, the maximum
magnitudes have L/2/2 candidates. Furthermore, it was
metric is given by:
shown in [15] that Cross-shaped L-QAM constellations ac-
tually have a better performance compared to Square L-QAM
dm = max dRe Imj
m + max dm .
i
(14)
i∈{1,2} j∈{1,2} schemes. We note that the Algorithm 2 of Table II may be
adopted for detecting the family of Cross LQAM aided SM
Then the antenna activation index detection of Eq. (11) may schemes.
be invoked, and a streamlined Square 16QAM detection may
be carried out as follows: III. S OFT-D ECISION -A IDED SM D ETECTION

1, if Im(Zenm̂ ) < 0 A. Conventional Soft-Decision-Aided SM Detection
b̂1 = ,
0, otherwise For the soft-decision-aided detection, the classic Log-MAP

1, if ĵ = 1 for dm̂ algorithm is given by [16]:
b̂2 = , "P #
0, otherwise
 (15) Si ∈Sbk =1 exp (di )
1, if Re(Zenm̂ ) < 0 Lp (bk ) = ln P , (16)
b̂3 = , Si ∈Sb =0 exp (di )
0, otherwise k

1, if î = 1 for dm̂ where Lp (bk ) refers to the a posteriori LLRs, while Sbk =1
b̂4 = ,
0, otherwise and Sbk =0 denote the SM codeword sets, when the specific
bit bk is fixed to 1 and 0, respectively. The probability metric
where the optimum 16QAM magnitude indices î and ĵ have {di }Ii=1 in Eq. (16) is given by:
been obtained in Eq. (14). log2 I
kYn − Si Hn k2 X
We summarize the hard-decision Square LQAM aided SM di = − + bj La (bj ), (17)
detection in Table I. Furthermore, when either a high-order N0 j=1
LPSK (L > 4) or a Star L-QAM constellation [13], [14] is
employed, the real and imaginary parts of the decorrelating where La (bk ) refers to the a priori LLRs gleaned from a
variable Zenm cannot be detected separately, as seen in Table I. channel decoder. Similar to Eq. (7), Eq. (17) may be calculated
However, as long as we have two bits, which determine the 2 We deliberately rotate all the Star L-QAM constellations of [13], [14]
signs of the real and imaginary parts of the transmitted symbol, anti-clockwise by a phase angle of (π/LP ), where LA and LP refers to the
a similar detection algorithm may be conceived, which is number of constellation rings and the number of phasors, respectively.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. 4

TABLE I
A LGORITHM 1: R EDUCED -C OMPLEXITY H ARD -D ECISION S QUARE LQAM AIDED SM DETECTION .
(1) Define the metrics that tests
˛ the real and
˛ imaginary parts separately
˛ as: ˛
˛ em ˛ ˛ em ˛
dRe i 2 2
m = ˛Ai Re(Zn )˛ − Ai µm , dIm i 2 2
m = ˛Ai Im(Zn )˛ − Ai µm ,
√
L/2
where {Ai }i=1 are the positive real PAM magnitudes on the x-axis and y-axis of Square LQAM constellation diagram.
(2) For a specific m, the maximum metric over all Square LQAM constellations is given by testing the real and imaginary parts separately as:
Re Im
dm = maxi∈{1,··· ,√L/2} dmi + maxj∈{1,··· ,√L/2} dmj ,
where the optimum PAM magnitudes indices î and ĵ obtained for each {dm }M m=1 may be recorded.
(3) The optimum antenna activation index may be found by:
m̂ = arg maxm∈m̄ dm ,
log2 I
and then the corresponding (log2 M ) bits {b̂k }k=log L+1 assigned to activate m̂ may be obtained accordingly.
2
log L
(4) The first bit and the ( 22 + 1)-th bit which determine the signs may be demodulated as:
1, if Im( Zem̂ ) < 0 1, if Re(Z em̂ ) < 0
b̂1 = n , b̂(log2 L)/2+1 = n .
0, otherwise 0, otherwise
(log2 L)/2 log 2 L
(5) For the (log2 L − 2) bits {b̂k }k=2 and {b̂k }(log L)/2+2 which determine the magnitudes, the optimum bit mapping is correspond-
2
ing to the imaginary and real magnitude indices ĵ and î, respectively, which were obtained when estimating dm̂ in Step (2).
TABLE II
A LGORITHM 2: R EDUCED -C OMPLEXITY H ARD -D ECISION G ENERAL LPSK/QAM AIDED SM DETECTION .
(1) Define the new testing metrics as: ˛ ˛ ˛ ˛
˛ enm )˛ + ˛ enm )˛ − (A2 + B 2 )µ2 ,
dim = ˛Ai Re(Z ˛ ˛Bi Im(Z ˛ i i m
L/4
where {(Ai , Bi )}i=1 denote the coordinates of the LPSK/QAM constellation points in the first quadrant, and (A2i + Bi2 = 1) is a const-
ant when LPSK is emploed.
(2) For a specific m, the maximum metric over all LPSK/QAM constellations is given by:
dm = maxi∈{1,··· ,L/4} dim ,
where the optimum constellation index î obtained for each {dm }M m=1 may be recorded.
(3) The optimum antenna activation index may be found by:
m̂ = arg maxm∈m̄ dm ,
log 2 I
and then the corresponding (log2 M ) bits {b̂k }k=log assigned to activate m̂ may be obtained accordingly.
2 L+1
(4) The first bit and the second
 bit which determine the signsmay be demodulated as:
1, if Im(Z enm̂ ) < 0 1, if Re(Z enm̂ ) < 0
b̂1 = , b̂2 = .
0, otherwise 0, otherwise
log 2 L
(5) For the rest (log2 L − 2) bits {b̂k }k=3 which determine the magnitudes, the optimum bit mapping arrangement is corresponding to the
optimum constellation index î, which were obtained when estimating dm̂ in Step (2).

on a vector-by-vector basis as: the same magnitudes but are associated with the opposite
2 log2 I signs, the maximization operation gives the simple result of
kZn − µm Si k X (max{t, −t} = |t|). Similarly, we define the special case for
di = − + bj La (bj ). (18)
N0 j=1
the Jacobian algorithm as:

According to our previous results in Eqs. (4) and (6), the Λ(|t|) = jac(t, −t) = |t| + Γ{2|t|}. (22)
differences between Eq. (17) and Eq. (18) are all constants,
which are eliminated by the division operation in Eq. (16). In the following section, we aim for invoking Eq. (22) for a
The Log-MAP algorithm may be simplified by the Max- streamlined algorithm.
Log-MAP algorithm as [16]:
Lp (bk ) = max (di ) − max (di ) . (19)
Si ∈Sbk =1 Si ∈Sbk =0 B. Reduced-Complexity Soft-Decision-Aided SM Detection
Since only the pair of maximum a posteriori probabilities For producing a single soft-bit output, the conventional SM
are taken into account in Eq. (19), the Max-Log-MAP algo- detectors have to estimate a total number of (I = M · L) a
rithm imposes a slight performance degradation. In order to posteriori probabilities {di }Ii=1 . In this section, we once again
compensate for this performance loss, the Approx-Log-MAP aim for reducing the detection complexity by detecting m and
algorithm was introduced as [17]: l separately. For the (log2 M ) bits which are assigned to the
Lp (bk ) = jacSi ∈Sb (di ) − jacSi ∈Sb (di ) , (20) antenna index, the detection complexity order of the Approx-
k =1 k =0
Log-MAP algorithm of Eq. (20) may be lower bounded by
where jac denotes the Jacobian algorithm, which may be O(M ), where ideally the antenna index detector does not have
expressed as [6]: to visit the L-element PSK/QAM symbols set. Meanwhile,
jac(d1 , d2 ) = max {d1 , d2 } + Γ{|d1 − d2 |}, (21) for the (log2 L) bits which are assigned to an LPSK/QAM
symbol, the detection complexity may be lower bounded by
where the additional term of Γ{|d1 − d2 |} takes into ac- O(2M ), where ideally the antenna index detector only has
count the difference between d1 and d2 according to a to be invoked twice according to the updated LPSK/QAM
lookup table [6]. When comparing two variables, which have subsets, when a specific bit bk is set to 1 and 0, respectively.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. 5

Similar to Eq. (8), the a posteriori probability evaluation of the Approx-Log-MAP algorithm of Eq. (20) is formulated as:
Eq. (18) may be extended as: h i
Lp (b1 ) = jacm∈m̄ jacl∈l̄b =1 (dm,l )
enm )Re(sl ) Im(Z enm )Im(sl ) µ2m |sl |2 h 1 i
Re(Z
dm,l = + − − jacm∈m̄ jacl∈l̄b =0 (dm,l )
N0 N0 N0 
1

log2 I (23) µ2m (30)
X = jacm∈m̄ Λ(|tRe
m |) − t Im
m − + Pr m
+ bj La (bj ), N0
 
j=1
Re Im µ2m
− jacm∈m̄ Λ(|tm |) + tm − + Prm ,
 2
 N0
where a constant of − kZNn0k is discarded from Eq. (18).
where the imaginary term of Λ(|tIm m |) in Eq. (27) is replaced
For the (log2 M ) bits which are assigned to the antenna in- by (−tIm Im
m ) and (tm ), when b1 is fixed to 1 and 0, respectively.
dex, the soft decisions produced by Eq. (20) may be expressed Similarly, the second soft bit decision is given by:
as:  
Re Im µ2m
Lp (b2 ) = jacm∈m̄ −tm + Λ(|tm |) − + Prm
Lp (bk ) = jacm∈m̄b (dm ) − jacm∈m̄b (dm ) , (24) N0
k =1 k =0   (31)
µ2m
− jacm∈m̄ tRem + Λ(|t Im
m |) − + Pr m .
where m̄bk =1 and m̄bk =0 refer to the index set for m, when the N0
log2 I
specific bit {bk }k=log 2 L+1
is fixed to 1 and 0, respectively. For The corresponding Max-Log-MAP algorithm may be ob-
a specific antenna index m seen in Eq. (24), we have to obtain tained by replacing all the Jacobian operations of jac by the
the probability of dm = jacl∈l̄ (dm,l ). Considering QPSK as maximization operation of max, while the special case of
an example, the a posteriori probability for each antenna index Λ(|t|) may be replaced by |t|.
is given by Eq. (25), where hthe a priori probability of the i The complexity order of Eqs. (30) and (31) is O(2M ),
Plog2 I
antenna index m is given by Prm = j=log 2 L+1
bj La (bj ) . where the antenna index detector is invoked twice according
Let us define two variables to test the real and the imaginary to the updated symbol set, when the specific bit is fixed to 1
part separately as: and 0.
Let us further consider the example of Square 16QAM aided
Re(Zem ) La (b2 ) Im(Zem ) La (b1 )
tRe
m =
√ n
2N0
− 2 , tIm
m =
√ n
2N0
− 2 , (26) SM detection, where the a posteriori probability of a specific
antenna index is given by:
so that Eq. (25) may be further expressed as:
dm = jacl∈l̄ (dm,l ) = dRe Im
m + dm + Prm , (32)
µ2m where we aim for testing the real and the imaginary part
dm = Λ(|tRe Im
m |) + Λ(|tm |) − + Prm , (27)
N0 separately. The real part dRe m in Eq. (32) may be further
extended as:
where a constant of La (b1 )+L
2
a (b2 )
is discarded from Eq. (25).  Re(Zem ) µ2m


 √ n + La (b4 ) − , 

Therefore, the Approx-Log-MAP algorithm of Eq. (24) may  10N 0 10N 0 
Re Re(Zem ) µ2
be invoked by using the a posteriori probabilities {dm }M m=1 dm = jac − 10N + La (b3 ) + La (b4 ) − 10N0 ,
√ n m

 0 

of Eq. (27) in order to detect the last (log2 M ) bits  3Re( Zem ) 9µ2m −3Re(Z em ) 9µ2m 
log2 I √ n
− 10N0 , √ n
+ L a (b 3 ) − 10N0 ,
{Lp (bk )}k=log 2 L+1
.  10N0 10N0

When the Max-Log-MAP algorithm is invoked, the a pos- Re1 µ2m Re2 9µ2m
= jac Λ(|tm |) + La (b4 ) − , Λ(|tm |) −
teriori probability of Eq. (27) may be further simplified as: 10N0 10N0
La (b3 )
µ2m +
dm = |tRe Im 2
m | + |tm | − + Prm , (28)
N0  Re1 Re2 La (b3 )
= jac dm , dm + ,
2
while the (log2 M ) soft bit decisions may be made without (33)
invoking the Jacobian algorithms as:
where the constant of La 2(b3 ) may be deleted, while the two
Lp (bk ) = max (dm ) − max (dm ) . (29) test-variables are defined as:
m∈m̄bk =1 m∈m̄bk =0
em )
Re(Z La (b3 ) em )
3Re(Z La (b3 )
tRe1
m =
√ n
10N0
− 2 , tRe2
m =
√ n
10N0
− 2 .
It can be seen that the detection algorithms of (34)
Eqs. (24) and (29) only have to estimate and compare the Similarly, the imaginary term dIm
m in Eq. (32) may be formu-
M a posteriori probabilities {dm }M m=1 of Eqs. (27) and (28), lated as:
respectively. Therefore the complexity order has been reduced ( )
µ2m
Im Λ(|tIm1
m |) + La (b2 ) − 10N0 ,
from O(M · L) to O(M ). dm = jac 9µ2m
For the first (log2 L) bits, when a specific bit is set to 1 or Λ(|tIm2
m |) − 10N0 (35)
 Im1 
0 as seen in Eq. (20), the LPSK constellation set has to be dm
= jac ,
updated. For the sake of producing the first soft bit decision, dIm2
m
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. 6

 Re(Zem ) Im(Zem ) µ2 

 √ n + √ n − m + Prm 


 2N0 2N0 N0 


 − Re( Zem ) Im(Zem ) µ2 

√ n + √ n + La (b2 ) − m + Prm
2N0 2N0 N0
dm = jacl∈l̄ em ) em ) µ2
. (25)
 Re(

Z Im(Z
√ n − √ n + La (b1 ) − m + Prm 


 2N0 2N0 N0 

 Re(
 Zem ) Im(Zem ) µ2 

− √2N − √2N + La (b1 ) + La (b2 ) − Nm0 + Prm
n n
0 0

where the constant of La 2(b1 ) is discarded, while the two test- For the (log2 M ) bits assigned to the antenna index, the
variables are defined as: 1PSK/BPSK/QPSK aided SM detection achieves the complex-
em )
Im(Z La (b1 ) 3Im(Zem ) La (b1 ) ity order lower bound of O(M ), while the Square LQAM
tIm1
m = √ n
10N0
− 2 , tIm2
m = √ n
10N0
− 2 . aided SM detection of Table III and the LPSK/QAM aided
(36) SM√ detection of Table IV have the complexity order of
Therefore, for the Approx-Log-MAP algorithm, the antenna O( L · M ) and O( L4 · M ), respectively. For the two specific
index detector of Eq. (24) may be invoked by utilizing the M a bits, which determine the sign of the transmitted LPSK/QAM
posteriori probabilities {dm }Mm=1 defined in Eq. (32), in order symbol, the BPSK/QPSK aided SM detection complexity is
log2 I
to produce the (log2 M ) soft bit decisions {Lp (bk )}k=log 2 L+1
. lower bounded by the order of O(2M ), while the Square
The complexity order is reduced from O(M · 16) to O(M · LQAM aided SM detection complexity order and the general
4), where four comparison operations have been made in LPSK/QAM aided SM detection complexity order are given
Eqs. (33) and (35). √
by O( L · M ) and O( L2 · M ), respectively. For the remaining
When the first bit b1 which determines the sign of the (log2 L − 2) bits, which determine the specific magnitudes
imaginary part of a Square 16QAM symbol is fixed to 1 of the LPSK/QAM symbols, the complexity order of the
2
or 0, {Λ(|tImi
m |)}i=1 seen in Eq. (35) may be replaced by Square LQAM aided SM detection and that of√the general
Imi 2 2
{−tm }i=1 and {tImi m }i=1 , respectively. More explicitly, the LPSK/QAM aided SM detection are given by O( 2L · M ) and
Approx-Log-MAP produces the first soft bit as Eq. (37), where O( L4 · M ), respectively. In summary, the Square LQAM aided
dRe
m does not have to be estimated again. The complexity order SM detection of Algorithm 3 has a lower complexity compared
of detecting La (b1 ) is given by O(M · 4). to Algorithm 4, where the latter applies to high-order LPSK,
When the second bit b2 which determines the magnitude of Star LQAM and Cross LQAM aided SM schemes.
the imaginary part of a Square 16QAM symbol is fixed to 1 or
0, Eq. (35) should be updated as dImm = dm
Im1
or dIm
m = dm ,
Im2

respectively. As a result, the second soft bit decision is given IV. S PACE -T IME S HIFT K EYING
by: It was demonstrated in [8] that the SM detector may be
Lp (b2 ) = jacm∈m̄ (dRe Im1 invoked for STSK detection. In fact, a SM scheme may be
m + dm + P rm )
(38) seen as a special case of STSK in conjunction with (T = 1)
− jacm∈m̄ (dRe Im2
m + dm + P rm ),
and (Q = M ) [18]. In this section, we summarize the STSK
where there is no new variable to evaluate, i.e. only additions schemes having different parameters in three cases, so that
and comparisons are made in Eq. (38). Therefore, the com- our proposed reduced-complexity SM detector may be invoked
plexity order of detecting the second bit is given by O(M · 2). accordingly.
Similarly, the third bit which determines the sign of the
real part of a Square 16QAM symbol may be detected by the
Approx-Log-MAP algorithm as: A. STSK Encoding


For a STSK scheme, the (T × M )-element transmission
Lp (b3 ) = jacm∈m̄ dbm3 =1 − jacm∈m̄ dbm3 =0 , (39)
matrix
 is obtained
 by the so-called dispersion process [8] of
where dbm3 =1 and dbm3 =0 are obtained by replacing e
Sn = Aq sl , where log2 L bits are assigned to modulate
2 Rei 2
{Λ(|tRei Re
m |)}i=1 seen in dm of Eq. (33) by {−tm }i=1 a single LPSK/QAM symbol {sl }L
2 l=1 , while log2 Q bits are
and {tRei
m } i=1 , respectively. assigned to activate one out of a total number of Q dispersion
Furthermore, the fourth bit, which modulates the magnitude matrices {Ae q }Q .
q=1
of the real part of a Square 16QAM symbol may be detected In order to obtain the optimum dispersion matrix set,
as: (T × T ) full-rank unitary matrices
are randomly generated,
Lp (b4 ) = jacm∈m̄ (dRe1 Im
m + dm + P rm ) where we have T = max{M, T } . The set of dispersion
(40)
− jacm∈m̄ (dRe2 Im
m + dm + P rm ).
matrices is obtained by taking the first T rows or the first
M columns of the unitary matrices, for the caseq of (M > T )
We have summarized the Approx-Log-MAP conceived for T
Square LQAM aided SM detection in Table III, while the and (M < T ), respectively. A constant of ( M ) should be
general LPSK/QAM aided SM detection is summarized in used for multiplying all the dispersion matrices,
h when we havei
Table IV. The special case of 1PSK/BPSK aided SM detection e e H
(M < T ), so that the power constraint of tr(Aq · Aq ) = T
is detailed in the Appendix. As discussed right after Eq. (31), may be satisfied. The optimum dispersion matrices may be
the reduced-complexity Max-Log-MAP may be obtained ac- obtained by maximizing the minimum distance
between STSK
cordingly. codewords as max {det [(Sf − Sg )]}min [8].
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. 7

Lp (b1 ) = jacm∈m̄ dbm1 =1 − jacm∈m̄ dbm1 =0
  1 2 9 2

= jacm∈m̄ dRem + jac −tIm1
m + La (b2 ) − 10 µm , −tIm2
m − 10 µm + P rm (37)
  Im1 1 2 9 2

− jacm∈m̄ dRe m + jac tm + La (b2 ) − 10 µm , tIm2
m − 10 µm + P rm .

TABLE III
A LGORITHM 3: R EDUCED -C OMPLEXITY A PPROX -L OG -MAP A LGORITHM C ONCEIVED FOR S QUARE LQAM A IDED SM D ETECTION .
(1) Define the variables testing the real and imaginary en
parts ”of the decorrelating variable Z m separately as:
“
Re em )
Ai Re(Z La b(log L)/2+1 Im em )
Ai Im(Z La (b1 )
2
tmi = N0
n
− 2
, tmi = N0
n
− 2
,
√
L/2
where {Ai }i=1 are the positive real PAM magnitudes on the x-axis and y-axis of Square LQAM constellation diagram.
(2) The a posteriori probability of a specific antenna index m is given by:
dm = dRe Im
m + dm + Prm ,
Plog2 I
where the a priori probability of the antenna index m is given by Prm = j=log bj La (bj ), while the real and imaginary terms of
2 L+1
the decision metric are given by: “ ” “ ”
dRe √
m = jaci∈{1,··· , L/2} dRe
m
i
, dIm √
m = jac i∈{1,··· , L/2} dIm
m
i
.
√ √ √
Re L/2 Im L/2
The L/2 candidates of {dm i }i=1 and {dm i }i=1 may be evaluated by:
Re Re Plog2 L A2 2
i µm
dm i = Λ(|tm i |) + j=(log bj La (bj ) − N ,
2 L)/2+2 0
P(log2 L)/2 A2 2
i µm
dIm
m
i
= Λ(|tImi
m |) + j=2 b L (b
j a j ) − N0
.
(3) The (log2 M ) bits which determine the antenna activation index may be detected as:
Lp (bk ) = jacm∈m̄b =1 (dm ) − jacm∈m̄b =0 (dm ) , k ∈ {log2 L + 1, · · · , log2 I}.
“ k” k
log2 L
(4) The first bit and the 2
+ 1 -th bit which determine the signs may be detected as:
“ ” “ ”
Im,b1 =1 Im,b1 =0
Lp (b1 ) = jacm∈m̄ dRe m + dm + Prm − jacm∈m̄ dRe m + dm + Prm ,
“ ” „ «
Re,b(log L)/2+1 =1
Lp b(log2 L)/2+1 = jacm∈m̄ dm 2
+ dIm
m + Prm
„ «
Re,b(log L)/2+1 =0
−jacm∈m̄ dm 2
+ dIm
m + Prm ,
√ √ √
L/2 L/2 L/2
Im,b1 =1
where dm Im,b1 =0
and dm may be obtained by replacing {Λ(|tImi
m |)}i=1 in Step (2) by {−tIm
m√
i
}i=1 and {tImi
m }i=1 , respect-
√
Re,b(log L)/2+1 =1 Re,b(log L)/2+1 =0 L/2 L/2
dm
ively, while √ 2
and dm 2
may be obtained by replacing {Λ(|tRei
m |)}i=1 in Step (2) by {−tRei
m }i=1
Rei L/2
and {tm }i=1 , respectively.
(5) The rest (log2 L − 2) bits whichh determine the real PAM magnitudes
i may be detected as:
Imi
Lp (bk ) = jacm∈m̄ dRe
m + jac bk =1 (dm ) + Prm
h i
Imi
−jacm∈m̄ dRem + jac bk =0 (dm ) + Prm , k ∈ {2, · · · , (log2 L)/2},
h i
Lp (bk ) = jacm∈m̄ jacbk =1 (dRei Im
m ) + dm + Prm
h i
Rei
−jacm∈m̄ jacbk =0 (dm ) + dImm + Prm , k ∈ {(log 2 L)/2 + 2, · · · , log2 L}.

TABLE IV
A LGORITHM 4: R EDUCED -C OMPLEXITY A PPROX -L OG -MAP A LGORITHM C ONCEIVED FOR G ENERAL LPSK/QAM A IDED SM D ETECTION .
(1) Define the test-variables as:
Re em )
Ai Re(Z La (b2 ) Im em )
Bi Im(Z La (b1 )
tmi = N0
n
− 2
, tmi = N0
n
− 2
,
L/4
where {(Ai , Bi )}i=1denote the coordinates of the LPSK/QAM constellation points in the first quadrant.
(2) The a posteriori probability of a specific antenna index m is given by:
dm = jaci∈{1,··· ,L/4} (dim ) + Prm ,
L/4
where the L/4 candidates of {dim }i=1 are evaluated by:
Re Im Plog L (A2 +B 2 )µ2
dim = Λ(|tm i |) + Λ(|tm i |) + j=32 bj La (bj ) − i N i m .
0
(3) The (log2 M ) bits which determine the antenna activation index may be detected as:
Lp (bk ) = jacm∈m̄b =1 (dm ) − jacm∈m̄b =0 (dm ) , k ∈ {log2 L + 1, · · · , log2 I}.
k k
(4) The first two bits which determine the signs may be detected as:
i,b1 =1
Lp (b1 ) = jacm∈m̄ (dm ) − jacm∈m̄ (di,b
m
1 =0
),
Lp (b2 ) = jacm∈m̄ (di,b
m
2 =1
) − jac m∈m̄ (di,b2 =0
m ),
Imi L/4 Imi L/4 Imi L/4
where di,b
m
1 =1
and d i,b1 =0
m may be obtained by replacing {Λ(|tm |)}i=1 in Step (2) by {−tm }i=1 and {tm }i=1 , respectively,
i,b2 =1 i,b2 =0 Rei L/4 Rei L/4 Rei L/4
while dm and dm may be obtained by replacing {Λ(|tm |)}i=1 in Step (2) by {−tm }i=1 and {tm }i=1 , respectively.
(5) The rest (log2 L − 2) bits whichh determine the imagnitude of hthe transmitted iLPSK/QAM symbol may be detected as:
Lp (bk ) = jacm∈m̄ jacbk =1 (dim ) − jacm∈m̄ jacbk =0 (dim ) , k ∈ {3, · · · , log2 L}.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. 8

B. STSK Detection 10
0

The received signal model of STSK may also be represented SM(4,N)-QPSK

V-BLAST(2,N)-QPSK
by Eq. (2), where the STSK transmission matrix Sn has (T × -1
10 STSK(2,N,2,4)-Square 64QAM
M ) elements, while the received signal matrix Yn and the G2 STBC-Square 16QAM
AWGN matrix Vn are of size (T × N ). Therefore, the hard-
-2 R=4 N=1
decision matrix-by-matrix-based STSK detection is also given 10
N=2

BER
by Eq. (3), while the soft STSK detection algorithms may
invoke the matrix-by-matrix-based a posteriori probability of -3
10
Eq. (17). In order to operate STSK detection on a vector-by-
vector basis, Eq. (2) may be formulated as [8]: N=2
10
-4 N=1
Y n = Kn Hn + Vn , (41)
where the notations are given by: 10
-5

T 0 5 10 15 20 25 30 35 40 45 50
Y n = [rvec(Yn )] , Hn = χ(IT ⊗ Hn ), Eb/N0
T
Kn = [0 · · · 0, s ,
| {z } l | {z }0 · · · 0], Vn = [rvec(Vn )] ,
q−1 Q−q
Fig. 2. BER performance of the V-BLAST(2,N)-QPSK, SM(4,N)-QPSK,
h iT STSK(2,N,2,4)-Square 64QAM as well as Square 16QAM aided G2 STBC.
e
χ = rvec(A1 ), · · · , rvec(A e Q) ,
n oQ
(42) He q = Hn /µq , while the normalization variables
n
while ⊗ denotes the Kronecker product. As a result, the new q=1
q
decision metric becomes: are given by {µq = Hn }Q
q=1 .
2 2 q 2 H We note that among all the Generalized STSK schemes of [18],
Y n − Ki H n = Yn +|sl |2 Hn −2Re(Y n Hn KHi ), only STSK detection may proceed from Eq. (43) to Eq. (44),
n q oQ (43) which allows us to invoke our proposed reduced-complexity
where Hn denotes the q-th row in Hn , and we have SM detector.
 q=1 
q 2 2
Hn = A e q Hn . We note that detecting the antenna V. P ERFORMANCE R ESULTS
index m for SM is equivalent to detecting the dispersion matrix We provide our simulation results in this section. Our com-
index q for STSK. Therefore, the decision metrics seen in parison between the reduced-complexity hard STSK detection
Eqs. (7) and (18) should be expressed in the following form conceived in Sec. IV and the STSK detection proposed in [1]
for STSK detection: is portrayed in Fig. 1. The STSK detectors of [1] can only
2 2
kZn − µq Ki k = kZn k + |sl |2 µ2q − 2Re(µq Zn KH achieve the ML performance when Star LQAM is employed,
i ), (44)
while the Square LQAM aided STSK detectors impose a
which is equivalent to Eq. (43). In order to arrive at the performance loss, which is evidenced by Fig. 1(a). By contrast,
appropriate decorrelating vector Zn and at the normalization Fig. 1(a) shows that the new STSK detectors proposed in this
variables {µq }Q
q=1 , we further consider STSK in three scenar- work retain their optimal detection capability, despite their
ios: substantially reduced complexity. Moreover, in Fig. 1(b) we
(1) When we have (M ≤ T ) and LPSK is employed for quantify the complexity imposed in terms of the total number
STSK, the  fading channel’s output power in Eq. (43)  of real-valued calculations required for producing a single-bit
q 2 H eH e 2 decision. Explicitly, the complexity comparison of Fig. 1(b)
is given by Hn = tr{Hn Hn Aq Aq } = kHn k ,
demonstrates that the proposed STSK detector’s complexity
which is a constant.
 In such a H
case,
 the decorrelating is lower than that of its counterparts conceived in [1]. This
vector is given by Zn = Yn Hn , while the normal- is a benefit of the fact that the LQAM demodulators were
ization variables are given by a constant of {µq = 1}Q streamlined in the proposed Algorithms 1 and 2. It is also
q=1 .
(2) When we have (M ≤ T ) and LQAM is employed demonstrated in Fig. 1 that although the Star LQAM aided
for STSK, the symbol-power (|sl |2 ) is no longer a STSK detector using Algorithm 2 has a slightly better BER
constant. Therefore, the decorrelating vector may be performance, the Square LQAM aided STSK detection using
 
e H Algorithm 1 exhibits the lowest detection complexity.
obtained by Zn = Y n Hn , where the fading chan-
Fig. 2 demonstrates our performance comparison between
nel should be normalized by (H e n = Hn /µ), while a range of MIMO schemes associated with the same rate,
the normalization variables are given by a constant of where the transmission rates of the SM, V-BLAST, STSK
{µq = µ = kHn k}Q q=1 . and STBC schemes are given by log2 (L · M ), log2 (LM ),
(3) When we have (M > T ), the fading channel power log2 (L·Q)/T and log2 (LQ )/T , respectively. When no receive
q 2
of ( Hn ) is no longer a constant. As a result, we diversity is achieved owing to (N = 1), the SM scheme
  performs slightly worse than its V-BLAST counterpart, and the
eH ,
may have the decorrelating vector as Zn = Y n H n STSK scheme has an improved performance as a benefit of its
where each row in H e n should be normalized as diversity gain, while STBC G2 exhibits the best performance,
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. 9

Matrix-by-Matrix STSK(2,2,2,4) employing

STSK(2,N,2,4)-Square 64QAM, Matrix-by-Matrix Square/Star 64QAM
0 2
10 STSK(2,N,2,4)-Square 64QAM, Reduced-Complexity

Computational Complexity (per bit)

STSK(2,N,2,4)-Square 64QAM, Sugiura et al.
STSK(2,N,2,4)-Star 64QAM, Sugiura et al. Vector-by-Vector
-1 3
10 10

-2
10 5
BER

-3
10

2
N=1 Sugiura et al.
-4
10
N=2 Reduced-Complexity
2
10
-5
10 Square Star Square Star Square Star Square Star
0 5 10 15 20 25 30 35 40 45
Eb/N0

(a) BER performance comparison. (b) Detection complexity comparison.

Fig. 1. BER performance and detection complexity comparison between the reduced-complexity hard STSK detectors conceived in Sec. IV and the STSK
detectors proposed by Sugiura et al. in [1].

Matrix-by-Matrix based soft SM detection

Vector-by-Vector based soft SM detection Matrix-by-Matrix based soft SM detection
Reduced-Complexity soft SM detection Reduced-Complexity soft SM detection
4
SM(4,N) 10
SM(M,1) 2 Approx-Log-MAP
Computational Complexity (per bit)

5
Computational Complexity (per bit)

Computational Complexity (per bit)

5 Max-Log-MAP
SM(4,1)-1PSK/BPSK/QPSK/8PSK/
2 3
10 2 Square 16/32/64/128/256QAM
3
10
10
3 Algorithm 4
5
5 5

2 2

2 2
10
2 10

1 2 3 4 1 2 3 4 5
log2M N
0 1 2 3 4 5 6 7 8
log2 L
(a) (b)
Fig. 4. Complexity comparison between soft SM detection arrangments
Fig. 3. Complexity comparison between soft SM detection arrangments when when different LPSK/QAM scheme was employed for SM(4,1) scheme. The
different number of antennas are equipped. Square 16QAM was employed for Square LQAM aided SM detection invokes Algorithm 3 with a lower detection
SM(M,N) scheme, while the Approx-Log-MAP algorithm was invoked for soft complexity, while high-order LPSK, Star LQAM or Cross LQAM aided SM
SM detection. detection invokes Algorithm 4.

which is evidenced by Fig. 2. However, Fig. 2 also shows that significantly as N increases. Furthermore, Figs. 3(a) and 4
as N increases, the SM scheme and the V-BLAST scheme show that the complexity reduction provided by our proposed
perform better at low SNRs because the STSK scheme and the soft SM detectors becomes even more substantial as M and L
G2 STBC scheme have to employ high-order QAM in order increases. For the SM(4,1)-Square 64QAM scheme, a factor
to compensate for their thoughput loss owing to utilizing T 14 and a factor 19 complexity reduction are achieved by the
symbol periods for the sake of achieving full diversity. Approx-Log-MAP and by the Max-Log-MAP, respectively,
Our complexity comparisons made for different soft SM which is evidenced by Fig 4. Furthermore, as expected, Fig. 4
detection arrangements are portrayed in Figs. 3 and 4. It can shows that Square LQAM aided SM detection using Algo-
be seen in Fig. 3(b) that following the decorrelating process of rithm 3 has a lower detection complexity compared to high-
Eq. (5), the complexity of both the vector-by-vector based SM order LPSK, Star LQAM or Cross LQAM aided SM detection
detection and of the proposed SM detection no longer grows using Algorithm 4. Fig. 4 also confirms that Approx-Log-MAP
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. 10

TC-STSK and TC-V-BLAST systems. Since the SM scheme

1.0 has a high iteration gain, while the G2 STBC scheme has
R=4 a near-horizontal EXIT curve in Fig. 5, (Iouter = 10) and
Eb/N0 = 0 dB
0.9 (Iouter = 2) were used by the TC-SM system and the TC-G2
N=2 STBC system, respectively. The BER performance of Fig. 6
0.8
confirmed our EXIT chart based predictions of Fig. 5.
0.7

0.6 VI. C ONCLUSIONS

IEM

0.5 We have conceived both hard-decision and soft-decision

LPSK/QAM aided SM detection. The antenna index and the
0.4
N=1 LPSK/QAM symbol are detected separately, while their cor-
0.3 N=1 relation is taken into account, so that the optimal performance
N=2
is retained. Furthermore, our simulation results demonstrate
0.2 SM(4,N)-QPSK that although the SM/STSK schemes performed slightly worse
V-BLAST(2,N)-QPSK
0.1 STSK(2,N,2,4)-Square 64QAM
than the V-BLAST/STBC schemes, they have a substantially
G2 STBC-Square 16QAM reduced detection complexity, which offers them an appealing
0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
advantage in realistic MIMO systems.
IAM
VII. A PPENDIX
Fig. 5. EXIT charts of the V-BLAST(2,N)-QPSK, SM(4,N)-QPSK,
STSK(2,N,2,4)-Square 64QAM as well as Square 16QAM aided G2 STBC.
A. Reduced-Complexity Hard-Decision-Aided SSK Detection
The optimum antenna activation index is given by m̂ =
arg maxm∈m̄ Re(Zenm ) − µ2m .
SM(4,N)-QPSK
V-BLAST(2,N)-QPSK
10
0 STSK(2,N,2,4)-Square 64QAM
G2 STBC-Square 16QAM B. Reduced-Complexity Hard-Decision-Aided BPSK Based
-1
SM Detection
10
The antenna activation index may be found by evaluating
-2 m̂ = arg maxm∈m̄ Re(Z em ) − µ2 . Then the BPSK
10 n m
BER

em̂ ).
demodulator may be invoked for detecting the sign of Re(Z n
-3
10

-4
C. Reduced-Complexity Approx-Log-MAP Algorithm Con-
10 N=2 N=1
ceived for SSK Detection

10
-5 The Approx-Log-MAP algorithm conceived for SSK detec-
-2 -1 0 1 2 3 4 5 6 7 8 tion is given by Eq. (24), where we have k ∈ {1, · · · , log2 M }.
Eb/N0 (dB)
The a posteriori probability metric of Eq. (24) is given by
Fig. 6. BER performance of the turbo coded V-BLAST(2,N)-QPSK, em )
Re(Z µ2
SM(4,N)-QPSK, STSK(2,N,2,4)-Square 64QAM as well as Square 16QAM dm = N0n − Nm0 + Prm .
aided G2 STBC.

generally has a higher detection complexity than Max-Log- D. Reduced-Complexity Approx-Log-MAP Algorithm Con-
MAP. However, it is widely recognized that Approx-Log-MAP ceived for BPSK Aided SM Detection
outperforms Max-Log-MAP, and a performance comparison log M+1
The (log2 M ) bits {Lp (bk )}k=22 which determine the
between these two algorithms invoked for STSK detection may antenna activation index may be detected by Eq. (24). The a
be found in [2]. We invoke Approx-Log-MAP for all the soft posteriori probability metric of Eq. (24) is given by dm =
MIMO detectors in the rest of this section. µ2m
Λ(|tRe
m |) − N0 + Prm , while the test-variable is defined as
The EXIT curves of Fig. 5 demonstrate that both the SM em )
− La (b 1)
Re(Z
and the STSK exhibit an increased iteration gain. Furthermore, tRe
m = N0
n
2 . Furthermore, the first bit, which is
Fig. 5 also predicts that the SM scheme and the V-BLAST assigned to the BPSK symbol may be detected as:
scheme employing low-level QPSK may perform better with  
Re µ2m
the aid of channel coding, when (N = 2) receive antennas are Lp (b1 ) = jacm∈m̄ −tm − + Prm
N0
used. We applied our proposed design in a Turbo Coded (TC)   (45)
system [19]. Four inner TC iterations (Iinner = 4) were used, Re µ2m
− jacm∈m̄ tm − + Prm .
while (Iouter = 5) outer iterations were employed for the N0
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. 11

R EFERENCES Chao Xu (S’09) received a B.Eng. degree from

Beijing University of Posts and Telecommunica-
[1] S. Sugiura, C. Xu, S. X. Ng, and L. Hanzo, “Reduced-complexity tions, Beijing, China, and a BSc(Eng) with First
coherent versus non-coherent QAM-aided space-time shift keying,” Class Honours from Queen Mary, University of
IEEE Transactions on Communications, vol. 59, pp. 3090 –3101, Nov. London, London, UK, through a Sino-UK joint
2011. degree program in 2008, both in Telecommunica-
[2] C. Xu, S. Sugiura, S. X. Ng, and L. Hanzo, “Reduced-complexity soft- tions Enginneering with Management. In 2009, he
decision aided space-time shift keying,” IEEE Signal Processing Letters, obtained a MSc degree with distinction in radio
vol. 18, pp. 547 –550, Oct. 2011. frequency communication systems from the Univer-
[3] C. J. Foschini, “Layered space-time architecture for wireless commu- sity of Southampton, Southampton, UK, and he was
nication in a fading environment when using multiple antennas,” Bell awarded IEEE Communications Society UK&RI
Labs. Tech. J., vol. 1, no. 2, pp. 41–59, 1996. Chapter Best MSc Student in Broadband and Mobile Communication Net-
[4] S. Alamouti, “A simple transmit diversity technique for wireless commu- works. He is currently working towards the PhD degree with the Research
nications,” IEEE Journal on Selected Areas in Communications, vol. 16, Group of Communications, Signal Processing and Control, School of Elec-
pp. 1451 –1458, Oct. 1998. tronics and Computer Science, University of Southampton, UK. His research
[5] L. Hanzo, O. Alamri, M. El-Hajjar, and N. Wu, Near-Capacity Multi- interests include reduced-complexity MIMO design, non-coherent space-time
Functional MIMO Systems(Sphere-Packing, Iterative Detection and Co- modulation detection, EXIT-chart-aided turbo detection as well as cooperative
operation). John Wiley & Sons, May 2009. communications.
[6] L. Hanzo, T. Liew, and B. Yeap, Turbo Coding, Turbo Equalisation and
Space-Time Coding for Transmission over Fading Channels. Wiley-
IEEE Press, 2003.
[7] R. Mesleh, H. Haas, S. Sinanovic, C. W. Ahn, and S. Yun, “Spatial
modulation,” IEEE Transactions on Vehicular Technology, vol. 57,
pp. 2228 –2241, July 2008.
[8] S. Sugiura, S. Chen, and L. Hanzo, “Coherent and differential space-
time shift keying: A dispersion matrix approach,” IEEE Transactions on
Communications, vol. 58, pp. 3219 –3230, Nov. 2010.
[9] J. Jeganathan, A. Ghrayeb, and L. Szczecinski, “Spatial modulation: Shinya Sugiura (M06-SM12) received the B.S.
optimal detection and performance analysis,” IEEE Communications and M.S. degrees in aeronautics and astronautics
Letters, vol. 12, pp. 545 –547, Aug. 2008. from Kyoto University, Kyoto, Japan, in 2002 and
[10] J. Jeganathan, A. Ghrayeb, L. Szczecinski, and A. Ceron, “Space shift 2004, respectively, and the Ph.D. degree in mobile
keying modulation for MIMO channels,” IEEE Transactions on Wireless communications from the University of Southamp-
Communications, vol. 8, pp. 3692 –3703, July 2009. ton, Southampton, U.K., in 2010. Since 2004, he
[11] C. Xu, S. Sugiura, S. X. Ng, and L. Hanzo, “Reduced-complexity non- has been with Toyota Central R&D Laboratories,
coherently detected differential space-time shift keying,” IEEE Signal Inc., Aichi, Japan, where his research has covered
Processing Letters, vol. 18, pp. 153 –156, Mar. 2011. a range of areas in wireless communications, net-
[12] L. Hanzo, S. X. Ng, W. T. Webb, and T. Keller, Quadrature Amplitude working, signal processing, and antenna design. He
Modulation: From Basics to Adaptive Trellis-Coded, Turbo-Equalised authored/coauthored more than 45 refereed research
and Space-Time Coded OFDM, CDMA and MC-CDMA Systems, 3rd publications, including 21 IEEE journal and magazine papers.
Edition. John Wiley & Sons, Sept. 2004. Dr. Suguira has received a number of distinctions, including the 2011 IEEE
[13] X. Dong, N. Beaulieu, and P. Wittke, “Error probabilities of two- Communications Society Asia-Pacific Outstanding Young Researcher Award,
dimensional M -ary signaling in fading,” IEEE Transactions on Com- the 2011 Ericsson Young Scientist Award, and the 2008 IEEE Antennas and
munications, vol. 47, pp. 352 –355, Mar. 1999. Propagation Society Japan Chapter Young Engineer Award.
[14] T. May, H. Rohling, and V. Engels, “Performance analysis of Viterbi
decoding for 64-DAPSK and 64-QAM modulated OFDM signals,” IEEE
Transactions on Communications, vol. 46, pp. 182 –190, Feb. 1998.
[15] P. Vitthaladevuni, M.-S. Alouini, and J. Kieffer, “Exact BER compu-
tation for cross QAM constellations,” IEEE Transactions on Wireless
Communications, vol. 4, pp. 3039 – 3050, Nov. 2005.
[16] W. Koch and A. Baier, “Optimum and sub-optimum detection of
coded data disturbed by time-varying intersymbol interference,” in IEEE
Global Telecommunications Conference (GLOBECOM’90), pp. 1679 –
Soon Xin Ng (S’99-M’03-SM’08) received the
1684 vol.3, Dec. 1990.
[17] P. Robertson, E. Villebrun, and P. Hoeher, “A comparison of optimal and B.Eng. degree (First class) in electronics engineering
sub-optimal MAP decoding algorithms operating in the log domain,” in and the Ph.D. degree in wireless communications
IEEE International Conference on Communications (ICC’95), vol. 2, from the University of Southampton, Southampton,
U.K., in 1999 and 2002, respectively. From 2003 to
pp. 1009 –1013 vol.2, June 1995.
[18] S. Sugiura, S. Chen, and L. Hanzo, “Generalized space-time shift keying 2006, he was a postdoctoral research fellow working
designed for flexible diversity-, multiplexing- and complexity-tradeoffs,” on collaborative European research projects known
IEEE Transactions on Wireless Communications, vol. 10, pp. 1144 – as SCOUT, NEWCOM and PHOENIX. Since Au-
1153, Apr. 2011. gust 2006, he has been a member of academic
[19] C. Berrou and A. Glavieux, “Near optimum error correcting coding staff in the School of Electronics and Computer
and decoding: Turbo-codes,” IEEE Transactions on Communications, Science, University of Southampton. He is involved
in the OPTIMIX and CONCERTO European projects as well as the IU-ATC
vol. 44, pp. 1261 –1271, Oct. 1996.
and UC4G projects. He is currently a senior lecturer at the University of
Southampton.
His research interests include adaptive coded modulation, coded modula-
tion, channel coding, space-time coding, joint source and channel coding,
iterative detection, OFDM, MIMO, cooperative communications, distributed
coding and quantum error correction codes. He has published over 140 papers
and co-authored two John Wiley/IEEE Press books in this field. He is a
senior member of the IEEE, a Chartered Engineer and a fellow of the Higher
Education Academy in the UK.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. 12

Lajos Hanzo (M’91-SM’92-F’04) FREng, FIEEE,

FIET, Fellow of EURASIP, DSc received his degree
in electronics in 1976 and his doctorate in 1983. In
2009 he was awarded the honorary doctorate “Doc-
tor Honoris Causa” by the Technical University of
Budapest. During his 35-year career in telecommu-
nications he has held various research and academic
posts in Hungary, Germany and the UK. Since 1986
he has been with the School of Electronics and
Computer Science, University of Southampton, UK,
where he holds the chair in telecommunications. He
has successfully supervised about 80 PhD students, co-authored 20 John
Wiley/IEEE Press books on mobile radio communications totalling in excess
of 10,000 pages, published 1250+ research entries at IEEE Xplore, acted both
as TPC and General Chair of IEEE conferences, presented keynote lectures
and has been awarded a number of distinctions. Currently he is directing an
academic research team, working on a range of research projects in the field of
wireless multimedia communications sponsored by industry, the Engineering
and Physical Sciences Research Council (EPSRC) UK, the European IST
Programme and the Mobile Virtual Centre of Excellence (VCE), UK. He is an
enthusiastic supporter of industrial and academic liaison and he offers a range
of industrial courses. He is also a Governor of the IEEE VTS. Since 2008
he has been the Editor-in-Chief of the IEEE Press and since 2009 a Chaired
Professor also at Tsinghua University, Beijing. For further information on
research in progress and associated publications please refer to http://www-
mobile.ecs.soton.ac.uk

View publication stats