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

net/publication/224563047

Space shift keying modulation for MIMO Channels

Article in IEEE Transactions on Wireless Communications · August 2009

DOI: 10.1109/TWC.2009.080910 · Source: IEEE Xplore

CITATIONS READS
836 2,375

4 authors, including:

Ali Ghrayeb Leszek Szczecinski

Texas A&M University at Qatar Institut National de la Recherche Scientifique
279 PUBLICATIONS 7,089 CITATIONS 138 PUBLICATIONS 3,297 CITATIONS

SEE PROFILE SEE PROFILE

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

Physical (PHY) layer security (PLS) for visible light communications (VLC) View project

Distributed Collaborative Beamforming (DCB) for Multi-Node Wireless Relays and Sensor Networks View project

All content following this page was uploaded by Leszek Szczecinski on 28 June 2014.

The user has requested enhancement of the downloaded file.

 Generalized Space Shift Keying Modulation
for MIMO Channels
Jeyadeepan Jeganathan and Ali Ghrayeb Leszek Szczecinski
ECE Department, Concordia University INRS-EMT, University of Quebec
Montreal, Quebec, H3G 1M8, Canada Montreal, Quebec, H5A 1K6, Canada
Email: {j_jegana, aghrayeb}@ece.concordia.ca Email: [email protected]

Abstract— A fundamental component of spatial modulation remains active during transmission so that ICI is avoided, and
(SM), termed generalized space shift keying (GSSK), is pre- IAS is no longer needed.
sented. GSSK modulation inherently exploits fading in wireless Although SM is shown to reduce the receiver’s complexity
communication to provide better performance over conventional
amplitude/phase modulation (APM) techniques. In GSSK, only as compared to V-BLAST [3], this is under a sub-optimal
the antenna indices, and not the symbols themselves (as in the SM detection rule that is only valid under some constrained
case of SM and APM), relay information. We exploit GSSK’s assumptions about the channel. Due to the sub-optimality of
degrees of freedom to achieve better performance, which is done the detection, SM does not exhibit the best performance in
by formulating its constellation in an optimal manner. To support [3]. As well, the constrained assumptions about the channel
our results, we also derive upper bounds on GSSK’s bit error
probability, where the source of GSSK’s strength is made clear. questions the validity of the performance comparison with V-
Analytical and simulation results show performance gains (1.5−3 BLAST in [6]. The optimal detector for SM is, however, de-
dB) over popular multiple antenna APM systems (including rived in [7] under conventional channel assumptions, and SM
Bell Laboratories layered space time (BLAST) and maximum is shown to outperform many schemes including V-BLAST (at
ratio combining (MRC) schemes), making GSSK an excellent the expense of increased receiver complexity). Also, the trade-
candidate for future wireless applications.
off between the number of transmit antennas versus the APM
constellation size is chosen heuristically in [6]. All of this
I. I NTRODUCTION
motivates our presentation of a simpler modulation technique,
Using multiple antennas in wireless communications allows namely GSSK, which can be used to build a stronger SM
unprecedented improvements over single antenna systems. foundation.
One example is the vertical Bell Laboratories layered space- Contribution: We analyze GSSK as a fundamental com-
time (V-BLAST) architecture [1], where multiple symbols are ponent of SM, in which the spatial domain is exploited to
multiplexed in space, and transmitted at the same time over modulate information. The presentation of SM in [6] does
all antennas. Due to inter-channel interference (ICI), caused not fully explore the idea of using antenna indices as the
by coupling multiple symbols in time and space, V-BLAST only means to relay information, as is the case for our GSSK
maximum likelihood (ML) detection increases exponentially scheme. The transmitted symbols in GSSK are just a means of
in complexity with the number of transmit antennas. Hence, identifying the activated antenna. In doing so, we achieve all of
practical integration of V-BLAST requires sub-optimal, low the aforementioned advantages comprising SM, while reduc-
complexity receivers [2]. For adequate performance, these ing transceiver overhead. GSSK’s constellation is thoroughly
receivers require the number of receive antennas to be larger or analyzed, where we present the underlying idea that allows
equal to the number of transmit antennas, which is not practi- GSSK to outperform APM schemes (such as V-BLAST and
cal for downlink transmission to small mobile devices. Conse- MRC). In particular, we show that GSSK takes advantage of
quently, avoiding ICI greatly reduces receiver complexity, and the fading process by increasing the constellation’s dimension,
results in performance gains. Also, the V-BLAST algorithms whose points result to be well spread apart. This analysis
assume that all symbols are transmitted at the same time. opens the door to understanding how SM parameters may be
Hence, inter-antenna synchronization (IAS) is necessary to chosen to obtain better performance gains. We also consider
avoid performance degradation, which consequently increases the performance of the GSSK scheme, where upper bounds
transmitter overhead. on the bit error rate (BER) are derived. Through our analysis,
Prior Work: In [3]–[6], the so-called spatial modulation we design optimal constellations, where it is apparent that
(SM) seems to be an effective means to remove ICI, and the tremendous degrees of freedom is available for practical imple-
need for precise time synchronization amongst antennas. SM is mentation. Simulation results are also presented to support our
a pragmatic approach for transmitting information, where the findings, and illustrate the future research potential of GSSK
modulator uses well known APM techniques, such as PSK and modulation.
quadrature amplitude modulation (QAM), but also employs Organization: Section II introduces the basic GSSK system
the antenna index to convey information. Only one antenna model, including a detailed analysis of GSSK’s constellation

978-1-4244-2644-7/08/$25.00 © 2008 IEEE

 TABLE I
space. We present analytical results on the bit error probability
E XAMPLE OF THE GSSK M APPER R ULE .
in Section III, followed by Section IV’s development of
constellation design rules. Section V then provides simulation £ ¤ £ ¤T
b = b1 b2 b3 j x = x1 x2 ··· x5
results on performance, and we conclude the paper in Section £ ¤ h iT
0 0 0 (1, 2) √1 √1 0 0 0
VI. 2 2
Notation: Italicized symbols denote scalar values while £ ¤ h iT
0 0 1 (1, 3) √1 0 √12 0 0
bold lower/upper case symbols denote vectors/matrices. ¡ ¢ We h
2
iT
£ ¤
use (·)T for transpose, (·)H for conjugate transpose, ·· for 0 1 0 (1, 4) √1 0 0 √12 0
2
the binomial coefficient, and k·kF¡ for the ¢ Frobenius norm £ ¤ h iT
of a vector/matrix. We use CN m, σ2 for the complex 0 1 1 (1, 5) √1 0 0 0 √12
2
Gaussian distribution of a random variable, having indepen- £ ¤ h iT
1 0 0 (2, 3) 0 √1 √1 0 0
dent Gaussian
³ ´ distributed real and imaginary2 parts denoted h
2 2
iT
2 £ ¤
by N m, σ2 , with mean m and variance σ2 . We use P (·) 1 0 1 (2, 4) 0 √1 0 √1 0
2 2
for the probability of an event, pY (·) for the probability £ ¤ h iT
1 1 0 (2, 5) 0 √1 0 0 √1
density function (PDF) of a random variable Y, and Ex [·] 2 2
£ ¤ h iT
for the statistical expectation with respect to x. We use Re {·} 1 1 1 (3, 4) 0 0 √1 √1
0
for the real part of a complex variable, and X to represent a 2 2

constellation of size M .
II. GSSK M ODULATION For example, with nt = 2 and Nt = 7, there are M 0 = 21
The general system model consists of a MIMO wire- possible combinations. Since we require a constellation size
less link with Nt transmit and Nr receive antennas, and M in multiples of 2, we only use 16 of the possible 21
is shown£ in Fig. 1. A random sequence of independent combinations. The set of antenna combinations, X , may be
¤ chosen at random, but we will see in Section IV that more
bits b = b1 b2 · · · bk enters a GSSK mapper, where
groups of m bits are optimal selection rules exist.
£ ¤T mapped to a constellation vector x = Once X is formulated, the GSSK’s mapper rule is straight-
x1 £ x2 ¤ · · · xNt , with a power constraint of unity (i.e.
forward. Groups of m = log2 (M ) bits are collected and
Ex xH x = 1). In GSSK, only nt antennas remain active
mapped to a vector xj , where j ∈ X specifies the antenna
during transmission, and hence, only nt of the xj ’s in x are
combination for the given m bit pattern. The symbols in xj
nonzero. The signal is transmitted over an Nr × Nt wireless
do not contain information, but can be designed to optimize
channel H, and experiences an£ Nr − dim additive ¤T white transmission.1 The vector xj specifies the activated antennas,
Gaussian (AWGN) noise η = η1 η2 · · · ηNr . The
√ during which all other antennas remain idle, and has the
received signal is given by y = ρHx + η, where ρ is the
following form:
average signal to noise ratio (SNR) at each receive antenna, h iT
and H and η have independent and identically distributed (iid) xj , √1nt 0 · · · 0 √1nt · · · √1nt 0 .
entries according to CN (0, 1). | {z }
nt of Nt non-zero values
η1
An example of 8-ary GSSK modulation is given in Table I,
1 1
η2
where we use Nt = 5, nt = 2, and X is chosen randomly. The
2 2
output of the channel is therefore given by
b GSSK GSSK b̂ p
MAPPER
M M ηN
DETECTOR
y = ρ0 hj,eff + η, (1)
r

Nt Nr where ρ0 = nρt , and hj,eff = hj(1) + hj(2) + · · · + hj(nt) (j (·) =

x y j ∈ {1, 2, . . . , Nt } specifies the column index of H). We refer
to hj,eff as an effective column, which represents the sum of
Fig. 1. GSSK system model. nt distinct columns in H.
Remark 1: Only nt columns of H are activated, and these
At the receiver side, the GSSK detector estimates the columns change depending on the transmitted information.
antenna indices that are used during transmission, and demaps Now that we have seen the mapping rule, let us describe
the symbol to its component bits b̂. how the receiver estimates the transmit antenna indices j.

A. Transmission B. Detection
The underlying concept in GSSK is using only antenna The detector’s main function is obtaining the antenna in-
indices to relay information. In general, combinations of dices used at the transmitter. Since the channel inputs are
antenna indices can be used.
¡ t¢Therefore, for GSSK using nt 1 For our purposes, we consider real values for x , and consider more
j
antennas, there are M 0 = Nnt possible constellation points. general (adaptive) symbol design in ongoing work.
assumed equally likely, the optimal detector is ML, which is assumed to be known at the receiver, and remains the same
given by regardless of the transmitted information. If, on the other
hand, both the antenna indices and the symbols conveyed
k = arg max pY (y | xj , H)
j information, the modulation scheme would not be APM, but
°
° p °2
° rather a form of SM. In this case, X eff is similar to GSSK, but
= arg min °y − ρ0 hj,eff ° with the possibility of having more than one scaled version of
j F
½³ √ 0 ´H ¾ xj,eff along the same direction.
ρ
= arg max Re y − 2 hj,eff hj,eff , (2)
j III. P ERFORMANCE A NALYSIS
where k ∈X represents the estimated antenna indices, and
pY (y | xj , H) is given by A. Error Probability
1 ³ √ ´ GSSK’s performance is derived using the well known union
pY (y | xj , H) = Nr exp − ky − ρHxj k2F . (3) bounding technique [8, p. 261-262]. The average BER for
π
Remark 2: The detection rule is a maximization problem GSSK is union bounded as
" #
over all effective columns of H (there are M of them). Es- X
sentially, these effective columns act as random constellation Pe,bit ≤ Exj N (j, k)P (xj → xk )
k
points for GSSK modulation. X X N (j, k)
= P (xj → xk ) , (4)
C. Constellation M
j k
We now look into GSSK’s constellation in more detail
and highlight some of its strength. Consider a fixed channel where N (j, k) is the number of bits in error between the
realization H, and the effective Nr −dim constellation symbol constellation vector xj and xk , and P (xj → xk ) denotes the
xj,eff = Hxj , shown in Fig. 2 for APM and GSSK. In APM, pairwise error probability (PEP) of deciding on xk given that
the effective constellation X eff is composed of scaled versions xj is transmitted. By using (2), the PEP conditioned on H is
of the vector hxj . But in GSSK, X eff is made up of scaled given by
¡√ ¢
versions of all effective columns of H. Decisions for APM P (xj → xk | H) = P (dk > dj | H) = Q κ ,
are performed in the 1 − dim complex space, independent ½³ ´H ¾
√ 0
of which antenna is used (since after matched filtering, the where dj = Re y − 2 hj,eff ρ
hj,eff and Q (x) =
sufficient statistics are scalar). On the other hand, GSSK R ∞ 1 − t2
decisions are made in the Nr −dim space. We therefore expect x 2π
e 2 dt. We define κ as
GSSK to outperform APM schemes for increasing M and Nr . 2Nr
Also, better performance is achieved in GSSK for channel ρ0 X
κ, khj,eff − hk,eff k2F = α2n , (5)
realizations having effective columns that are widely spread 2 n=1
apart in the Nr − dim space, which depend on the stochastic ¡ ¢
properties of the channel and can be capitalized upon for where αn ∼ N 0, σ2α , σ2α = ρd(j,k) 4nt
, and d (j, k) is the
adaptive forms of GSSK. As well, X can be formulated to number of distinct columns of H between hj,eff and hk,eff .
optimize the distance spectrum of GSSK’s X eff , by exploiting Remark 4: The metric affecting the system performance is
GSSK’s degrees of freedom, namely nt and Nt . the distance between the effective columns of H. We hinted
at this observation earlier in Section II-C, when analyzing
x2,eff = hx2 x3,eff = h3,eff GSSK’s constellation. Also, we can choose hj,eff ’s (i.e. X )
x1, eff = h1,eff such that large d (j, k)’s are obtained, which translates into
x1, eff = hx1 achieving better performance.
h The random variable κ in (5) is chi-squared distributed with
APM GSSK
s = 2Nr degrees of freedom, and PDF pκ (v) given by [8, p.
x2,eff = h2,eff 41]. The PEP can then be formulated as
P (xj → xk ) = Eκ [P (xj → xk | H)]
Fig. 2. Illustration of the effective constellation space X eff .
Z∞
¡√ ¢
= Q v pκ (v) dv, (6)
Remark 3: It is using the column indices of H as the
source of information that results in the improved constellation v=0

space, and not the fact that different columns are being used which has a closed form expression given in [9, Eq. (64)].
for transmission. For example, if APM with transmission Thus,
on alternating antenna indices is considered, X eff would not Nr −1 µ ¶
X Nr − 1 + k
change since at any given time, the receiver explores only P (xj → xk ) = γ N r
[1 − γ α ]k , (7)
α
all possible transmit symbols. The actual antenna indices are k=0
k
 ³ q 2 ´
σα
where γ α = 12 1 − 1+σ 2 , and is a function of j and k. we consider only minimizing the parameter γ α instead. Hence,
α
Plugging (7) into (4), we obtain the second constellation design rule is given by
XX
µ ¶ θ̂2 = arg max σ 2α
1 XX
NXr −1
Nr − 1 + k
Pe,bit ≤ N (j, k)γ Nα
r
[1 − γ α ]k . θ2
j k
M k
j k k=0 1 XX
(8) = arg max d (j, k) (12)
θ2 nt j k
B. Diversity ³ ´
In order to clearly show the system diversity, we re-derive where θ̂2 = N̂t , n̂t , X̂ , θ 2 = (Nt , nt , X ). Therefore, for a
the error probability with
³ a2 ´loose upper bound. Specifically, given Nt and nt value, we maximize the values of d (j, k).
we use Q (x) ≤ 2 exp − x2 [8, p. 54], and upper bound (6)
1 This maximization can be interpreted as having antenna com-
by binations that are as different from one another as possible.
³ ´ Widely varying antenna combinations imply having a large
Z ∞ exp ¡− v ¢ vNr −1 exp − v2 M 0 , which can be increased by choosing a larger Nt , and
2 2σα
P (xj → xĵ ) ≤
N +1 Nr
dv thus increasing the transmitter’s overhead. Note that a large
0 2 r σα Γ (Nr ) nt will also help, but will have adverse effects as well due
1¡ 2 ¢−Nr
to the n1t factor in (12). Depending on system requirements
= σ +1
2 α (i.e. low hardware overhead, or high performance), the range
µ ¶−Nr
d (j, k) of requirements for Nt and nt can be specified, from which
≤ 22Nr +1 ρ−Nr .
nt the optimal combination of Nt and nt can be obtained by (12)
Therefore, the bit error probability is given by through computer search.
Remark 5: The optimal set of parameters may not be
Pe,bit ≤ Cρ−Nr , (9) unique, since several sets may result in identical performance.
X X N(j,k) ³ ´−Nr Also, this trade-off between transmitter complexity and per-
2Nr +1 d(j,k)
where C = M 2 nt . We clearly see formance provides design flexibility, which can be exploited
j k in adaptive type systems.
from (9) that a diversity order of Nr is achieved, which is Figure 3 illustrates GSSK’s performance bounds given by
the same as that of an MRC-APM system using Nr receive (8), with M = 8 and M = 32, and for varying Nt ’s. For each
antennas. plot, nt and X are obtained by computer simulations using
IV. O PTIMAL C ONSTELLATION D ESIGN (12).
We consider the optimal formulation of GSSK’s constel- 10
0

lation X , in terms of minimizing the bit error rate. The

probability of bit errors given by (8) is minimized by the
following joint optimization problem: 10
-1

θ̂opt = arg min Pe,bit , (10)

θopt
³ ´ 10
-2

GSSK bound (N =8, n =3)

where θ̂ opt = N̂t , n̂t , X̂ , μ̂ , θ opt = (Nt , nt , X , μ) with t t
GSSK bound (N =9, n =2)
Pe,bit

constraint nt < Nt and M 0 ≥ M , μ is the labeling rule t t

GSSK bound (N =10, n =2)
t t
for the constellation X (i.e. the rule for labeling m bits to 10
-3
GSSK bound (N =32, n =1)
t t
a symbol vector xj ), and Pe,bit is given by (8). We note that GSSK bound (N =5, n =2)
t t

in (8), N (j, k) is affected by μ, whereas γ α is affected by GSSK bound (N =6, n =2)

t t
GSSK bound (N =7, n =2)
nt , Nt , and X . To simplify the design process, we present two 10
-4 t
GSSK bound (N =8, n =1)
t

sub-optimal constellation design rules. t t

In the first algorithm, we simplify the optimization problem M=32

M=8
by considering relatively high SNRs. Hence, (10) reduces to
0 5 10 15 20 25 30
XX ρ (dB)
θ̂ 1 = arg min N (j, k)γ α , (11)
θ1
j k
Fig. 3. GSSK bounds for varying M , Nt , and nt (Nr = 2).
where we only keep the terms contributing most to Pe,bit ,
and θ̂1 and θ 1 have the same parameters as θ̂opt and θopt , As expected, the performance degrades as Nt decreases, and
respectively. Still, the optimization is fairly complex (remem- as nt increases. We also note that certain antenna transitions
ber, for each Nt chosen, there are Nt − 1 possibilities for nt , do not gain much in performance. For example, only a few
¡ Nn t)¢
from which there are (M t available antenna combinations to tenths of a dB is gained from the transition of Nt = 9 to
formulate X ). Therefore, to simplify the optimization further, Nt = 10 (M = 32).
 V. S IMULATION RESULTS complexity reduction is attributed to the fact that symbols
In this section, we compare GSSK’s performance for vary- do not carry information (as in SM, MRC, and V-BLAST),
ing Nt and nt . Monte Carlo simulations are run for at least 105 therefore reducing the optimization problem’s overhead when
channel realizations. We use Gray (or quasi-Gray) mapping detecting the message. We also note that, as expected, GSSK’s
when appropriate (i.e. for PSK and QAM modulation). performance degrades as Nt is decreased, but still outperforms
In Fig. 4, we demonstrate GSSK’s performance versus V- APM in most cases.
BLAST, MRC, and SM. We target m = 3 bits/s/Hz trans- VI. C ONCLUSION
mission, and consider Nr = 4. For reference, we use three
different transmission setups. The first one is APM, 8-QAM In this paper, we introduced a new modulation method
(referred to as GSSK) for MIMO wireless links by exploit-
transmission with Nt = 1 (single antenna transmission), and
ing the inherent fading process. Rather than transmitting
M = 8. The second is V-BLAST with BPSK modulation,
information through symbols, the transmitter antenna indices
Nt = 3, and ordered successive interference cancellation
were used as the sole information conveying mechanism.
(OSIC) with the minimum mean squared error (MMSE) re-
ceiver [2]. Third, SM [3] using optimal detection [7] with Throughout the paper, we laid out GSSK fundamentals as
the building ground for hybrid modulation schemes (which
BPSK modulation, and Nt = 4 antennas is shown.
combine GSSK and APM) such as in [6]. We presented GSSK
We plot GSSK for two different constraints. The first is with
improvements over APM (up to 3 dB), and derived closed
Nt = M = 8, and hence, nt = 1. The second is with Nt = 5,
form upper bounds on the bit error probability. All of SM’s
nt = 2, and X obtained from (12). The bounds of (8) are also
merits mentioned in [6] are also inherent in GSSK (at similar
plotted for comparison. The simulation and analytical results
performance), but with lower computational overhead, and
are a close match, especially for high SNRs. The bounds are
greater design flexibility. These advantages make GSSK a
tighter for nt = 2 than nt = 1, and this is due to the fact that
promising candidate for low complexity transceivers in next
summing over all possible constellation points (for the union
generation communication systems. Future research directions
bound) is more justified in the case of nt = 2. When nt = 1,
will involve the adaptive case, where GSSK’s constellation
the constellation points are all unique (i.e. distinct columns
can take advantage of channel conditions. We also intend to
of H), resulting in a union bound that is more loose (in this
investigate GSSK’s robustness to non-ideal channel conditions,
case, the nearest neighbor approximation is better suited). On
and practical GSSK implementation issues in current MIMO
the other hand, with nt = 2, the constellation points share
communication standards.
common columns of H, and a larger number of constellation
points have an effect on performance. Therefore, the union R EFERENCES
bound for these higher nt values result in tighter bounds. [1] P. Wolniansky, G. Foschini, G. Golden, and R. Valenzuela, “V-BLAST:
an architecture for realizing very high data rates over the rich-scattering
10
0
wireless channel,” in Proc. International Symposium on Signals, Systems,
GSSK (N =5, n =2)
t t
and Electronics (ISSSE’98), Pisa, Italy, pp. 295-300, Sept.-Oct. 1998.
GSSK (N =7, n =2)
t t
[2] R. Böhnke, D. Wübben, V. Kühn, and K. D. Kammeyer, “Reduced
MRC-QAM (M=8, N =1) complexity MMSE detection for BLAST architectures,” in Proc. IEEE
t
10
-1
V-BLAST (M=2, N =3) Globecom’03, San Francisco, California, USA, Dec. 2003.
t
GSSK (N =8, n =1)
[3] R. Mesleh, H. Haas, C. Ahn, and S. Yun, “Spatial modulation - a
t t
new low complexity spectral efficiency enhancing technique,” First in-
SM [7] (M=2, N =4)
t ternational Conference on Communications and Networking in China
10
-2
GSSK bound (N =5, n =2)
(ChinaCom’06), pp. 1-5, Oct. 2006.
t t [4] R. Mesleh, H. Haas, C. W. Ahn, and S. Yun, “Spatial modulation-
GSSK bound (N =7, n =2)
Pe,bit

t t OFDM,” In the Proceedings of the 11th International OFDM-Workshop

GSSK bound (N =8, n =1) 2006 (InOWo’06), pp. 288–292, August 2006.
t t
-3
10 [5] S. Ganesan, R. Mesleh, H. Haas, C.W. Ahn, S. Yun, “On the performance
of spatial modulation OFDM,” Fortieth Asilomar Conference on Signals,
Systems and Computers(ACSSC ’06), pp. 1825-1829, Oct.-Nov. 2006.
[6] R. Mesleh, H. Haas, S. Sinanović, C. W. Ahn and S. Yun, “Spatial
-4
10 modulation,” IEEE Trans. Vehicular Technology, to appear.
[7] J. Jeganathan, A. Ghrayeb, and L. Szczecinski, “Spatial Modulation:
Optimal Detection and Performance Analysis,” IEEE Commun. Letters,
to appear.
-5
10 [8] J.G. Proakis, Digital Communications, (4th ed.) McGraw-Hill, New York,
0 2 4 6 8 10 12 14 16 18 20
ρ (dB) 2001.
[9] M.-S. Alouini and A. Goldsmith, “A unified approach for calculating error
rates of linearly modulated signals over generalized fading channels,”
Fig. 4. BER performance of GSSK versus MRC-QAM, V-BLAST, and SM, IEEE Trans. Commun., vol. 47, no. 9, pp. 1324–1334, September 1999.
for m = 3 bits/s/Hz transmission (Nr = 4).

GSSK’s performance improvements is clearly shown in the

figure, where we observe gains of 3 dB over APM, and 1 dB
over V-BLAST (for Pe,bit = 10−5 ). GSSK has almost identical
performance to that of SM, but with lower complexity. This

View publication stats