Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

EuCAP 2012 1569523381

An Improved Spatial Modulation Scheme for


MIMO Channels
J.M. Luna-Rivera and M. G. Gonzalez-Perez
1
Facultad de Ciencias, UASLP, Av. Salvador Nava s/n, Zona Universitaria, San Luis Potosi, México
Phone/Fax: +52 (444) 8262491/8262384, ext. 2920, Email: mlr@uaslp.mx

Abstract—Spatial modulation (SM) has recently emerged as modulation method. The contribution of this paper is two-fold:
a very promising technique to exploit further the potential 1) to provide a new and simple spatial modulation mapping
of multiple-input multiple-output (MIMO) systems using low- rule, referred as ESM, which allows all transmit antennas at a
complexity transceivers. Based on this concept, this paper pro-
poses a novel spatial modulation technique named Extended time to convey information even those that remain inactive; 2)
Spatial Modulation (ESM). The proposed method maps the to propose a two-stage adaptive detector that maximizes the
incoming bits to the transmission vector in a new and simple attainable receive diversity gain by increasing the minimum
way. Results show that ESM inherits most of the SM properties distance between constellation points at the receiver. It will
but more important increases the system’s spectral efficiency be shown that ESM enhances the system’s spectral efficiency
linearly with the number of transmit antennas rather than
a base-two logarithm factor like other existing SM schemes. linearly with the number of transmit antennas, moreover, it
Furthermore, ESM eliminates the constraint of SM to work with does not restrict the transmitter to configurations of antennas
a number of transmit antennas in power of two. In addition, elements which are only powers of two, in fact it extends its
a two-stage adaptive detector is also introduced for SM-based application to work with any number of transmit antennas.
techniques. From simulation results, it is shown that the proposed More importantly, ESM provides a more flexible structure
ESM scheme achieves better tradeoffs between capacity and
performance/complexity as compared with standard SM schemes suitable for a wider range of applications than SM, where
over time-varying Rayleigh flat-fading channels. a tradeoff between performance and diversity gains must
Keywords— Spatial Modulation, MIMO Channels, Digital be chosen with affordable complexity. Simulations results of
Modulation. bit error rate (BER) and spectral efficiency demonstrate the
benefits of the proposed scheme as compared with other SM-
I. I NTRODUCTION based schemes.
Past work has shown that MIMO antenna systems represent The organization of the paper is described next. First,
a key technology for current and future wireless commu- in Section II, we introduce the proposed ESM transmission
nication standards [1]- [2]. SM is a recently transmission scheme. Section III presents the two-stage adaptive detector
scheme for MIMO systems proposed in [3]. Unlike other which is intended to improve the spatial diversity at the
MIMO schemes, SM exploits the spatial position of each receiver of SM-based techniques. In Section IV, an analyzes
transmit antenna as an extra source of information. In addition of the overall complexity for the proposed ESM scheme is
to this, SM enjoys of relatively low-complexity structures at provided and compared with the complexity demanded by the
the transmitter where it activates only one transmit antenna original SM scheme. Simulation results are reported in Section
at each time, avoiding the use of multiple radio frequency V. Finally Section VI draws the conclusions and future work.
chains. The simplest form of SM, referred as Space Shift
Keying (SSK) modulation, is introduced in [4] where the II. E XTENDED S PATIAL M ODULATION
information is only represented by the presence or absence Figure 1 considers a generic MIMO system with 𝑁𝑇
of energy at the antenna activated for transmission. Because transmit antennas and 𝑁𝑅 receive antennas. The input bit
SSK uses only the antenna indices to relay information, its stream, 𝑑1 , 𝑑2 , ⋅ ⋅ ⋅ , is sent through the ESM modulator where
complexity for detection is lowered as compared to SM. groups of 𝑛𝑏 bits are mapped to the transmission vector
In general, SM schemes abstain from the use of advanced x = [𝑥1 , ⋅ ⋅ ⋅ , 𝑥𝑁𝑇 ]𝑇 ∈ ℂ𝑁𝑇 . The basic concept of ESM is to
detection mechanisms due to the absence of interference in activate an antenna at the transmitter based on the input data
the system. In [5], the optimum maximum-likelihood (ML) bits unique pattern. Different to SM, where only one antenna is
detector is derived for both SM and SSK. A generalized SSK activated to carry information at each time, ESM may activate
(GSSK) scheme is presented in [6], GSSK basically allows the more than one transmit antenna depending on the incoming
activation of a subset of the transmit antennas rather than just bits pattern at the modulator. ESM will allow all transmit
one. Other research studies conducted to analyze or improve antennas to convey information even those that remain inactive
the spatial modulation performance can be found in [7]- [10] during transmission. To illustrate this idea easily but without
and references therein. loss of generality, let us consider the example given in Table
Based on the SM concept, this paper tackles some limita- I for a system using 𝑁𝑇 = 3. Observe that when the input bit
tions of SM by approaching this concept as a classic digital is one, it activates the corresponding antenna directly without

1
1 1 η
1
1
2 MIMO 2 Ns out 2 η2 ^
d Extended ML d
Channel of N r

...
...
SM Detector

...
Input (ESM) (H) switch ηS
data Ns
bits NT NR
x y

Fig. 1. Block diagram of the considered MIMO system.

the need of a given pre-defined mapping table. Notice also that antenna combinations. For instance, the 𝑁𝑇 × 1 transmission
the elements of the transmission vector are normalized to√keep vector x representing the 𝑛𝑏 -bit zero word for 𝑁𝑇 = 3 and
−1 √1 √1 𝑇
with the energy constraints of the system, i.e. 𝑥𝑗 = (𝑑𝑗 / 𝑤), 𝑀 = 1 can be defined as x = [ √ 3
, 3 , 3 ] . Similarly, any
where 𝑤 denotes the hamming weight of the input word, d. combination of positive and negative amplitudes may be used
for 𝑀 > 2, however, a further analysis is required to define an
TABLE I optimal criterion. Table II illustrates an example of the ESM
ESM SCHEME USING 𝑁𝑇 = 3.
mapping procedure for a MIMO system where 𝑁𝑇 = 3 and
d= [𝑑1 𝑑2 𝑑3 ] x= [𝑥1 𝑥2 𝑥3 ]𝑇 𝑀 = 2 are assumed.
0 0 0 [0 0 0 ]𝑇 TABLE II
0 0 1 [0 0 1 ]𝑇 ESM SCHEME USING 𝑁𝑇 = 3 AND 𝑀 = 2.
0 1 0 [0 1 0 ]𝑇
1 1 𝑇
0 1 1 [0 √ √ ]
2 2 d= [𝑑1 𝑑2 𝑑3 𝑑4 ] x= [𝑥1 𝑥2 𝑥3 ]𝑇
1 0 0 [1 0 0]𝑇 1
0 0 0 0 √ [+1 −1 −1]𝑇
1 0 1 [ √1 0 1 𝑇
√ ] 1
3
2 2
0 0 0 1 √ [−1 +1 +1]𝑇
1 1 0 [ √1 1
√ 0 ]𝑇 3
2 2 0 0 1 0 [0 0 −1]𝑇
1 1 1 [ √1 1
√ 1 𝑇
√ ]
3 3 3 0 0 1 1 [0 0 +1]𝑇
0 1 0 0 [0 −1 0]𝑇
0 1 0 1 [0 +1 0]𝑇
It is the aim of ESM to generalize the application of SM √1 [0
0 1 1 0 −1 −1]𝑇
by removing the constraint of 𝑁𝑇 being a power of two. 2
√1 [0
0 1 1 1 +1 +1]𝑇
Moreover, in the same fashion that SM if 𝑁𝑇 is fixed, the 2
1 0 0 0 [−1 0 0]𝑇
spectral efficiency of ESM can still be enhanced by increasing 1 0 0 1 [+1 0 0]𝑇
the number of constellation points, 𝑀 , that can be transmitted 1 0 1 0 [−1 0 −1]𝑇
on each antenna. Therefore, the total number of data bits 1 0 1 1 [+1 0 +1]𝑇
1
transmitted with ESM at each time is 1 1 0 0 √ [−1 −1 0]𝑇
2
1
1 1 0 1 √ [+1 +1 0]𝑇
2
𝑛𝑏 = 𝑁𝑇 + 𝑙𝑜𝑔2 (𝑀 ). (1) 1 1 1 0 1

3
[−1 −1 −1]𝑇
1
1 1 1 1 √ [+1 +1 +1]𝑇
3
Recall that the transmission efficiency for SM grows loga-
rithmically with 𝑁𝑇 and 𝑀 , i.e. 𝑛 ˆ 𝑏 = 𝑙𝑜𝑔2 (𝑁𝑇 ) + 𝑙𝑜𝑔2 (𝑀 ).
In Table I, 𝑀 is equal to one. When 𝑀 > 1, ESM uses After the transmission of x, the received signal is given by
only the first 𝑁𝑇 bits of the 𝑛𝑏 -bit input word as the pattern √
y = 𝜌Hx + 𝜼 (2)
for the antenna activation. Furthermore, adding the zero as
a spatial constellation point reduces the Euclidean distance where H denotes the 𝑁𝑅 × 𝑁𝑇 channel matrix with complex
between all constellation points and degrades the performance gains ℎ𝑟,𝑡 ∼ 𝒞𝒩 (0, 1), 1 ≤ 𝑟 ≤ 𝑁𝑅 , 1 ≤ 𝑡 ≤ 𝑁𝑇 , and 𝜼 ∈
significantly. To overcome this disadvantage, the zero trans- ℂ𝑁𝑅 the noise vector whose components are 𝒞𝒩 (0, 1). The
mission vector x = [0, ⋅ ⋅ ⋅ , 0]𝑇 is replaced by a vector with signal to noise ratio (SNR) at each receive antenna is denoted
non-zero elements and different from the other 𝑁𝑇 −1 possible by 𝜌.

2
problem:
2
d̂ = arg max 𝑝𝑌 (𝒚∣x𝑖 , 𝑯), x𝑖 ∈ ℂ𝑁𝑇 , 1 ≤ 𝑖 ≤ 𝐿 (3)
𝑖
1.5
where d̂ defines the estimated word of 𝑛𝑏 bits over the whole
1 set of 𝐿 possible transmit sequences, and 𝑝𝑌 (𝒚∣x𝑖 , 𝑯) the
probability density function for the received signal, which is
0.5
h h given by
1,4 1,2
( √ )
exp −∥𝒚 − 𝜌Hx𝑖 ∥2𝐹
imag

0 𝑝𝑌 (𝒚∣x𝑖 , 𝑯) = . (4)
h
1,3 𝜋 𝑁𝑅
−0.5 Equivalently, we see that the maximization problem in (3) can
be rewritten as the following minimization problem:
−1 h √
1,1
d̂ = arg min ∥𝒚 − 𝜌Hx𝑖 ∥2𝐹 . (5)
𝑖
−1.5
From the above optimization problem, it can be shown that the
error performance will be dictated by the Euclidean distance
−2
−2 −1.5 −1 −0.5 0 0.5 1 1.5 2 among the vectors, i.e.
real
a𝑖 ≜ Hx𝑖 (6)
Fig. 2. Illustration of the constellation points at the receiver for a time- where a𝑖 ∈ ℂ𝑁𝑅 for all 𝑖. Since the constellation points for
invariant Rayleigh flat-fading MIMO channel using SSK with 𝑁𝑇 = 4 and
𝑁𝑅 = 1. ESM at the receiver depend on the complex channel gains,
ℎ𝑟,𝑡 , then the constellation diagram behaves randomly accord-
ing to these values. Observe that for a larger 𝑁𝑅 , the Euclidean
distance between all constellation points is increased. How-
III. ESM D ETECTION ever, this advantage is affected when the transmit-receive wire-
less links suffer large attenuations, causing the constellation
Previous works in SM emphasize the fact that informa- points to get closer and closer to each other which degrades the
tion can be carried not only by the symbols themselves error performance. Therefore, we propose an adaptive antenna
but also by the antenna indices. However, from the error selection (AAS) scheme before the ML detection process to
performance perspective and assuming a maximum likelihood compensate for this loss, see Figure 1. The selection criterion
(ML) strategy at the receiver, it can be shown that the error consists on choosing the best 𝑁𝑆 out of the 𝑁𝑅 antenna
performance is essentially bounded by the difference among elements available at the receiver with the highest gains. To
the complex gains of the wireless MIMO channel regardless do this, we estimate the vector v ∈ ℝ𝑁𝑅 given by
of the antenna used during transmission. In other words, the 𝑁𝑇

achievable error performance of SM under energy constraints v(𝑟) = ∥ ℎ𝑟,𝑡 ∥2 , 1 ≤ 𝑟 ≤ 𝑁𝑅 . (7)
depends on the Euclidean space constructed by the constel- 𝑡=1
lation points diagram of size 𝑀 ⋅ 2𝑁𝑇 which is determined We hint that the elements in v are used here as indicators of
by the MIMO channel realizations. Therefore, like any other the spatial diversity gain on each receive antenna. Thus, the
conventional digital modulation scheme a trade-off between 𝑁𝑆 antennas at the receiver with the highest spatial diversity
spectral efficiency (i.e. constellation size) and performance gains (highest 𝑁𝑆 values in v), where 𝑁𝑆 < 𝑁𝑅 , are used as
must be defined. In Figure 2, the complex received signal input to the ML detector.
(circles) in (2) is depicted for several transmissions over a
time-invariant Rayleigh flat-fading MIMO channel using SSK IV. C OMPLEXITY
with 𝑁𝑇 = 4 and 𝑁𝑅 = 1. This example shows a transmission In this section, we discuss and compare the complexity
scheme that achieves a spectral efficiency of 2 bits/s/Hz for a required by the proposed ESM scheme. We first address the
𝑆𝑁 𝑅 = 20 dB. For reference, the noiseless received points are hardware requirements of ESM at the transmitter which is
also plotted (asterisk). Therefore, we conclude that the BER clearly summarized with the number of RF chains needed.
performance will be dictated by the smallest distance between Although ESM uses a random number of antenna elements up
the noiseless received points, which, in this case depends on to 𝑁𝑇 during transmission, only one RF chain is needed since
the links using the transmit antennas 2 and 3, see Figure 2. In the same symbol is transmitted through all activated antennas.
this context, we highlight the fact that ESM maintains these Observe that this condition is preserved for the zero vector
same features. where ESM replaces it with a vector that combines positive
We now develop the optimal detector for the proposed ESM and negative versions of the transmitted symbol but with a
scheme under the framework presented in Figure 1. The ML deterministic pattern. Therefore, ESM works with only one
detector [5] is realized by solving the following optimization RF chain in the same fashion that the SM and SSK schemes.

3
TABLE III
From the receiver viewpoint, the computational effort of S PECTRAL EFFICIENCY FOR SM AND ESM AS A FUNCTION OF 𝑁𝑇 .
the ML detection is given in the order of 𝑂(𝐿) due to the
searching process over all possible noiseless received signals Spectral efficiency (𝛾) b/s/Hz
𝑀 =1 𝑀 =2
(constellation size). Therefore, it is necessary to find a compro-
𝑁𝑇 SM ESM SM ESM
mise between the constellation size (spectral efficiency) and 2 1 2 2 3
the complexity of this detector which increases exponentially 3 NA 3 NA 4
with the number of transmit antennas. However, it is remarked 4 2 4 3 5
5 NA 5 NA 6
that ESM can achieve considerable high data rates with a 6 NA 6 NA 7
relatively small number of antenna elements. Alternatively, 7 NA 7 NA 8
suboptimum detection approaches may be favored against the 8 3 8 4 9
NA= Not Available
ML scheme but at the expense of some error performance
degradation. In addition, the AAS scheme will reduce the
hardware complexity at the receiver by using fewer RF chains 0
10
with the benefit of some performance gain. SM (NR=1)
ESM (NR=1)
SM (NR=2)
V. S IMULATION R ESULTS
−1 ESM (NR=2)
10
Some performance and spectral efficiency results are now
carried out for the propose ESM scheme over time-varying
Rayleigh flat-fading channels. Let us first discuss the spectral
BER
−2
10
efficiency behavior of ESM which is given by
𝑛𝑏
𝛾 =
𝑇𝑠 ⋅ 𝑊 −3

𝑁𝑇 + log2 (𝑀 )
10

= (bits/s/Hz) (8)
𝑇𝑠 ⋅ 𝑊
where 𝑛𝑏 denotes the number of bits transmitted in a given −4
10
bandwidth, 𝑊 , for a system that uses 𝑁𝑇 transmit antennas 0 2 4 6 8 10
SNR (dB)
12 14 16 18 20

and 𝑀 constellation points per transmit antenna. The scalar 𝑇𝑠


defines the duration of the transmitted signal that for simplicity Fig. 3. BER performance of ESM versus SM for a 2 b/s/Hz with 𝑁𝑇 = 2
it is set here to 𝑇𝑠 = 1. Using (8), the spectral efficiency and 𝑁𝑅 = 1 and 2.
of ESM is presented in Table III for different number of
transmit antennas (𝑁𝑇 ). These results are compared with those
obtained by the SM scheme [3]. Two cases are illustrated: 𝑁𝑅 = 2. The better performance of ESM is mainly due to the
i) using 𝑀 = 1, i.e. SM reduces to SSK, and ii) using Euclidean space constructed at the receiver which it clearly
𝑀 = 2, i.e. BPSK signals are transmitted on each antenna. increases the minimum distance between the signal points of
Clearly, the results in Table III show that the proposed ESM the constellation diagram as compared with conventional SM.
scheme outperforms SM in terms of spectral efficiency. For Let us now consider the performance of ESM using the AAS
example, ESM achieves more than twice the spectral efficiency scheme introduced in Section III. We consider an effective
of SM under the configuration of 𝑁𝑇 = 8 and 𝑀 = 2. 2 × 2 MIMO system with 𝑁𝑅 = 2, 4 and 8 receive antennas,
Moreover, ESM works for any number of antenna elements i.e. from (7) the best 𝑁𝑆 = 2 antenna elements are chosen
at the transmitter while SM is restricted to a number which is from the different set of array with 𝑁𝑅 = 2, 4 and 8 at the
a power of two. Although it is not presented in Table III, SM receiver. The results in Fig. 4 show a significant additional
has the disadvantage of requiring a large number of transmit gain with the two-stage adaptive detector. It can be seen a
antennas for high data rates, which grows exponentially with performance gain of 4 dB for a 𝐵𝐸𝑅 = 10−3 when 𝑁𝑅 = 8
𝑁𝑇 and 𝑀 , a feature that is alleviated with ESM. Observe also receive antennas are considered but where only two RF chains
that the antenna hardware requirements for both ESM and SM are employed. It is important to remark that the proposed two-
remain unchanged regardless of the value of 𝑀 . stage adaptive detector can be applied to any other SM-based
Figure 3 shows the ESM’s BER performance over a transmission scheme with similar results.
Rayleigh fading channel operating at the spectral efficiency Finally, we highlight the benefits of the ESM scheme by
of 𝛾 = 2 bits/s/Hz using 𝑁𝑇 = 2 and 𝑀 = 1. These results showing its performance for a spectral efficiency of 4 bits/s/Hz
are compared with those obtained by SM employing the same in Fig. (5) and compares it with the performance achieved
number of transmit antennas. Notice that in order to achieve by the standard SM. The following two cases are considered:
the spectral efficiency of 2 bits/s/Hz SM has to set 𝑀 = 2. At ESM uses 𝑁𝑇 = 3 and 𝑀 = 2 with the two-stage adaptive
the receiver both schemes perform only ML detection without detector at the receiver; the second setup is SM with 𝑁𝑇 = 8
the AAS block for 𝑁𝑅 = 1 and 𝑁𝑅 = 2. It can be seen that transmit antennas and 𝑀 = 2, ML detection is only performed
ESM outperforms SM about 4 dB for 𝑁𝑅 = 1 and 3 dB for at the receiver. The aim of the results presented in Fig. (5) is

4
−1
10
in practice when the physical size of the transmission unit is
ESM (N =2, N =2)
R S restricted.
ESM (N =4, N =2)
R S In addition, a two-stage adaptive detector has been proposed
ESM (NR=8, NS=2)
and its performance studied over time-varying Rayleigh flat-
−2
fading channels. Although the AAS block in the two-stage de-
10
tector augment the number of antenna elements at the receiver,
the RF chains remain unchanged. Improved spatial diversity
BER

gains are obtained by the proposed two-stage detector.


Finally, it is concluded that the proposed ESM scheme
−3
10 achieves more flexible and practical tradeoffs between capacity
and performance/complexity as compared with existing SM
schemes. Future work will focus on improving the energy
efficiency of the ESM scheme.
−4
10
0 2 4 6 8 10 12 14 16 18 20 ACKNOWLEDGEMENT
SNR (dB)
This research work was supported by CONACYT under the
Grant No. 83811.
Fig. 4. BER Performance of ESM with AAS for a 2 b/s/Hz considering
𝑁𝑇 = 2 and 𝑁𝑆 = 2. R EFERENCES
[1] P. Wolniansky, G. Foschini, G. Golden, and R. VAlenzuela, “V-BLAST:
0
10 an architecture for realizing very high data rates over the rich-scattering
SM (NR=2) wireless channel”, in Proceedings International Symposium Signals,
ESM (NR=4, NS=2) Systems, Electronics (ISSSE’98), Pisa, Italy, pp. 295–300, Sept. 1998.
[2] B. Hassibi and B. Hochwald, “High-Rate Codes that are Linear in Space
−1
and Time”, IEEE Transactions on Information Theory, Vol. 48, No. 7,
10
pp. 18041824, July 2002.
[3] R. Mesleh, et. al., “Spatial Modulation”, IEEE Transactions on Vehicular
Technology, Vol. 57, No. 4, pp. 2228–2241, July 2008.
[4] J. Jeganathan, A. Ghrayeb, L. Szczecinski, and A. Ceron, “Space
BER

−2
10 Shift Keying Modulation for MIMO Channels”, IEEE Transactions on
Wireless Communications, Vol. 8, No. 7, pp. 3692–3703, July 2009.
[5] J. Jeganathan, A. Ghrayeb, and L. Szczecinski, “Spatial Modulation:
optimal detection and performance analysis”, IEEE Communications
−3
10 Letters, Vol. 12, No. 8, pp. 545–547, 2008.
[6] J. Jeganathan, A. Ghrayeb, and L. Szczecinski, “Generalized Space
Shift Keying Modulation for MIMO Channels”, IEEE International
Symposium Personal, Indoor, Mobile Radio Communications, pp.1-5,
−4 September 2008.
10
0 2 4 6 8 10 12 14 16 18 20 [7] M. Di Renzo, H. Haas, “Improving the Performance of Space Shift
SNR (dB) Keying (SSK) Modulation via Opportunistic Power Allocation”, IEEE
Communications Letters, Vol. 14, No. 6, pp. 500–502, June 2010.
[8] M. Di Renzo, H. Haas, ‘On the Performance of SSK Modulation
Fig. 5. BER performance of ESM versus SM for a 2 b/s/Hz with 𝑁𝑇 = 2 over Multiple-Access Rayleigh FAding Channels”, In Proceeding of
and 𝑁𝑅 = 1, 2, 4 and 8 (𝑁𝑆 = 2). GLOBCOM 2010, pp. 1–6, June 2010.
[9] R. Mesleh, I. Stefan, H. Haas, M. Di Renzo, P.M. Grant, “Trellis Coded
Spatial Modulation”, IEEE Transactions on Wireless Communications,
to illustrate that ESM can achieve the same performance than Vol. 9, No. 7, pp. 2349–2361, July 2010.
[10] S. Sugiura, S. Chen, and L. Hanzo, “Generalized Space-Time Shift
SM but with a smaller number of transmit antennas which in Keying Designed for Flexible Diversity, Multiplexing and Complexity
this case is 3 instead of 8. The price paid for this reduction in Tradeoffs”, IEEE Transactions on Wireless Communications, Vol. 10.
antenna elements at the transmitter comes at the expense of No. 4, pp. 1144–1153, 2011.
increasing the number of antenna element at the receiver from
2 to 4 but where the number of RF chains remain unchanged.
VI. C ONCLUSIONS
In this paper, a novel spatial modulation scheme named
ESM has been proposed. It is shown that ESM achieves a
spectral efficiency which increases linearly with the number
of transmit antennas rather than a base-two logarithm factor
like other existing SM schemes. Moreover, ESM eliminates the
restriction of SM where the transmitter works only when the
number of transmit antennas is a power of two. Thus, ESM
yields a significant spectral efficiency gain using a reduced
number of transmit antennas. Such property may be critical

You might also like