Вы находитесь на странице: 1из 4

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. Citation information: DOI 10.1109/LCOMM.2017.2733540, IEEE
Communications Letters 1

Precoding-Aided Spatial Modulation with Increased


Robustness to Channel Correlations
Mutlu Koca, Member, IEEE and Hikmet Sari, Fellow, IEEE

Abstract—We present a precoding approach for spatial modu- erroneous version of the channel. However, these works still
lation (SM) for increased robustness against spatial correlations rely on fixed or block-fading channel realizations. That is why,
over slow fading Rayleigh channels. This approach, based on in this work we extend this constellation shaping approach to
phase-rotation of the transmitted symbol according to the active
transmit antenna, can be implemented without changing the slow-fading and correlated Rayleigh channels and present a
power budget or needing any explicit knowledge of the channel rotation-based precoding approach against transmit correlation
state information at the transmitter. The optimum precoding effects as a cost and complexity effective alternative. This
coefficients are determined so as to minimize the asymptotic method is based on imposing a different phase rotation on
average bit-error rate (ABER). Both theoretical analysis and the symbols according to the activated antenna, hence it
simulation results indicate significant performance improvements
even in the case of heavily correlated transmit antennas. continues to avoid ICI, does not change the transmit power
and can still be implemented in a single radio frequency (RF)
structure with only a marginal increase in the circuitry as
I. I NTRODUCTION opposed to the coding based approaches mentioned above. The
rotation optimization is based on minimizing the asymptotic
Spatial modulation (SM) [1], where part of the incoming average bit-error probability (ABEP) at high signal-to-noise
bits are used to choose a single transmit antenna while the ratio (SNR) values. The presented approach is also similar to
rest are used to select the modulation symbol to be sent from the random phase rotation approach of [11]. However, whereas
the active antenna, has emerged as a novel communication the method in [11] requires a random precoding phase set to
technology.Despite having fundamental advantages such as be shared between the ransmitter and the receiver in every
the avoidance of inter-channel interference (ICI) and the signalling instant, the precoding presented in this work relies
simplification of the transmit chain, the joint utilization of on the sharing of a single optimum phase set throughout
both the antenna and signal spaces to convey information in the entire transmission and therefore is more practical. The
SM is also the cause of a main drawback. Specifically, if proposed precoder also differs from multi-antenna precoder
there is spatial correlation among the transmit antennas due approaches given in works such as [12], which achieve ICI
to insufficient separation and/or a direct line-of-sight channel free signals at the receiver at the cost of losing the single
component (i.e., Rician fading), the receiver performance in RF structure at the transmitter, as it can be implemented with
resolving the active transmit antenna degrades significantly. both conventional and generalized versions of SM and while
This in turn results in a large increase in the overall average preserving the single RF structure during precoding.
bit-error rate (ABER) of the SM transceiver as well. This Compared with the existing precoding approaches, the con-
problem, that is the unequal error vulnerability of the antenna tributions of this work are as follows: i) Slow fading Rayleigh
and symbol bits against adverse fading channel effects, has channels are considered as opposed to fixed channel realiza-
been addressed in [2]–[4] and various unequal error protection tions where the precoder is designed for robustness against
(UEP) solutions based on trellis-coded modulation [2], bit- the spatial correlation effects among the transmit antennas; ii)
interleaved coded modulation [3] and block-Markov superpo- Unlike the approaches presented in [6]–[12], only the channel
sition coding [4] have been proposed. Even though they are type and the correlation parameters are needed at the trans-
not designed specifically for UEP, other alternatives such as mitter and not the explicit channel knowledge. Both analysis
the trellis coding approach of [5] are also shown to provide and simulation results show that precoded SM overcomes
some robustness against these effects as well. However, in significant performance losses even in the case of severe inter-
all of these approaches, an encoder-decoder pair is added antenna correlations in comparison to conventional SM.
to the transceiver which increases both the latency and the
computational complexity.
Recently, another line of work in [6]–[10] has addressed II. S YSTEM M ODEL
the signal shaping by precoding at the transmitter to provide We consider the conventional SM system model with Nt
transmit diversity and performance improvement for SM sys- transmit and Nr receive antennas employing maximum likeli-
tems. Here, the precoder is designed to shape the constellation hood (ML) detection. In this scheme, the number of transmit
observed at the receiver in a way to enhance the minimum antennas is an integer power of 2, i.e., Nt = 2n , and the
Euclidean distance or the bit-error probability. However, these transmitter employs M -ary digital modulation to an m-bit
precoders are designed only for time-invariant channels and m
message where M = 2 and the modulation symbol set is
for cases where the full channel information is available at the

X = X1 , . . . Xk , . . . XM . At each transmission
transmitter. In [9], [10], partial channel state information (CSI) instant, each set of n + m bits is split into groups of n and m
is also considered where the transmitter is assumed to know an bits, and the former is used to select a single transmit antenna
M. Koca is with the Wireless Communications Laboratory, Department
and the latter is mapped onto one of M possible constella-
of Electrical and Electronics Engineering, Boǧaziçi University, Bebek 34342 tion points. The method used in the bit-to-index mapping is
Istanbul, Turkey, E-mail: mutlu.koca@boun.edu.tr. immaterial to the central discussion, so throughout the paper
H. Sari is with Sequans Communications, Les Portes de la Défense, we assume uniform mapping. The transmitted signal vector is
15-55 Boulevard Charles de Gaulle, 92700 Colombes, France,  T
E-mail:hsari@ieee.org. x= x 1 , . . . xℓ , . . . xNt where all but one entry
1089-7798 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
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. Citation information: DOI 10.1109/LCOMM.2017.2733540, IEEE
Communications Letters 2

is zero because only one antenna is active for transmission. where N (u, û, v, v̂) denotes the number of bit errors between
In other words, the position of the non-zero element denotes the channel and symbol pairs, (hu , Xv ) and (hû , Xv̂ ), and
the active antenna index and its value indicates thetransmitted log2 (Nt M ) is the total number of bits. P̄s (u, û, θu , θû , v, v̂)
symbol. We assume unit transmit power, i.e., Ex x† x = 1 .

denotes the corresponding average pairwise error probability
The proposed precoding approach is based on rotating (APEP) that can be written as
the symbols according to the antenna selected for signal " r !#
transmission. This can be seen as multiplying the transmit- k z k2
P̄s (u, û, θu , θû , v, v̂) = EH Q (5)
ted symbol vector x with a diagonal phase  rotation matrix, 2
Θ = Diag ejθ1 ejθ2 ... ejθNt , which does not
change power budget and the single RF property of SM. The where the Nr × 1 vector z is defined as

corresponding received signal model is expressed as z = ρ hu Xv ejθ − hû Xv̂ ejθû ,

(6)
y = HΘx + ν (1) with ρ = 1
being the average SNR.
N0
Notice that under the signal model presented in the previous
where H is the Nr × Nt dimensional channel matrix and
section, z forms a zero-mean proper complex Gaussian random
y and ν are the Nr × 1 dimensional received signal and
vector with the covariance matrix Σz that can be computed as
channel noise vectors, respectively. The elements of ν are
modelled as independent and identically distributed (i.i.d.) Σz = ργ(u, û, v, v̂, θu , θû )Σr , (7)
complex Gaussian variables with zero mean and double side-
where ρ is as defined before, and
band variance N0 , i.e., νk ∼ CN (0, N0 ) for k = 1, . . . , Nr .
If the active antenna and transmitted symbol indices are γ(u, û, v, v̂, θu , θû ) =| Xv |2 + | Xv̂ |2 −2Re{σu,
t
û e
j(θu −θû )
Xv Xv̂∗ }
given as u and v, respectively, (1) can also be written as (8)
is the correlation contribution of the transmission parameters
y = hu Xv ejθu + ν (2) t
with σu,û denoting the (u, û)-th entry of Σt . Notice from (7)
that the transmit and receive correlation effects are separable
where hu is the u-th column of H for u = 1, . . . , Nt and Xv in the covariance matrix expression. Then following the steps
is the v-th element of X for v = 1, . . . , M . given in [14], the APEP in (5) can be computed by
We assume correlated Rayleigh fading in the MIMO chan- Z π2 −1
nel. To describe the spatial correlation effects, the well-known 1 Σz
Kronecker correlation model is used where H is expressed as P̄s (u, û, θu , θû , v, v̂) =
2 + I dθ. (9)
π 0 4 sin θ

1 1
T Notice that (9) can be computed in closed form. Suppose
H = Σr2 H̆Σt2 (3)
λk for k = 1, . . . , Nr denote the eigenvalues of the matrix
with Σt and Σr being the real valued, Hermitian symmetric Σr . In the fully correlated receiver case rank [Σr ] = Nr and
transmit and receive correlation matrices, respectively. H̆ is the all the eigenvalues of Σz are distinct. Then
independent Rayleigh fading channel matrix whose elements −1 Y Nr  −1 X Nr  −1
are described as i.i.d. complex Gaussian random variables, Σz ργλk ργλk
4 sin2 θ + I = +1 = ξk +1

i.e., [H̆]p,q = h̆p,q ∼ CN (0, 1) for p = 1, . . . , Nr and q =
k=1
4 sin2 θ k=1
4 sin2 θ
1, . . . , Nt . Throughout this letter, the exponential correlation  −1
model of [13] is employed where the elements of Σt and Σr Q
where ξk = j6=k 1 − λkj
λ
is the k-th residual in the
t |q−q̂|
are defined as [Σt ]q,q̂ = σq,q̂ = αt for q, q̂ = 1, . . . , Nt partial fraction expansion for k = 1, . . . , Nr . With this result,
r |p−p̂|
and [Σr ]p,p̂ = σp,p̂ = αr for p, p̂ = 1, . . . , Nr where (9) can be rewritten as
0 ≤ |αt |, |αr | ≤ 1. However, the central contributions of this Nr Z π2  −1
work will also be valid with other models. X 1 ργλk
P̄s (u, û, θu , θû , v, v̂) = ξk + 1 dθ
Given the signal model in (2), the most-likely antenna and
k=1
π 0 4 sin2 θ
symbol index pair (uML , vML ) is expressed as Nr
s !
1X ργλk
(uML , vML ) = arg min D(y|hu , Xv , Θ) = ξk 1 − . (10)
u,v 2 4 + ργλk
k=1
jθu 2 The second line of (10) is from ( [15], Appdx. B) stating that
where D(y|hu , Xv , Θ) =k y − hu Xv e k is the distance
metric used in the detection. Z π2  n
1 sin2 θ
Jn (c) = dθ
π 0 sin2 θ + c
III. P ERFORMANCE A NALYSIS AND O PTIMIZATION n−1
n
X n−1+k  k
The central task in the design of the precoding matrix = [P (c)] [1 − P (c)] (11)
k
Θ is to determine the optimum rotation angles θℓ for ℓ = k=0
1, 2, . . . , Nt . Because we consider time-varying fading chan-  q 
nels without any CSI at the transmitter, a channel-aware where P (c) = 21 1 − 1+c c
. If the correlation matrix has
optimization is not possible. Instead, as proposed here, the repeated eigenvalues, the partial fraction
 expansion in (10) −t
precoder design is based on the minimization of the asymptotic

has terms in the form of 4ργλ k
sin2 θ
+1 where t > 1 for
ABEP, which is not available in closed-form but can be tightly
approximated by the upper bound given by which the expansion coefficients can be found and the ABEP
can still be computed. The extreme case is when the receiver
Nt XNt X M X M correlation does not exist, which corresponds to t = Nr . In
1 X N (u, û, v, v̂)
P̄b ≤ P̄s (u, û, θu , θû , v, v̂) this case, the APEP can be expressed directly as
Nt M u=1 log2 (Nt M )
û=1 v=1 v̂=1
(4) P̄s( u, û, θu , θû , v, v̂) = JNr (ρ̄) (12)
1089-7798 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
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. Citation information: DOI 10.1109/LCOMM.2017.2733540, IEEE
Communications Letters 3

where ρ̄ = ργ(u, û, v, v̂, θu , θû )/4 and JNr (·) is as in (11). SNR regime can be written as
Once the APEP’s between all channel and symbol pairs are 2 2 2 2
computed, they can be used to obtain the tight ABEP upper B(Σr ) X X X X N (u, û, v, v̂)
P̄b ≤
bound in (4). Unfortunately, in many cases, it is not possible 8ρ u=1 1 − αt cos(π(v − v̂) + θ(u − û))
to obtain a closed-form solution for the optimum rotation  û=1 v=1 v̂=1 
B(Σr ) 1 1 2
angle set Θ̂ that minimizes this union bound. However, Θ̂ = + + . (14)
can easily be found by employing numerical search algorithms 2ρ 1 + αt 1 − αt cos θ 1 + αt cos θ
and several observations can be made regarding the proposed It is clear from (14) that when θ is 0 or π radians, the
ABER minimizing precoding approach as follows: ABEP upper bound is maximized. That is, conventional SM
First, when there is no spatial correlation among the transmit transmission with no rotation or that employing full rotation
t
antennas Σt = I or σu,û = δu,û , and Σz used in (9) becomes are the two worst strategies. On the other hand, √ a unique
minimum is achieved when αt cos θ = γ0 = 3 − 8 radians.
ρ | Xv |2 + | Xv̂ |2 −2Re{Xv Xv̂∗ } Σr , u = û
 
Σz = More importantly, so long as αt ≥ γ0 , this minimum is
ρ | Xv |2 + | Xv̂ |2 Σr , u 6= û. achievable by some rotation angle θ̂ = arccos αγ0t even when
the two transmit antennas are fully correlated ( αt = 1).
Notice that any dependence on the rotation angles disappears.
Notice that the 2 × Nr dimensional SM system employ-
That is why, rotation-based precoding provides benefits only
ing BPSK is a special case where approximate closed-form
if there is some spatial correlation among transmit antennas
solution for the optimum rotation angle is possible. Such
in the case of slow-fading, time-varying Rayleigh channels.
closed-form solutions are not available for higher dimensional
Second, notice also that the optimum rotation angle set Θ̂
MIMO systems or larger constellations but this does not pose
is independent of the receiver correlation effects. To evaluate
a significant problem as the evaluation of (4) is relatively easy
this, consider the case where the receiver is fully correlated.
which makes it possible to find the optimum rotation angle set
Using the
q first
 order Taylor expansion, the approximation
1

c 1
conveniently by numerical search methods.
2 1− 1+c ≈ 4c for c >> 1 can be obtained and used to
rewrite the APEP in (10) at large SNR values as IV. S IMULATION R ESULTS
1 In this section we present the effect of the optimum ro-
P̄s (u, û, v, v̂) ≈ B(Σr ) (13)
ργ(u, û, v, v̂, θu , θû ) tation angles used in the precoding on the simulated ABER
performance. For ease of illustration, we consider the 2 × 2
where B(Σr ) = N
P r ξk
k=1 λk is only dependent on the eigen- SM system where BPSK or QPSK symbols are sent from the
values of Σr . That is, the impact of the transmit and receive first antenna without any rotations (θ1 = 0) and only those
parameters on the APEP are separable. As a result, the ABEP transmitted from the second are rotated with θ (θ2 = θ).
expression that appears in (4) at high SNR values becomes In all configurations, Σr = I is assumed since the receive
correlation does not have any impact on the optimum rotation
Nt X
Nt XM XM
B(Σr ) X N (u, û, v, v̂) angle(s) and the theoretical ABEP upper bounds are evaluated
P̄b ≤ . at 24 dB SNR where the bound is in tight agreement with the
ρNt M log2 (Nt M ) u=1 v=1
γ(u, û, v, v̂, θu , θû )
û=1 v̂=1
actual ABER simulation results. The respective union bounds
It is clear that any optimization of the rotation angles involves for various values of the transmit correlation coefficient αt
only the terms inside the summations and not the terms out- are plotted as a function of θ and in each case the optimum θ̂
side. Therefore, the size and the structure of receive correlation value minimizing the ABER is determined.
matrix does not change the optimum set. Figures 1(a) and (b) show the θ vs. the ABEP upper bounds
Finally, notice that the number of receive antennas only for BPSK and QPSK, respectively, for a number of αt values.
affects the number of terms in B(Σr ) which does not have In all cases, θ = 0 corresponds to the conventional SM
an impact on the optimum rotation matrix Θ̂. Therefore, the where all symbols are transmitted with the same constellation
precoder optimization can be done assuming a single receive orientation regardless of the activated antenna. As seen in
antenna for simplicity even if multiple receive antennas are Fig. 1(a) for BPSK, when the transmit correlation is high, the
used. The optimum angle set remains the same. conventional SM performance deteriorates significantly, but it
Despite not being easy to obtain in closed-form for the is possible to align the constellations for a minimum ABER
general SM scenarios, the optimum rotation angles can be regardless of the severity of the correlation. For the QPSK
computed analytically for small scale MIMO systems and/or transmission shown in Fig. 1(b), the optimized rotation helps
constellations. Below we consider such a case that is useful in in avoiding significant performance losses even if convergence
depicting the potential benefits of the proposed phase rotation. to a single minimum as in BPSK is not possible.
Next we depict the impact of optimum precoding on the
ABER vs. SNR performance of SM. Figures 2(a) and (b) show
A. Example Case: Precoding for 2 × Nr SM with BPSK the respective performances of the 2 × 2 system employing
Consider a 2 × Nr SM system employing BPSK. Suppose BPSK and QPSK. As shown above, for BPSK, independently
the BPSK symbols transmitted from the first antenna are of the correlation between the two transmit antennas, the
sent without any rotation and a constellation rotation of θ ABER performance of the precoded SM system converges
is employed for the symbols transmitted from the second to that of the uncorrelated SM whereas conventional SM
antenna. Then γ(u, û, v, v̂, θu , θû ) in (8) can be expressed as performance suffers a significant loss especially for large cor-
relations. For QPSK, as in the case of BPSK, the performance
 |u−û| 
degradation of the conventional SM can be compensated
γ(u, û, v, v̂, θu , θû ) = 2 1 − αt cos π(v − v̂) + θ(u − û)
significantly by optimum precoding. Even for large transmit
where αt is the correlation coefficient between the two trans- correlations, the ABER’s can be brought within a few dB’s
mit antennas and the approximate ABEP in (14) in the high of the uncorrelated Rayleigh performance even if convergence
1089-7798 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
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. Citation information: DOI 10.1109/LCOMM.2017.2733540, IEEE
Communications Letters 4

−1 0
10 10
α =0.99
t
α =0.9
t
−1
α =0.8 10
t
αt=0.7
−2
10 α =0.5 −2
t 10
αt=0
α =0.99, Conventional SM
ABEP Upper Bound

t
10
−3
α =0.99, Rotation optimized SM
t

ABER
−3 α =0.9, Conventional SM
10 t
α =0.9, Rotation optimized SM
−4 t
10
α =0.8, Conventional SM
t
α =0.8, Rotation optimized SM
t

10
−5 αt=0.7, Conventional SM
−4
10 αt=0.7, Rotation optimized SM
αt=0.5, Conventional SM
−6
10 α =0.5, Rotation optimized SM
t
α =0, Conventional SM
t

−5 −7 Theoretical ABEP Bounds


10 10
0 0.5 1 1.5 2 2.5 3 0 5 10 15 20 25 30
θ (rad) SNR (dB)

(a) (a)

−1 0
10 10

α =0.99
t
α =0.9
t −1
10
α =0.8
t

10
−2
α =0.7
t
α =0.5
t −2
10
α =0
t
ABEP Upper Bound

ABER
−3 −3
10 10

αt=0.99, Conventional SM
α =0.99, Rotation optimized SM
−4 t
10
α =0.9, Conventional SM
t
10
−4
α =0.9, Rotation optimized SM
t
α =0.8, Conventional SM
−5 t
10
α =0.8, Rotation optimized SM
t
α =0, Conventional SM
t
Theoretical ABEP Bounds
−5 −6
10 10
0 0.5 1 1.5 0 5 10 15 20 25 30
θ (rad) SNR (dB)

(b) (b)

Figure 1: Optimization of ABER in 2 × 2 SM-MIMO system Figure 2: ABER comparison of precoded and conventional
employing a) BPSK, b) QPSK. 2 × 2 SM-MIMO systems employing a) BPSK, b) QPSK.

to a single minimum ABER is not possible. The proposed [5] E. Basar, U. Aygolu, E. Panayirci and H. V. Poor, ”New trellis code
precoding approach also provides similar benefits for higher design for spatial modulation,” IEEE Transactions on Wireless Commu-
nications, vol. 10, no. 8, pp. 2670-2680, Aug. 2011.
order MIMO systems or larger constellations but results for [6] C. Masouros, “Improving the diversity of spatial modulation in MISO
these cases are not presented here for space limitations. channels by phase alignment,” IEEE Communications Letters, vol. 18,
no. 5, pp. 729-732, May 2014.
[7] P. Yang, Y. Xiao, B. Zhang, M. El-Hajjar, S. Li and L. Hanzo,
V. C ONCLUSION “Phase rotation-based precoding for spatial modulation systems,” IET
We have presented a constellation rotation based precoding Communications, vol. 9, no. 10, pp. 1315-1323, 2015.
[8] P. Yang, Y. L. Guan, Y. Xiao, M. D. Renzo, S. Li and L. Hanzo, “Trans-
approach for SM-MIMO over correlated Rayleigh channels. mit precoded spatial modulation: Maximizing the minimum Euclidean
Both analysis and simulation results indicate significant im- distance versus minimizing the bit error ratio,” IEEE Transactions on
provements over conventional SM with very small increase Wireless Communications vol. 15, no. 3, pp. 2054-2068, Mar. 2016.
[9] C. Masouros and L. Hanzo, “Constellation randomization achieves
in the computational complexity at the transmitter. Notice transmit diversity for Single-RF spatial modulation,” IEEE Transactions
that a more general precoding approach where amplitude (or on Vehicular Technology, vol. 65, no. 10, pp. 8101-8111, Oct. 2016.
power) scaling in addition to phase rotations as given in [6], [10] M. C. Lee, W. H. Chung and T. S. Lee, “Generalized precoder design
formulation and iterative algorithm for spatial modulation in MIMO
[8], [9] can also be considered for further improvements in systems with CSIT,” IEEE Transactions on Communications, vol. 63,
the performance. Future work may also involve the extension no. 4, pp. 1230-1244, April 2015.
of the proposed approach to other fading channel types such [11] E. G. Larsson, ”Improving the frame-error-rate of spatial multiplexing
in block fading by randomly rotating the signal constellation,” IEEE
as Rician, doubly-selective and multipath fading channels. Communications Letters, vol. 8, no. 8, pp. 514-516, Aug. 2004.
[12] J. Li, M. Wen, X. Cheng, Y. Yan, S. Song and M. H. Lee, “Generalized
R EFERENCES precoding-aided quadrature spatial modulation,” IEEE Transactions on
Vehicular Technology, vol. 66, no. 2, pp. 1881-1886, Feb. 2017.
[1] R. Mesleh, H. Haas, S. Sinanovic, C. W. Ahn and S. Yun, “Spatial [13] S. L. Loyka, “Channel capacity of MIMO architecture using the expo-
modulation,” IEEE Transactions on Vehicular Technology, vol. 57, no. nential correlation matrix,” IEEE Communications Letters, vol. 5, pp.
4, pp. 2228-2241, July 2008. 369-371, Sept. 2001.
[2] R. Mesleh, M. Di Renzo, H. Haas and P. Grant, “Trellis coded spatial [14] M. Koca and H. Sari, “Performance analysis of spatial modulation over
modulation,” IEEE Transactions on Wireless Communications, vol. 9, correlated fading channels,” in Proc. 2012 IEEE VTC-Fall, Quebec City,
no. 7, pp. 2349–2361, July 2010. QC, Canada, 2012, pp. 1-5.
[3] M. Koca and H. Sari, ”Bit-interleaved coded spatial modulation,” in [15] M.-S. Alouini and A. Goldsmith, “A unified approach for calculating er-
Proc. IEEE PIMRC 2012, Sydney, NSW, 2012, pp. 1949-1954. ror rates of linearly modulated signals over generalized fading channels,”
[4] Z. Yang, C. Liang, X. Xu and X. Ma, ”Block Markov superposition IEEE Trans. Commun., vol. 47, no. 9, pp. 1324-34, Sep. 1999.
transmission with spatial modulation,” IEEE Wireless Communications
Letters, vol. 3, no. 6, pp. 565-568, Dec. 2014.
1089-7798 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Вам также может понравиться