Professional Documents
Culture Documents
signal vector and v ∈ CNr ×1 is the additive white Gaussian Substitute (3),(4),(6) and (7) into (5), and the mutual
noise with zero mean and covariance matrix σ 2 INr . information between x and y can be expressed as
Under the assumption of SM MIMO system, the channel I(x; y|H = H)
matrix can be stated as H = [h1 , h2 , · · · , hNt ] with hi ∈
1 t
MN M Nt
CNr ×1 , i = 1, · · · , Nt . Since SM is employed, the baseband −dm1 ,m2
= log2 (M Nt ) − Ev log2 e ,
input signal could be further denoted as M Nt m =1
1 m2 =1
x = [0, 0, · · · , xi , · · · , 0]T , (2) (8)
where xi , i = 1, · · · , Nt , represents the single nonzero element where dm1 ,m2 = σ −2
||H(xm1 − xm2 ) + v||2 − ||v|| . 2
x ∈ C and is launched from the ith transmit antenna with
power Ex {xx† } = P . In the limit of very low signal to noise ratio regime (SNR,
γ = σP2 ), i.e. γ → 0, by assuming a fixed noise power σ 2 = 1,
A. Mutual Information with Finite Alphabet we have
Let we assume that (p + q)-bits information is transmitted lim I(x; y|H = H)
γ→0
at a time slot, p-bits for antennas selection and q-bits for
= lim I(x; y|H = H)
discrete constellations mapping such as M -ary pulse-amplitude P → 0
σ2 = 1
modulation (PAM), phase shift keying (PSK) modulation, or
1 t
quadrature amplitude modulation (QAM) of cardinality M i.e., MN M Nt
−dm1 ,m2
Nt = 2p , p ∈ N and M = 2q , q ∈ N. The finite alphabet with = log2 (M Nt ) − lim Ev log2 e ,
P → 0 M Nt m =1
cardinality M is denoted as SM = {S1 , S2 , · · · , SM }. Assume σ2 = 1 1 m2 =1
that the information bits to be transmitted are randomly gen- (9)
erated with P(ρ = 0) = P(ρ = 1) = 12 , ∀ρ, the probability
that the symbol Sk is chosen to be sent out is P(x = Sk ) = Obviously, to get the limit of (9), the exponential terms in
1 the second term of (9) should be analyzed first. When m1 =
M , k = 1, 2, · · · , M and the probability that the ith antenna
is selected to transmit is P(i = l) = N1t , l = 1, 2, · · · , Nt . m2 , considering P → 0 and σ 2 = 1, it can be easily got
||H(xm1 − xm2 ) + v||2 → ||v||2 , resulting that dm1 ,m2 → 0. If
For a given channel realization, the probability density m1 = m2 , we can get ||H(xm1 − xm2 ) + v||2 = ||v||2 , hence
function (PDF) of the corresponding received signal y when dm1 ,m2 = 0. Therefore, (9) can be written as
sending x writes
lim I(x; y|H = H) = 0 (10)
1 ||y − Hx||2 γ→0
P(y|x, H = H) = exp(− ). (3)
(πσ 2 )Nr σ2
In the limit of very high SNR regime, similarly, the limit
Define new vectors, xm ∈ CNt ×1 , m = 1, 2, · · · , M Nt and of the mutual information can be written as
let xm = x(k−1)Nt +l = [0, 0, · · · , Sk , 0, · · · ]T represent
lim I(x; y|H = H)
lth position γ→+∞
sending Sk at the lth position of xm . Therefore, the probability = lim I(x; y|H = H)
of sending xm is P(x = xm ) = P(x = Sk )P(i = l) = M1Nt . P → +∞
σ2 = 1
Then, the PDF of the corresponding received signal y can be = lim I(x; y|H = H) (11)
written as m1 = m2 , ||H(xm − xm ) + v||2 → ∞
1 2
m1 = m2 , ||H(xm − xm ) + v||2 = ||v||2
P(y|H = H) = Ex {p(y|x, H = H)} 1
σ2 = 1
2
1 t
MN
(4) = log2 M Nt .
= P(y|xm , H = H).
M Nt m=1
In general, the mutual information in (8) has no closed form
The mutual information between the input and output expression. Although the value can be got through numerical
signals can be given by [12] simulations, it is difficult to maximize the mutual information
based on the present expression, which we always expect to.
I(x; y|H = H) = H(y|H = H) − H(y|x, H = H) Thus in the following part, we derive a lower bound of the
(5)
= H(y|H = H) − H(v), mutual information, and based on the obtained lower bound
we set an approximated expression for the mutual information
where the entropy of the received vector y can be expressed with finite signal alphabets.
as
H(y|H = H) = Ey {− log2 (P(y|H = H))}
B. Lower bound and Approximation
=− P(y|H = H) log2 (P(y|H = H))dy,
CNr Jensen’s Inequality (JI) [13] is well known in solving lower
(6) bound of expression in (6). Employing JI in (6), we have
and the entropy of noise is
Ey {− log2 (P(y|H = H))} ≥ − log2 (Ey {P(y|H = H)}).
H(v) = Nr log2 (πeσ 2 ). (7) (12)
488
Proceedings of the 2014 IEEE ICCS
⎧ M N 2 ⎫
⎨ t 1 ||y−Hxm ||2
⎬
−
H(y|H = H) ≥ − log2 e σ2 dy
⎩ CNr (πσ 2 )Nr M Nt ⎭
m=1
⎛ ⎞
(14)
⎜ 1
M
Nt M Nt ||y−Hxm ||2 ||y−Hxm ||2 ⎟
= − log2 ⎜ dy⎟
− 1 − 2
e σ2 e σ2
⎝ (πσ 2 )2Nr (M Nt )2 Nr
⎠.
m1 =1 m2 =1 C
Θ
Therefore, lower bound of entropy of the received vector y at Apparently, the difference ΔI(x;y|H=H) between
given channel realization can be written as I(x; y|H = H) and I(x; y|H = H) for very low or
very high SNR is given by
H(y|H = H) ≥ − log2 P(y|H = H)P(y|H = H)dy e
N ΔI(x;y|H=H) = Nr log2 ( ) (21)
C r 2
2
= − log2 P(y|H = H) dy .
CNr Owing to this, we propose an approximation I(x; y|H =
(13) H) to the mutual information I(x; y|H = H) and have
Substituting (4) into (13) leads to a loose bound of the entropy I(x; y|H = H) ≈ I(x; y|H = H)
of y as shown in (14), where the integral Θ can be calculated (22)
y|H = H) + ΔI(x;y|H=H)
= I(x;
as [14]
Further substituting the (18) into (22), the approximation
I(x; y|H = H) can be rewritten in a closed form expression
π Nr †
as
1
H † H(xm1 −xm2 )
Θ= e− 2σ2 (xm1 −xm2 ) . (15) − 2σ12 Δ†x H † HΔx
2 Δx ∈D e
I(x; y|H = H) = − log2 . (23)
Further apply (15) in (14) and define Δx = xm1 − xm2 , we (σ 2 )Nr (M Nt )2
achieve the loose lower bound
− 2σ12 Δ†x H † HΔx
C. Constellation Design Criterion for SM MIMO Systems
Δx ∈D e
H(y|H = H) ≥ − log2 , For traditional MIMO systems with finite-size inputs, it
(2π)Nr (σ 2 )2Nr (M Nt )2 (16) has been proved that the mutual information is a function
of the distance between different constellation points [15]. In
H(y|H=H) SM MIMO, where the input distribution is constrained to be
where the set of all difference vectors D is given by of finite-size and to belong to the special subset of spatial
modulation (i.e., the inputs vector has, at most, one nonzero
D = {Δx |Δx = xm1 − xm2 , ∀m1 , m2 = 1, · · · , M Nt }. (17) entry), the relationship between the mutual information and
the finite alphabet set SM is described in (8), where we can
Within (16), a rather loose lower bound H(y|H = H) has been see that the difference vectors Δx = xm1 − xm2 , ∀m1 , m2
y|H = H) of
formulated, which leads to a lower bound I(x; are the key factors that dominate the mutual information
the mutual information I(x; y|H = H). And it can be easily get that those difference
vectors are determined by the finite alphabet set SM . As a
y|H = H) = H(y|H
I(x; = H) − H(v) result, how to design the finite alphabet set SM to maximize
− 2σ12 Δ†x H † HΔx
the mutual information becomes a problem for practical SM
Δx ∈D e
(18) MIMO systems. Unfortunately, we should say that it is hard
= − log2 .
( 2e )Nr (σ 2 )Nr (M Nt )2 to analyze the impact of SM on the mutual information
theoretically because of the non-closed form of (8). Hereafter,
Given a fixed noise power σ 2 = 1, in the limit of very small we will analyze the constellation design based on the obtained
y|H = H)
and very large transmit power P , the limits of I(x; approximated mutual information I(x; y|H = H) instead of
can be easily got. Since it is obvious that all exponential terms I(x; y|H = H).
in numerator of (18) tend to be one for P → 0 and zero for Since the PDF of Δx , P(Δx ) = (M N 1
2 , ∀Δx ∈ D,
t)
P → ∞ and m1 = m2 , the limits can be written as
I(x; y|H = H) in (23) can be rewritten as
! "
lim y|H = H) = Nr log2 ( 2 )
I(x;
1 † †
I(x; y|H = H) = 2Nr log2 σ − log2 (EΔx e− 2σ2 Δx H HΔx )
P → 0 e (19)
σ2 = 1 ! 1 2 2
"
≤ 2Nr log2 σ − log2 (EΔx e− 2σ2 ||H|| ||Δx || )
lim y|H = H) = log2 (M Nt ) + Nr log2 ( 2 ).
I(x; ≤ 2Nr log2 σ − log2 (e− 2σ2 ||H||
1 2
EΔx {||Δx ||2 }
)
P → ∞ e (20)
σ2 = 1 (24)
489
Proceedings of the 2014 IEEE ICCS
Mutual Information(bit/s/Hz)
1 † † y|H), Nt = 4, Nr = 2
I(x;
SM (H)|A−CDC = arg min EΔx e− 2σ2 Δx H HΔx . 5
Ex {xx† }=P,x∈SM I(x; y|H), Nt = 8, Nr = 4
(25) 4
y|H), Nt = 8, Nr = 4
I(x;
Since the constellation design criterion (CDC) in (25) is a func-
tion of channel realizations, the design criterion can be applied 3
for any channel conditions only if the channel is known. We
2
therefore name it adaptive CDC (A-CDC). Considering it is
computational complex to calculate the constellation according 1
to (25), especially when the channel is fast fading, we then
propose another criterion, named maximizing the average 0
−20 −15 −10 −5 0 5 10 15 20
pairwise distance constellation design criterion (MAPD-CDC), SNR(dB)
designed according to the third line of (24) to maximize the
upper bound of I(x; y|H = H) instead of maximizing the Fig. 2: Mutual information with QPSK symbols as inputs of
I(x; y|H = H) itself. Then the criterion can be expressed as 4 × 4, 4 × 2 and 8 × 4 SM MIMO systems
# $
SM |M AP D−CDC = arg max EΔx ||Δx ||2 .
Ex {xx† }=P,x∈SM
(26) 6
In this way, the constellation design is independent of the
channel conditions. With this criterion, the computational 5
complexity can be greatly reduced, but in the same time the
Nt=16,Nr=16
system performance in terms of the mutual information will be
Mutual Information(bit/s/Hz)
4
degraded as well. Future, based on the two optimization criteri-
ons, the constellation design problem for SM MIMO systems
3
can be solved through a heuristic search algorithm, such as Nt=4,Nr=4
Genetic algorithm, particle swarm optimization algorithm, etc.
2
Nt=2,Nr=2
It is also shown by Fig. 2 that the mutual information is mono- To show the mutual information of SM MIMO systems
tonically increasing until it hits the upper bound log2 (M Nt ). when employing different constellation design criteria, by set
Comparing 4 × 4 SM MIMO system to 4 × 2 SM MIMO M = 4 we first generate 1000 SM s finite alphabet sets
system, we can see that as the number of receive antennas randomly guaranteeing Ex {xx† } = P, x ∈ SM , the total set
(1) (2) (1000)
increases, the maximum mutual information does not change, is denoted as Stotal = {SM , SM , · · · , SM }. Based on
but it saves 8dB transmit power to achieve the maximum the two proposed constellation design criteria, A-CDC and
mutual information. Further compare 8 × 4 SM MIMO system MAPD-CDC, the proper finite alphabet is chosen for data
with 4 × 4 SM MIMO system, we can see that 8 × 4 SM transmission. The numerical results are shown in Fig.3. For
MIMO system outperforms 4×4 SM MIMO system in term of comparison purpose, the mutual information when employing
maximum mutual information by 1 (bit/s/Hz). This is because random finite alphabet set selection is also included in Fig.3.
8 transmit antennas offers more antenna selection gain than 4 It is shown that the finite alphabet selection using A-CDC
490
Proceedings of the 2014 IEEE ICCS
and MAPD-CDC outperforms the Random Selection (RS) in [9] A. Lozano, A. M. Tulino, and S. Verdú, ”Optimum power allocation
terms of mutual information. And the system with A-CDC for parallel Gaussian channels with arbitrary input distributions”, IEEE
Trans. Inf. Theory, vol. 52, pp. 3033-3051, July 2006.
offers larger mutual information than that with MAPD-CDC
as analyzed, but the performance loss is trivial compared the [10] C. Xiao and Y. R. Zheng, ”On the mutual information and power
allocation for vector Gaussian channels with finite discrete inputs”, in
reduced computational complexity, at least for small number Proc. IEEE Globecom. 08, New Orleans, LA, Nov. 30 Dec. 4 2008.
of transmit antennas. [11] C.Xiao, Y.R. Zheng and Z. Ding, ”Globally Optimal Linear Precoders
for Finite Alphabet Signals Over Complex Vector Gaussian Channels”,
IV. C ONCLUSION IEEE Trans. Singal Processing. vol.59, no.7, July, 2011.
[12] I. E. Telatar, ”Capacity of multi-antenna gaussian channels”, Europ.
In this paper, we have proposed an SM MIMO systems Trans. Telecommu., vol. 10, pp.585 -595 1999
oriented information-theoretic framework and derived an ex- [13] T. M. Cover and J. A. Thomas, ”Elements of Information Theory”,
pression of the mutual information based on the proposed 1991,Wiley
frame work. Since the expression of mutual information can [14] C. Bockelmann, D. Wübben, and K.-D. Kammeyer, ”Mutual informa-
is not in a closed form, we further proposed a closed form tion based rate adaptation for MIMO-OFDM systems with finite signal
approximation for the mutual information with finite input alphabets,” in International ITG Workshop on Smart Antennas (WSA),
Darmstadt, Germany, Feb. 2008.
alphabets. Numerical results show that the approximated mu-
[15] F. Pérez-Cruz, M. Rodrigues, and S. Verdú, ”MIMO Gaussian channels
tual information can approach the real value well especially with arbitrary inputs: Optimal precoding and power allocation,” IEEE
in very high and low SNR regimes. Moreover, based the Trans. Inf. Theory, vol. 56, no. 3, pp. 1070-1084, Mar. 2010.
obtained approximated expression of mutual information, two
constellation design criteria, A-CDC and MAPD-CDC, have
been proposed to maximize the system mutual information
with finite alphabet. Numerical results show that both of the
two criteria perform well, especially MAPD-CDC is of lower
complexity while keeps low performance degradation. We also
simulated the mutual information of 4 × 4, 4 × 2 and 8 × 4
SM MIMO systems over Rayleigh channels, results show that
although increasing the number of receive antennas does not
improve the maximum mutual information, it saves transmit
power in achieving the same mutual information.
ACKNOWLEDGMENT
The work presented in this paper was supported in part
by the National Natural Science Foundation of China (No.
61371109, 61271229), the Outstanding Youth Fund of Shan-
dong province with No. JQ201315 and the New Century
Excellent Talents from the Ministry of Education of China
(NCET-11-0316).
R EFERENCES
[1] J. Jeganathan, A. Ghrayeb and L. Szczecinski and A. Ceron, ”Space Shift
Keying Modulation for MIMO Channels”, IEEE Trans. Wireless Comm.,
vol. 8, no. 7, pp.3692-3703, July 2009
[2] A. Younis, N. Serafimovski, R. Mesleh, and H. Haas, ”Generalised spatial
modulation”, in Proc. IEEE Asilomar Conf. Signals Syst. Comput., Nov.
2010, pp. 1498-1502.
[3] N. Valliappan, A. Lozano, and R. W. Heath, Jr., ”Antenna subset
modulation for secure millimeter-wave wireless communication”, IEEE
Trans. Commun., vol. 61, no. 8, pp. 3231-3245, Aug. 2013.
[4] Y. Yang and B. Jiao, ”Information-guided channel-hopping for high data
rate wireless communication”, IEEE Commun. Lett., vol. 12, no. 4, pp.
225-227, Apr. 2008.
[5] M. D. Renzo, H. Haas, A. Ghrayeb, S. Sugiura, and L. Hanzo, ”Spatial
modulation for generalized MIMO: challenges, opportunities, and imple-
mentation”, Proceedings of the IEEE, vol. 102, no. 1, pp.56-103, Jan.
2014
[6] M. Di Renzo and H. Haas, ”Performance analysis of spatial modulation”,
in Proc. ICST/IEEE Int. Conf. Commun. Networking China, Aug. 2010,
pp. 1-7.
[7] J. Jeganathan, A. Ghrayeb, and L. Szczecinski, ”Spatial modulation:
optimal detection and performance analysis”, IEEE Commun. Lett., vol.
12, no. 8, pp. 545-547, Aug. 2008.
[8] X. Guan, Y. Cai, and W. Yang, ”On the Mutual Information and
Precoding for Spatial Modulation with Finite Alphabet”, IEEE Comm.
Letters, vol. 2, no. 4, pp. 383-386, Aug. 2013.
491