Professional Documents
Culture Documents
Abstract-In this paper, the spatial modulation (SM) is tion (IAS) and multiple radio frequency (RF) chains make
considered as a novel physical layer technique to the practical implementation of these schemes difficult.
improve the physical layer security of wireless networks.
Exploiting the index of antenna as an additional Therefore, a clear research goal is to develop new
dimension to convey part of source information bits, SM approaches to achieve physical layer security.
helps to achieve multiplexing gain and realize higher On the other hand, spatial modulation (SM) has recently
data rates. Moreover, by properly designing the received much attention since it can achieve a spatial
transmission signal, SM can also improve physical layer multiplexing gain while mitigating the aforementioned
security. The secrecy mutual information of SM with
finite alphabet input is derived in this work, and a limitations [8]. By mapping information bits into the
precoding scheme is proposed to improve the physical antenna indices, a higher rate is realized with lower
layer security performance of SM. Analysis and hardware complexity. At the receiver, the message
numerical results verify the validity and effectiveness of conveyed on the indices can be recovered by detecting
the proposed scheme, and it also shows that the which antenna transmits the signal. The optimal receiver
maximum achievable secrecy mutual information is the
logarithm of the number of transmit antennas. design and error performance analysis are presented in
[9-10], while the capacity of SM with Gaussian input over
Index Terms-physical layer security, spatial Rayleigh fading channels has been derived in [11].
modulation, finite alphabet, secrecy mutual information After careful analysis, it can be found that both physical
layer security and spatial modulation rely on the
I. INTRODUCTION
randomness and uniqueness properties of wireless channels.
MN m 1 n 1 ³v SV2 ¨ V2 ¸ m2 1 ¨¨
using the SM mapping table. log2 M ¸dv
© ¹ V2 ¸ (7)
© ¹
TABLE I. SM MAPPING TABLE WITH M 4 AND BPSK
ª 2 ·º
§
2
1 M N « ¨ hm hm2 xn v v ¸»
M
=log2 M ¦¦ v 2 ¦
E log
MN m 1 n 1 « m2 1 ¨¨
exp
V2 ¸»
Input Antenna Transmit Input Antenna Transmit «¬ ¸»
© ¹¼
bits index symbol bits index symbol
With similar algebraic manipulations, the mutual
00 0 1 +1 10 0 3 +1 information between y and x is derived as
00 1 1 -1 10 1 3 -1 ª º
« M N § d v2 · »
01 0 2 +1 11 0 4 +1 M N
«
«
¦ ¦ exp ¨ D 2 ¸
¨ V ¸
»
»
(8)
1 m2 1 n 2 1
© ¹
I x; y log 2 N ¦ ¦ E v « log 2 »
MN m 1 n 1 « §
·»
2
01 1 2 -1 11 1 4 -1 M h h x v
¦ exp ¨¨ ¸»
m m2 n
«
« V2 ¸»
m2 1 ¨ ¸
¬« © ¹ ¼»
2
in which d D hm xn hm 2 xn 2 . Based on (7) and (8), the V 2 hm xn hm xn / 2 . Accordingly, the average pairwise
2 2
mutual information over the main channel is written as error probability (APEP) for the decoding at the destination
I x, a; y D I a; y x I x; y is obtained as
(9)
1 M N ª M N § d v 2 v 2 ·º 1 M N M N § d ·
log2 MN ¦¦ Ev «log2 ¦ ¦ exp ¨ D 2
MN m 1 n 1 « ¨ V
¸»
¸» PD ¦¦
MN MN 1 m 1 n 1
¦¦ Q¨ D ¸ (14)
¬ m2 1 n2 1
© ¹¼ m2 1 n2 1 © 2V ¹
( m2 , n2 ) z ( m, n )
Observing that d D 0 when m2 m and n2 n , (9)
can be rewritten as Similarly, the APEP at the eavesdropper is
ª §
f (V )
·º
2
1 M N M N § d ·
« ¨
§ 2 ¸»
dD v v ·¸» (10)
2 PE ¦¦
MN MN 1 m 1 n 1
¦¦ Q¨ E ¸ (15)
© 2V ¹
M N M N
1 «log ¨1
I x, a; y D log2 MN ¦¦ v 2 ¦¦ ¨
E
MN m 1 n 1 « ¨ m2 1 n2 1
exp ¨
V2
¸
¸¸»
m2 1 n2 1
( m2 , n2 ) z ( m, n )
« ¨ (m2 ,n2 )z(m,n) © ¹¸»
«¬ © ¨ ¸»
¹¼
C. Precoding
where (m2 , n2 ) z (m, n) denotes that the events m2 m
and n2 n do not occur simultaneously. Denote the As we mentioned, selecting an antenna is equivalent to
signal-to-noise ratio (SNR) by SNR 1/ V2 of, then we have selecting a channel. Hence, the process of SM, i.e. selecting
f (V 2 ) o 0 , thus the upper bound I x, a; y D an antenna to transmit a chosen symbol according to the
UP
log2 MN is
obtained. source information block, can be viewed as a mapping I
On the other hand, the mutual information over the given by
eavesdropper’s channel is expressed as
I : Au X RD where RD ^h m nx m, n` (16)
1 M N ª § d u u ·º
2 2
M N
(10)
I x, a; yE log2 MN ¦¦Eu «log2 ¦¦exp¨ E V2
MN m 1 n 1 « m2 1n2 1 ¨
¸»
¸»
in which RD can be viewed as the received signal space at
¬ © ¹¼ the destination for non-noise channel. There are totally
MN signal points in RD , and d D is the mutual
in which d E g m xn g m 2 xn 2 and u is the noise at the
Euclidean distance between any two of them. For the
eavesdropper. As a result, the secrecy mutual information is
eavesdropper’s channel, there is a similar mapping M
obtained as
given by
ISEC I x, a; y D I x, a; y E
ª
M : Au X RE where RE ^g m nx m, n` (17)
M N § dE v 2 · º
« ¦¦ ¨ V 2 ¸¸ »»
exp ¨
(11) An effective approach to improve the secrecy mutual
1 M N «
¦¦Ev «log2 M N §© ¹
m2 1 n2 1
= » information is to degrade the detection performance at the
MN m 1 n 1 « dD v · »
2
eavesdropper, so we pursue a precoding scheme to reduce
« ¦¦exp¨ V 2 ¸ »
¨ ¸
d E , based on the observation from (15).
«¬ m2 1 n2 1
© ¹ »¼
We assume that the CSI of the eavesdropper’s channel is
B. Pairwise Error Probability available at the source. Therefore, the aim is to develop a
method to adjust the MN signal points in space RE , so
The task of decoding at the destination is to detect not that the Euclidean distance d E and the mutual
only the traditional symbol x but also the index of information I x, a; y E decreases. Recalling what is
transmit antenna a , i.e. to decide which channel is used. mentioned in section I, if the channels are all the same, the
The maximum likelihood (ML) decoding algorithm can be messages carried on the indices of antennas would be
expressed as following undistinguishable. Inspired by this, we multiply the
ml , xˆ
n arg min
m A , xn X
y hm xn
2
(12)
symbols transmitted by the mth antenna by a complex
precoding coefficient D m to rotate and amplify the signal,
Denoting rm , n hm xn , thus we have the pairwise error which is determined by
probability (PEP) P(rm , n o rm2 , n2 ) , i.e. the probability of g1
Dm (18)
detecting the m2 th antenna transmits xn 2 when instead, gm
the m1th antenna transmits xn1 as following With precoding, we have Dm gm xn g1xn for all m , i.e. no
matter which antenna transmits xn , the received signal
2 2
P rm,n o rm2 ,n2 Pr y hm xn ! y hm2 xn2 point in RE would always be g1 xn . Consequently, it
holds that Dm gm xn Dm2 gm2 xn 2 0 if we have n n2 , thus
ª hm xn hm2 xn2 º
2
a
the mutual information over the eavesdropper’s channel can
= Pr «Re v hm xn hm2 xn2 » (13)
*
!
« 2 » be rewritten as
¬ ¼
b § d · ª § § d 'u 2 u 2 ··º
1 M N « ¨ M N
¸¸» (19)
=Q¨ D ¸ I x, a; y E log2 MN ¦¦
MN m 1 n 1
Eu log2 M ¦¦exp¨ E 2
« ¨ ¨ V ¸¸¸»
© 2V ¹ «¬ ¨© © ¹¹»¼
m2 1 n2 1
n2zn
where (a) is derived by substituting y hm xn v and where dE ' Dm gm xn Dm2 gm2 xn2 . Naturally, the eavesdropper
arranging the terms, and (b) is derived by using the fact that can not decode the message mapped on the indices.
*
Re v hm xn hm2 xn2 is still a RV following the complex However, this would not happen to the destination because
Gaussian distribution with zero mean and variance of of the difference between the main channel and the
eavesdropper’s channel. With (19), the maximum value of 3
ISEC I x, a; y D I x, a; yE
UP UP M=4
log2 M (21)
1
It implies that the maximum secrecy mutual information is
the logarithm of the number of transmit antennas.
0.5
M=2
IV. NUMERICAL RESULTS
0
In this section, the performance of the proposed scheme, -20 -15 -10 -5 0 5 10 15 20 25 30
SNR(dB)
which exploits SM to improve physical layer security, is
investigated numerically. The parameter is set as O 1 and Fig. 2. Secrecy mutual information VS SNR for M=2, 4, 8 and N=1, 2, 4
V 2 1 . We perform 10000 independent trials of Monte
Carlo experiments to obtain the average results. 0
10
Fig. 2 illustrates the secrecy mutual information for the
proposed SM based scheme with precoding. As expected,
the maximum secrecy mutual information is the logarithm
of the number of transmit antennas, e.g. it achieves 3
-1 M=2, N=1
bits/s/Hz at 30 dB for M 8 . In the special case N 1 , 10
M=2, N=2
all of the source information bits are conveyed on the M=4, N=2
APEP
indices of antennas, and it outperforms other cases with the M=8, N=2