Content uploaded by Tong-Xing Zheng
Author content
All content in this area was uploaded by Tong-Xing Zheng on Nov 29, 2017
Content may be subject to copyright.
0090-6778 (c) 2016 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/TCOMM.2017.2672661, IEEE
Transactions on Communications
1
Secure Transmissions in Millimeter Wave Systems
Ying Ju, Hui-Ming Wang, Senior Member, IEEE, Tong-Xing Zheng, Member, IEEE, and Qinye Yin
Abstract—Exploiting millimeter wave is an effective way to
meet the data traffic demand in the 5G wireless communication
system. In this paper, we study secure transmissions under slow
fading channels with multipath propagation in millimeter wave
systems. Concerning the new propagation characteristics of mil-
limeter wave, we investigate three transmission schemes, namely
maximum ratio transmitting (MRT) beamforming, artificial noise
(AN) beamforming and partial MRT (PMRT) beamforming.
We evaluate the secrecy performance by analyzing both the
secrecy outage probability (SOP) and the secrecy throughput
for each scheme. Particularly, for the AN scheme, we derive a
closed-form expression for the optimal power allocation ratio of
the information signal power to the total transmit power that
minimizes the SOP, as well as obtain an explicit solution on
the optimal transmission parameters that maximize the secrecy
throughput. By comparing the secrecy performances achieved by
different strategies, we demonstrate that the secrecy performance
of the millimeter wave system is significantly influenced by the
relationship between the legitimate user’s and the eavesdropper’s
spatially resolvable paths, which is different from the wireless
systems with statistically independent channel models. In the
absence of the common path between the legitimate user and
the eavesdropper, MRT beamforming is the best scheme. In
the presence of common paths, AN beamforming and PMRT
beamforming show their respective superiorities depending on
the transmit power and the number of common paths. Numerical
results are provided to verify our theoretical analysis.
Index Terms—Physical layer security, millimeter wave, multi-
path, secrecy outage, secrecy throughput, artificial noise.
I. INT ROD UC TI ON
WITH the rapid and continuous development of the
wireless communication technology, enormous amount
of confidential messages will be transmitted through wireless
systems. Nevertheless, the radio characteristics of electromag-
netic wave and the open feature of the wireless channel make
the design of secure communication challenging. Physical
layer security, exploiting the wireless channel characteristics to
maximize the difference between the legitimate user’s channel
This work was partially supported by the National High-Tech Research and
Development Program of China under Grant 2015AA01A708, the National
Natural Science Foundation of China under Grants 61671364 and 61172093,
the Author of National Excellent Doctoral Dissertation of China under
Grant 201340, the New Century Excellent Talents Support Fund of China
under Grant NCET-13-0458, the Young Talent Support Fund of Science
and Technology of Shaanxi Province under Grant 2015KJXX-01, and the
Specialized Research Fund for the Doctoral Program of Higher Education
of China under Grant 20130201130003. The associate editor coordinating
the review of this paper and approving it for publication was C.-X. Wang.
(Corresponding author: Hui-Ming Wang.)
Y. Ju is with the MOE Key Laboratory for Intelligent Networks and
Network Security, School of Electronic and Information Engineering, Xi’an
Jiaotong University, Xi’an 710049, China, and also with Shaanxi Monitor-
ing Station, State Radio Monitoring Center, Xi’an 710200, China (e-mail:
juyingtju@163.com).
H.-M. Wang, T.-X. Zheng and Q. Yin are with the MOE Key Laboratory for
Intelligent Networks and Network Security, School of Electronic and Infor-
mation Engineering, Xi’an Jiaotong University, Xi’an 710049, China (e-mail:
xjbswhm@gmail.com; txzheng@stu.xjtu.edu.cn; qyyin@mail.xjtu.edu.cn).
and the eavesdropper’s channel, provides a powerful tool to
improve the secrecy performance of wireless communication
systems [1], [2]. In recent years, multiple-antenna technique
is considered as an efficient way to enhance the physical
layer security [3]-[7]. By adopting multiple-antenna technique,
various studies have deduced and calculated secrecy capacity
from the perspective of information theory. In these early
studies, a significant assumption is that the eavesdropper’s
channel state information (CSI) is accessible to the transmitter.
When the eavesdropper’s instantaneous CSI is not available at
the transmitter in practice, artificial noise (AN) beamforming
has been proposed in [8] where the transmitter emits the infor-
mation bearing signal together with AN to interfere with the
eavesdropper. The power allocation between the information
signal and the AN is studied both under the fast fading channel
[9], [10] and the slow fading channel [11], [12]. Under the
slow fading channel, transmission schemes are designed to
reduce secrecy outage probability (SOP) and increase secrecy
throughput. In [11], the analytical power allocation that mini-
mizes the SOP is obtained in the scenario where the legitimate
user’s channel is available. In [12], the code rate and power
allocation are optimized to maximize the secrecy throughput in
both the non-adaptive and adaptive schemes under a given SOP
constraint. We should point out that, the design and analysis
of physical layer security schemes rely heavily on the wireless
channel. However, all the aforementioned works are only valid
for the channel which has rich scattering in the frequency band
below 3 GHz.
The explosive growth in demand for data traffic calls
for significant increases in the capacity of wireless systems.
Millimeter wave communication is a particularly promising
approach for meeting this challenge because of the huge
amounts of available spectrum possessed by millimeter wave
[13], [14]. An increasing number of recent literatures show
their interests in millimeter wave [15], [16]. Millimeter wave
channel models have been studied based on their propaga-
tion characteristics theoretically [17] and experimentally [18],
where the ray cluster model is demonstrated as the most
suitable model for the millimeter wave propagation environ-
ment. Adopting the ray cluster model, many beamforming
strategies are designed for millimeter wave systems in [19]-
[24]. Specifically, a joint spatial division and multiplexing
(JSDM) strategy is proposed for a millimeter wave multiuser
system in [19]. The beamforming schemes in millimeter wave
cellular systems are investigated in [20], [21]. Furthermore, a
hybrid RF/baseband transmission scheme is proposed in [22]
to reduce the design and implementation complexity, and the
scheme is further simplified in [23], [24]. The propagation
characteristics of millimeter wave illustrated in these studies
are different from the traditional wireless communications with
spectral band below 3 GHz, which can be summarized as
0090-6778 (c) 2016 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/TCOMM.2017.2672661, IEEE
Transactions on Communications
2
follows:
1) Scattering and multipaths in the millimeter wave band
are sparse. The scattering environment makes many of the
traditional statistically independent fading distributions inac-
curate for the millimeter wave channel. Channels under the
millimeter wave band are not independent and identically
distributed (i.i.d.) Rayleigh but rather correlated fading.
2) Due to the short wavelength inherent characteristic of
millimeter wave, more antennas can be equipped in a limited
space. Large antenna array provides a high beamforming gain
which in turn overcomes the huge path loss caused by the
increase in frequency, as well as improves the directionality
and spatial resolution [25]-[27].
Concerning specific propagation features of millimeter
wave, physical layer security in millimeter wave systems
attracts new interests [28]-[31]. Specifically, in [29], a low-
complexity directional modulation technique, namely antenna
subset modulation (ASM), is proposed by exploiting the
potential of large antenna arrays to guarantee the security.
It is concluded that ASM achieves a narrow beamwidth in
the desired direction and a high symbol error rate (SER) in
undesired directions. In [30], secrecy throughputs are analyzed
from the perspectives of delay-tolerant transmission and delay-
limited transmission, and analog beamforming using phase
shifters is utilized at the millimeter wave base station to
reduce the system cost. Secrecy performance of noise-limited
and AN assisted millimeter wave cellular networks with the
stochastic geometry framework is analyzed in [31]. However,
neither a systematic secrecy transmission scheme has been
proposed, nor the achievable secrecy performance has been
analyzed in these works. To the best of our knowledge,
no previous work has provided secure transmission schemes
and comprehensive secrecy performance analysis under a ray
cluster channel model that characterizes multipath propagation
for a millimeter wave system. So far, how to safeguard the
physical layer security of a millimeter wave system is still an
open problem, which motivates our work.
A. Our Work and Contributions
In this paper, we investigate the secure transmissions under
slow fading channels in a millimeter wave system. We assume
that an eavesdropper intercepts the communication from the
transmitter to the legitimate destination and the eavesdropper’s
instantaneous CSI is not known at the transmitter. In our
previous conference papers [32], [33], we discuss how to apply
the AN scheme to millimeter wave systems. Building on our
prior works, we investigate three secure transmission schemes
by evaluating both the SOP and the secrecy throughput, and
give more insightful and comprehensive analysis in this paper.
Our contributions are summarized as follows:
1) A discrete angular domain channel model considering
spatially resolvable paths is established to perform theoretical
design and analysis of transmission schemes for millimeter
wave systems, where the channel is constituted by the spatially
resolvable paths and represented by the spatially orthogonal
basis. Furthermore, we find that the secrecy performance of
millimeter wave system is significantly influenced by the
relationship between the destination’s and the eavesdropper’s
spatially resolvable paths, which is unlike those wireless
systems with statistically independent channel models.
2) Concerning the channel condition of the millimeter wave
system, we investigate three transmission methods. We first
analyze the new secrecy properties of the maximum ratio trans-
mitting (MRT) beamforming for the millimeter wave system.
Then we propose two secure transmission schemes, namely
AN beamforming and partial MRT (PMRT) beamforming. For
the AN scheme, we design a new form of AN generation.
For the PMRT scheme, confidential signals are emitted by
a narrow beam which keeps away from the eavesdropper’s
resolvable paths. We analyze the secrecy performance of the
AN beamforming by deriving a closed-form SOP and an
accurate integral-form secrecy throughput. We further obtain
the power allocation ratio of the information signal power to
the total transmit power that minimizes the SOP or maximizes
the secrecy throughput. For the PMRT beamforming, we
derive a closed-form secrecy throughput to evaluate its secrecy
performance.
3) We illustrate that the MRT beamforming is the best
scheme in terms of achieving the minimum SOP and the
maximum secrecy throughput when there is no common path
between the destination and the eavesdropper. For the case
where common paths exist, the AN scheme has the best
secrecy performance among the three schemes when either
the transmit power is low or the number of common paths are
large, otherwise the PMRT scheme shows its superiority.
B. Organization and Notations
This paper is organized as follows. In section II, we establish
the channel model for the millimeter wave system and describe
the related security problem. In section III, we propose three
secure transmission schemes. In section IV and V, we analyze
the secrecy performance, including the SOP and the secrecy
throughput, achieved by AN beamforming, MRT beamforming
and PMRT beamforming. We further obtain the optimal power
allocation that minimizes the SOP or maximizes the secrecy
throughput for the AN scheme. In section VI, we provide
numerical results to verify our theoretical analysis. In section
VII, we conclude our paper.
We use the following notations in this paper: bold up-
percase (lowercase) letters denote matrices (vectors). (·)∗,
(·)T,(·)H,| · |,∥ · ∥,P{·} and EA{·} denote conjugate,
transpose, conjugate transpose, absolute value, Euclidean nor-
m, probability, and mathematical expectation with respect
to A, respectively. CN(µ, σ2),Exp(λ)and Gamma(N, λ)
denote circularly symmetric complex Gaussian distribution
with mean µand variance σ2, exponential distribution with
parameter λ, and gamma distribution with parameters Nand
λ, respectively. CM×Ndenotes the space of all M×Nma-
trices with complex-valued elements. RMand Z+denote M-
dimensional real number domain and positive integer domain,
respectively. log(·),lg(·)and ln(·)denote base-2, base-10
and natural logarithms, respectively. fu(·),Fu(·)and F−1
u(·)
denote probability distribution function (PDF), cumulative
distribution function (CDF) of uand inverse function of Fu(·),
0090-6778 (c) 2016 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/TCOMM.2017.2672661, IEEE
Transactions on Communications
3
respectively. The intersection and difference between two sets
Ω1and Ω2are denoted by Ω1∩Ω2and Ω1\Ω2, respectively.
Ei(−x) = ∞
x
e−t
tdt with x > 0.
II. SY ST EM MO DE L
We consider a millimeter wave system where the con-
fidential information transmission from a transmitter to a
destination is intercepted by an eavesdropper. The transmitter
has Ntantennas, while the destination and the eavesdropper
are equipped with one antenna1, respectively.
A. Channel Model
With the sparse scattering and multipaths, the millimeter
wave channel can be represented by a ray cluster based spatial
channel model, where the channel is constituted by Ncclusters
and each cluster contains Nrpaths [18]-[20], [22]-[24]. Thus
the channel can be described as
h=Ntβ
NcNr
lc
lr
glc,lra(Θlc,lr)H,(1)
where βis the average path loss between the transmitter and
the receiver, glc,lris the complex gain of the lth
rpath in the
lth
ccluster, a(Θlc,lr)is the normalized array response (steering
vector) at the azimuth angle of departure (AOD) of θlc,lr, and
Θlc,lr
,sin(θlc,lr). When an uniform linear array (ULA) is
adopted, the normalized array response can be given by
a(Θ) = 1
√Nt
[1, e−j2πd
λΘ, e−j22πd
λΘ,·· · , e−j(Nt−1) 2πd
λΘ]T,
(2)
where dis the antenna spacing, λis the wavelength, and
generally d=λ/2. We assume the channel contains one
cluster in this paper according to the experimental results in
[18], [34]. Because in the millimeter wave system, transmit
power is most probably concentrated on one specific cluster.
B. Discrete Angular Domain Channel Model
As discussed above, by characterizing individual multipaths,
ray cluster channel model extremely conforms to the practical
physical propagation environment, and is identified as the
physical channel model of millimeter wave. However, we need
to establish a mathematical model for the theoretical analysis
purpose. Then the discrete angular domain channel model
which considers spatially resolvable paths shows up [15], [35].
The angular domain resolvability is determined by the
length of the transmit antenna array M=Ntd
λ, due to the
finite angular resolution characteristic of the array-size-limited
system. As proved in [35], paths whose Θdiffer by less than
1
Mare not resolvable by the array. Thus the basic idea of
the discrete angular domain channel model is to sample the
angular domain at a fixed spacing of 1
Mat the transmitter and
1We consider a single receive antenna for tractability. In practice, multiple
receive antennas are equipped and they will form a receive beam, which is
equivalent to a directional single antenna. This will not influence the analysis
performed in the paper. Similar assumption has also been adopted by [19],
[29], [30], etc.
to represent the channel by the spatially orthogonal basis [15],
[35], [36]. The orthogonal basis can be defined as
U,[a(Ψ1),a(Ψ2),·· · ,a(ΨN t)],(3)
where Ψi,1
M(i−1−Nt−1
2)for i= 1,2,·· · , Nt. Each
column of Uis an array response vector that represents a
physical angle θwith θ= arcsin Ψ. We can easily find that
there is a one-to-one mapping between Ψ∈(−1,1) and the
AOD θ∈[−π/2, π/2]. The orthogonal basis provides a very
simple but approximate decomposition of the total transmitted
signal into the multipath transmitted along different physical
directions up to a resolution of 1
M. We represent the ith
channel gain in terms of comprehensive effect of all paths
whose Θis within a window of width 1
Maround Ψi. Thus
the discrete angular domain channel model can be described
as
h=Ntβ
LgUH,(4)
where g= [g1, g2,·· · , gN t]is the complex gain vector. We
assume all paths’ AODs are distributed within the angular
range [θmin, θmax ], where θmin ≤θmax ∈[−π/2, π/2]. If
Ψi∈[sin(θmin),sin(θmax )], the ith column of U(the ith
orthogonal basis vector) represents a spatially resolvable path.
Due to the limited scattering feature of the millimeter wave
system, only a few of the Ntorthogonal basis vectors represent
spatially resolvable paths. Lis denoted as the number of
spatially resolvable paths with L < Nt, and can be calculated
by L=⌊Msin(θmax) + Nt+1
2⌋−⌈Msin(θmin) + Nt+1
2⌉+ 1.
Following the channel models in [18, section III E] and [37,
section III D], since each resolvable path contains several
practical paths, if Ψi∈[sin(θmin),sin(θmax )], we assume that
the ith complex gain giis a complex Gaussian coefficient with
zero mean and unit variance, i.e., gi∼ CN(0,1); otherwise,
gi= 0.
We assume that the instantaneous CSI of the destination
is perfectly known at the transmitter [8]-[12]. We define a
suspicious area where the eavesdropper’s paths are possibly
distributed in. It means that the paths’ AODs of the eaves-
dropper are within the angular range Asus ,[θE,min , θE,max],
with θE,min ≤θE,max ∈[−π/2, π/2]. Since the eavesdropper
and its surrounding obstacles may not physically change their
positions for a long period, we treat AODs as unchanged value
within several channel coherent times, and the information
of suspicious area is known at the transmitter. However,
the complex gain of each path changes rapidly. Thus the
instantaneous information of gEis assumed to be unknown,
whereas its distribution is available.
According to (4), we describe channels of the destination
and the eavesdropper as hD=NtβD
LDgDUHand hE=
NtβE
LEgEUH, where βDand βEare the path loss, gDand
gEdenote the complex gains, LDand LEare numbers of
spatially resolvable paths through the destination’s channel the
eavesdropper’s channel. Obviously, LD< Ntand LE< Nt.
As shown in Fig. 1, we define the set ΩD,{IDi|IDi ∈
Z+, IDi ∈[1, Nt], ID1< ID2<··· < ID LD}, where IDi is
an index of the basis vector which represents a destination’s
resolvable path. Define the set ΩE,{IEi|IE i ∈Z+, IEi ∈
0090-6778 (c) 2016 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/TCOMM.2017.2672661, IEEE
Transactions on Communications
4
Transmitter
D
Destination
Eavesdropper
C
!
"
E
(a) LC= 0
D
E
Transmitter
Eavesdropper
Destination
P
C
A
(b) LC= 0
Fig. 1. Description of the spatially resolvable paths.
[1, Nt], IE1< IE2<·· · < IELE}, where IEi is an index of
the basis vector which represents an eavesdropper’s resolvable
path. The set ΩC,ΩD∩ΩEdescribes the common paths of
the destination and the eavesdropper, and LCis the number
of common paths. The set ΩA,ΩE\ΩCincludes all the
elements in ΩEbut not in ΩCand the set ΩP,ΩD\ΩC
includes all the elements in ΩDbut not in ΩC.
We define the function S(B,Ω) ,[bI1,bI2,·· · ,bIK],
where B,[b1,b2,·· · ,bN], and Ω,{Ii|Ii∈Z+, Ii∈
[1, N ], I1< I2<··· < IK, K ≤N}. The function S
generates a matrix whose columns are selected from B. Define
˜
gD,S(gD,ΩD)∈C1×LD,˜
gE,S(gE,ΩE)∈C1×LE,
˜
UD,S(U,ΩD)∈CNt×LD,˜
UE,S(U,ΩE)∈CNt×LE,
gD,(C),S(gD,ΩC)∈C1×LC,gD,(P),S(gD,ΩP)∈
C1×(LD−LC),gE,(C),S(gE,ΩC)∈C1×LCand gE,(A),
S(gE,ΩA)∈C1×(LE−LC). Thus we can rewrite hD=
NtβD
LD˜
gD˜
UH
Dand hE=NtβE
LE˜
gE˜
UH
E.
C. Problem Description
The capacities of the destination’s channel and the eaves-
dropper’s channel can be respectively given by CD= log(1 +
ξD)and CE= log(1 + ξE), with ξDand ξEdenoting the
received signal-to-interference-plus-noise ratios (SINRs). By
utilizing the well-known Wyner’s wiretap encoding scheme,
Rt,Rsand Re,Rt−Rsare defined as codeword rate,
secrecy rate and rate redundancy, where Reis exploited to
offer secrecy against eavesdropping.
The secure information can not be correctly recovered by
the destination when CD< Rt. Besides, once CEexceeds
Re, perfect secrecy is destroyed and a secrecy outage occurs.
To avoid the above two undesirable situations, we adopt an
adaptive on-off transmission scheme proposed in [12], where
the transmitter decides when to radiate signals based on the
instantaneous CSI of the destination. Throughout the paper, for
notational brevity, we define γ,∥˜
gD∥2as the overall channel
gain of the destination, γC,∥gD,(C)∥2and γP,∥gD,(P)∥2.
Obviously, γ=γC+γP. Define γ◦,(γC, γP)∈R2as the
instantaneous gains of the common and non-common paths of
the destination. Define Υ, the transmission region, as the set
of those γ◦satisfying the transmission constraint. Since the
channel gain γ◦varies from time to time, the transmitter emits
signals only when γ◦∈Υ; otherwise, the transmitter keeps
silence. Define the SOP for a given γ◦as
Pso =P{CE> Rt−Rs|γ◦},∀γ◦∈Υ.(5)
Secrecy throughput measures the effective average transmis-
sion rate of the confidential message, which is defined as
τ=Eγ◦[Rs(γ◦)] ,(6)
where Rs(γ◦) = 0 for γ◦/∈Υ.
III. TRA NS MI SS IO N SCH EM ES
In this section, we first analyze the secrecy performance of
the MRT beamforming in the millimeter wave system as a
benchmark. Next, by considering the channel features of mil-
limeter wave, we propose two transmission schemes, namely
AN beamforming and PMRT beamforming, to improve the
secrecy performance of the MRT beamforming for the case
where the destination and the eavesdropper have common
resolvable paths.
A. MRT Beamforming
The MRT beamforming strategy is a simple but effective
transmission method. Although it is studied by lots of the
researchers, we will study the new security properties that it
presents in the millimeter wave system. The transmitter sends
one data stream, i.e, Ns= 1. The signals received at the
destination and the eavesdropper are
yMRT
D=√PhDw1s+nD,(7)
yMRT
E=√PhEw1s+nE,(8)
where w1=hH
D/∥hD∥is the beamforming vector, Pis the
total transmit power, sis the information bearing signal with
E[|s|2] = 1,nDand nEare i.i.d. additive white Gaussian
noise with zero-mean and variance σ2
n, i.e., nD∼ CN (0, σ2
n)
and nE∼ CN(0, σ2
n). Then the SNRs of the destination and
the eavesdropper can be respectively described as
ξMRT
D=P∥hD∥2
σ2
n
=P βDNt
σ2
nLD∥gDUH∥2=P βDNt
σ2
nLD∥˜
gD∥2,
(9)
ξMRT
E=P|hEhH
D|2
σ2
n∥hD∥2=P βENt
σ2
nLE
|gEUHUgH
D|2
∥gDUH∥2
=P βENt|gE,(C)gH
D,(C)|2
σ2
nLE∥˜
gD∥2.(10)
0090-6778 (c) 2016 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/TCOMM.2017.2672661, IEEE
Transactions on Communications
5
Only if i∈ΩD∩ΩE, i.e., i∈ΩC,gEi g∗
Di = 0. Thus we have
gEgH
D=gE,(C)gH
D,(C).
From the above two equations, several observations can be
obtained.
1) In the situation that LC= 0, i.e., the destination
and the eavesdropper have no common path, array steer-
ing vectors in ˜
UDand ˜
UEare orthogonal. hEhH
D=
NtβE
LENtβD
LD˜
gE˜
UH
E˜
UD˜
gH
D= 0 leads to ξMRT
E= 0 and
CMRT
E= 0, such that Pso(γ◦) = 0. Therefore, when LC= 0,
the secrecy outage never occurs.
2) From the definitions of gEand gH
D, they are uncor-
related. gE,(C)is a vector whose elements are all selected
from gE, and gH
D,(C)is a vector whose elements are all
selected from gH
D. Therefore, gE,(C)and gH
D,(C)are un-
correlated. Moreover, as gE,(C)and gH
D,(C)are Gaussian
distribution vectors, we have that gE,(C)and gH
D,(C)are
independent. Since each element of gE,(C)follows a Gaus-
sian distribution with zero mean and unit variance, which
is independent of the unit-norm vector gH
D,(C)
∥gD,(C)∥, we have
that |gE,(C)
gH
D,(C)
∥gD,(C)∥|2∼Exp(1) [27]. Therefore, we ob-
tain EξMRT
E=P βENt∥gD,(C)∥2
σ2
nLE∥˜
gD∥2E|gE,(C)
gH
D,(C)
∥gD,(C)∥|2=
P βENt∥gD,(C)∥2
σ2
nLE∥˜
gD∥2. When LCincreases, ∥gD,(C)∥2at the nu-
merator of EξMRT
Eincreases, which leads to an increase
in EξMRT
E. The secrecy performance becomes worse when
LCincreases.
3) The conclusion implies that the secrecy performance
of millimeter wave system is dramatically influenced by the
relationship between spatially resolvable paths of the desti-
nation and the eavesdropper, which is unlike those wireless
systems with statistically independent channel models. That is
because array steering vectors are the key constituent part of
the channel formulas.
For the situation of LC= 0, there are two basic ideas
to improve the secrecy performance. One idea is emitting
interference to deteriorate the eavesdropper’s channel. The
other idea is to avoid signal leakage to the eavesdropper’s
channel. Based on those two ideas, we propose two transmis-
sion schemes, namely the AN beamforming and the PMRT
beamforming.
B. AN Beamforming
The idea of AN beamforming is to emit information bearing
signal together with AN in order to confuse the eavesdropper.
AN is transmitted in the null space of the destination’s channel
to guarantee the fact that the destination is not interfered with
AN. Utilizing AN is an effective way to improve secrecy
performance when the eavesdropper’s instantaneous CSI is
unknown at the transmitter. However, the complexity of the
null space calculation which includes singular value decom-
position (SVD) depends on the number of transmit antennas.
Considering the expectation of the 5G millimeter wave system,
there will be hundreds of transmit antennas, which leads to a
high complexity. Compared with the statistically independent
channel, millimeter wave channel has specific propagation
characteristics, so we propose a new AN generating method in
this paper which is much simpler but effective. In our method,
AN is transmitted to the directions of the eavesdropper’s
resolvable paths excluding the directions of common paths.
The transmitted signal can be given by
x=ηP w1s+(1 −η)P
LE−LC
W2z,(11)
where W2=S(U,ΩA) = [w2,1,w2,2,·· · ,w2,(LE−LC)]∈
CNt×(LE−LC)is the AN beamforming matrix, z∈
C(LE−LC)×1is the AN bearing signal with E[zzH] =
ILE−LC,ηis the power allocation ratio of the information
signal power to the total transmit power with 0≤η≤1. From
the definition of W2, we find that the generating method of
the AN beamforming matrix is simple as we only select some
columns from U. Since ΩA= ΩE\ΩC, we have hDW2=0.
Therefore, utilizing specific characteristics of the millimeter
wave channel is beneficial for reducing the complexity of the
design and application for AN. It should be noted that the
case η= 1 is equivalent to the MRT beamforming where the
transmitter does not transmit AN but only information bearing
signal with the total transmit power P. Signals received by the
destination and the eavesdropper can be described as
yAN
D=ηP hDw1s+(1 −η)P
LE−LC
hDW2z+nD,(12)
yAN
E=ηP hEw1s+(1 −η)P
LE−LC
hEW2z+nE.(13)
Then the SINRs of the destination and the eavesdropper can
be respectively formulated as
ξAN
D=ηP ∥hD∥2
σ2
n
=ηP βDNtγ
σ2
nLD
,(14)
ξAN
E=ηP |hEw1|2
(1−η)P
LE−LC∥hEW2∥2+σ2
n
=
ηP βENt
LE∥˜
gD∥2|gEUHUgH
D|2
(1−η)P βENt
(LE−LC)LE∥gEUHW2∥2+σ2
n
=
ηP βENt
LE∥˜
gD∥2|gE,(C)gH
D,(C)|2
(1−η)P βENt
(LE−LC)LE∥gE,(A)∥2+σ2
n
=
ηP βENtγC
σ2
nLEγ|gE,(C)
gD,(C)H
∥gD,(C)∥|2
(1−η)P βENt
σ2
n(LE−LC)LE∥gE,(A)∥2+ 1 .(15)
By denoting gEUHW2=χA,1, χA,2,·· · , χA,(LE−LC), we
have χA,j =Nt
i=1 gEia(Ψi)Hw2,j . Since W2=S(U,ΩA)
and Uis a unitary matrix, we get gEUHW2=gE,(A).
AN transmission effectively interferes with the eavesdrop-
per, but decreases the power utilized for data transmission
and reduces the signal power at the destination. Therefore, the
power allocation ratio ηis critical to ensure a good secrecy
performance. Thus we will study the optimal power allocation
that minimizes the SOP or maximizes the secrecy throughput
in section IV.
0090-6778 (c) 2016 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/TCOMM.2017.2672661, IEEE
Transactions on Communications
6
C. PMRT Beamforming
For the MRT scheme, the transmitting beams steer in
all directions of the destination’s resolvable paths, which
contain LCpaths to the eavesdropper in the situation that
LC= 0. That means the eavesdropper also receives the
confidential signal. We propose a narrow-beam transmission
scheme, which is called partial MRT (PMRT), aiming at
the above problem to improve the secrecy performance. The
PMRT scheme transmits signals only in partial directions
of the destination’s resolvable paths excluding the directions
of common paths. The PMRT beamforming matrix w3=
(gD,(P)UH
(P))H/∥gD,(P)UH
(P)∥, where U(P)=S(U,ΩP) =
uP,1,uP,2,··· ,uP,(LD−LC)∈CNt×(LD−LC). Signals re-
ceived by the destination and the eavesdropper are yP
D=
√PhDw3s+nDand yP
E=√PhEw3s+nE, respectively.
Then the SINRs of the destination and the eavesdropper can
be respectively described as
ξP
D=P βDNt
σ2
nLD
|gDUHU(P)gH
D,(P)|2
gD,(P)UH
(P)U(P)gH
D,(P)
=P βDNt
σ2
nLD
γP,(16)
ξP
E=P βENt
σ2
nLE
|˜
gE˜
UH
EU(P)gH
D,(P)|2
gD,(P)UH
(P)U(P)gH
D,(P)
= 0.(17)
Denoting gDUHU(P)=χP,1, χP,2,··· , χP,(LD−LC), we
have χP,j =Nt
i=1 gDia(Ψi)HuP ,j . Since U(P)=S(U,ΩP),
we get gDUHU(P)=gD,(P). From ˜
UE=S(U,ΩE)and
ΩE∩ΩP= Ø, we have ˜
UH
EU(P)=0.
There is a special case LD=LE=LCwhich means
that all the destination’s and eavesdropper’s resolvable paths
overlap. For the case LE=LC, i.e., ΩE= ΩC,W2= Ø and
no AN is emitted in the AN scheme. It is the same as the MRT
scheme and all the power is allocated to the transmission of
information bearing signal, i.e., η= 1. For the case LD=LC,
i.e., ΩD= ΩC,w3= Ø in the PMRT scheme. Since LCis
large, the eavesdropper receives a large amount of confidential
messages, hence both the AN scheme and the PMRT scheme
do not work any more. However, in practice, this extreme
case is hard to occur due to the following two reasons. 1)
Since the reflection and scattering in practical propagation
environment are complicated, signals transmitted to receivers
at different locations encounter various reflection objects and
go through different paths. Even in the same direction, the
AODs of receivers may be different. 2) Furthermore, in the
5G millimeter wave system, when the number of transmit
antennas becomes large, the spacing 1
Mgets small and the
spatial horizon is divided exquisitely. Thus little difference
of paths between the destination and the eavesdropper can be
found. Therefore it is reasonable for us to assume LE−LC>0
or LD−LC>0in our study. For the situation LD=LCand
LE−LC>0, we can utilize the AN scheme to guarantee the
secure transmission. Whereas for the situation LE=LCand
LD−LC>0, we can exploit the PMRT scheme to improve
the secrecy performance. For the situation LD−LC>0and
LE−LC>0, both two schemes work.
IV. SEC RE CY PE RF OR MA NC E OF AN BE AM FO RM IN G AN D
MRT BEAMFORMING
In this section, we investigate the secrecy performance
of AN beamforming. We first analyze the secrecy outage
performance by deriving a closed-form SOP and design the
optimal power allocation η◦to minimize the SOP. Then we
maximize the secrecy throughput under a given SOP constraint
and obtain the optimal power allocation ratio η∗. In addition,
we analyze the secrecy performance of MRT beamforming as
a special case of AN beamforming when the power allocation
ratio η= 1.
A. SOP Minimization and Optimal Power Allocation
We first derive the CDF of ξAN
Ein (15). Define u,
|gE,(C)
gH
D,(C)
∥gD,(C)∥|2, we have u∼Exp(1). Define v,
∥gE,(A)∥2, we have v∼Gamma(LE−LC,1). Since ΩC∩
ΩA= Ø,uand vare independent to each other. Moreover,
we define a(γ◦),P βENtγC
σ2
nLE(γC+γP),b,P βENt
σ2
n(LE−LC)LEand
c(γ◦),P βDNt(γC+γP)
σ2
nLD. For notational brevity, we omit γ◦
from a(γ◦)and c(γ◦), and treat them as functions of γ◦by
default. The CDF of ξAN
Ecan be given by
FAN
ξE(x) = Pηau
(1 −η)bv + 1 < x
=Pu < (1 −η)bxv +x
ηa = 1 −Eve−(1−η)bx
ηa v−x
ηa
= 1 −e−x
ηa ∞
0
vLE−LC−1
Γ(LE−LC)e−[1+ (1−η)bx
ηa ]vdv
(a)
= 1 −e−x
ηa 1 + (1 −η)bx
ηa −(LE−LC)
,
(18)
where (a)holds for the integration formula [38, 3.326.2].
Given that the instantaneous CSI of the destination is avail-
able at the transmitter, we set Rtequal to CDand minimize
the SOP under each destination’s channel realization. Based on
the on-off transmission scheme, to get a better secrecy perfor-
mance, CDshould exceed Rsin order to guarantee a positive
Reagainst the eavesdropper, i.e., only if CD> Rs, the
transmitter radiates signals, otherwise, transmission should be
suspended. Since ξAN
D=ηc, we obtain that η > ηmin ,T−1
c,
where T,2Rs. From (18), the SOP defined in (5) can be
expressed as
Pso(γ◦),P{CE> CD−Rs|γ◦}
=PξAN
E>ξAN
D−(T−1)
T
= exp −ηc −(T−1)
ηaT
×1 + (1 −η)b[ηc −(T−1)]
ηaT −(LE−LC)
.
(19)
We derive the following properties from the above equation:
1) Pso monotonically decreases with P. Since
dPso
dP =−T−1
ηT a2da
dP +(1−η)γ
γCT δ
dc
dP Pso, where
δ= 1 + (1−η)b[ηc−(T−1)]
ηaT >0, and da
dP >0,dc
dP >0,
0090-6778 (c) 2016 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/TCOMM.2017.2672661, IEEE
Transactions on Communications
7
we have dPso
dP <0. It implies that secrecy outages are more
likely to occur in low transmit power scenarios.
2) Pso monotonically decreases with βD,
while increases with βE. On one hand, since
dPso
dβD=−1
aT +(1−η)γ
γCT δ Pso dc
dβDand dc
dβD>0, we have
dPso
dβD<0. On the other hand, since dPso
dβE=ηc−(T−1)
ηT a2Pso da
dβE
and da
dβE>0, we obtain dPso
dβE>0. As βdecreases with
the increase of the distance between the receiver and the
transmitter, we find either a closer destination or a farther
eavesdropper will be beneficial for decreasing Pso.
Then the problem of the SOP minimization can be formu-
lated as
min
ηPso(γ◦),
s.t. ηmin < η ≤1.
(20)
The minimum of Pso(γ◦)can only be achieved at the
boundary η= 1 or those zero-crossing points of the first-
order derivative of Pso(γ◦). The derivative of Pso(γ◦)can be
given by
dPso
dη =bc(LE−LC)η−3
aT exp −ηc −(T−1)
ηaT
×1 + (1 −η)b[ηc −(T−1)]
ηaT −(LE−LC)−1
×η3+ε1η2+ε2η+ε3,
(21)
where ε1=T−1
aT (LE−LC),ε2=−(T−1)[aT +b(c+T−1)]
aT bc(LE−LC)−T−1
c
and ε3=(T−1)2
aT c(LE−LC). Denote K(η) = η3+ε1η2+ε2η+
ε3. We find the sign of dPso
dη follows that of K(η). In other
words, to investigate the monotonicity of Pso on η, we need
to examine the sign of K(η). For notational brevity, we define
¯a,P βENt
σ2
nLE=γC+γP
γCaand ¯c,P βDNt
σ2
nLD=1
γC+γPc. In the
following theorem, we provide the solution to problem (20).
Theorem 1: The optimal η◦that minimizes the SOP is
η◦=
Ø, γ ≤T−1
¯c,
1,T−1
¯c< γ ≤(¯a+ 1)(T−1)
¯a¯c,
η⋆,otherwise,
(22)
where η⋆=3
−q
2+p3
27 +q2
4+3
−q
2−p3
27 +q2
4−ε1
3
with p=ε2−ε2
1
3,q=ε3−ε1ε2
3+2ε3
2
27 , and η◦= Ø means
that the transmission is suspended.
Proof: Please see Appendix A.
Theorem 1 indicates that when γis small which corresponds
to a poor quality of the destination’s channel, the transmitter
either suspends the transmission or transmits information bear-
ing signals with full power. When γbecomes large enough,
it is wise to emit AN to decrease the SOP. The resulting
minimum Pso(γ◦)is obtained by substituting η◦into (19).
Since we minimize the SOP under each destination’s channel
realization, the overall SOP is consquently minimized.
B. Secrecy Throughput Maximization and Optimal Power Al-
location
In this subsection, we maximize the secrecy throughput
under a SOP constraint through a dynamic transmission pa-
rameter scheme, which means values of Rs,Rtand ηare
chosen dynamically for each destination’s channel realization.
From the definition in (6), the secrecy throughput will be
maximized if we maximize the secrecy rate Rs(γ◦)for each
γ◦. Therefore, the optimization problem can be formulated as
max
η,Rt
Rs(γ◦),
s.t. 0< Rs(γ◦)< Rt(γ◦)≤CD,(23a)
Pso(γ◦)≤ϵ, (23b)
0≤η≤1,(23c)
where ϵ∈[0,1] is a pre-set SOP threshold, (23a)-(23c) show
the constraints for reliable transmission, secrecy outage and
power allocation, respectively.
Lemma 1: The optimization problem in (23) is equivalent
to
max
ηRs(γ◦) = log 1 + ηc
1 + ηκ(η),
s.t. 0≤η≤1, c > κ(η),
(24)
where κ(η),F−1
ξE(1−ϵ)
η,c > κ(η)is the transmission
constraint of the on-off transmission scheme, which means
that the transmitter emits signals only when this constraint is
satisfied.
Proof: Please see Appendix B.
The definition of κ(η)yields FAN
ξE(ηκ(η)) = 1 −ϵ, which
is equivalent to
Q1(κ) = 0,(25)
where Q1(κ),e−κ(η)
a1 + (1−η)bκ(η)
a−(LE−LC)−ϵ.
Lemma 2: κ(η)is a monotonically increasing and convex
function of ηin the range η∈[0,1].
Proof: Please see Appendix C.
Although κis an implicit function of η, we can calculate
κefficiently by utilizing Lamma 2. We simply observe κ≥0
from the definition of κ. Lamma 2 shows that the maximum
κcan be get when η= 1, which is κmax =−aln ϵaccording
to (25). Given an arbitrary η,Q1(κ)monotonically decreases
with κ. Combined with this and Q1(0) = 1 −ϵ > 0,
Q1(κmax) = ϵ[1 −(1 −η)bln ϵ]−(LE−LC)−ϵ≤0, we derive
that Q1has the unique zero-crossing point. Therefore, for a
given η, we can calculate κthat satisfies (25) by utilizing the
bisection method within the range [0, κmax]instead of a one-
dimensional exhaustive search. Then we give the implemen-
tation steps of the bisection method. We start with an interval
[κl, κu]known to contain the solution of κ, and the initial
values of κland κuare 0and κmax respectively. We calculate
Q1(κ)at its midpoint κm= (κl+κu)/2, to determine whether
the solution is in the lower or upper half of the interval, and
update the interval accordingly, i.e., if Q1(κm)<0, we set
κu=κm; if Q1(κm)>0, we set κl=κm. This produces
a new interval, which also contains the solution of κ, but has
0090-6778 (c) 2016 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/TCOMM.2017.2672661, IEEE
Transactions on Communications
8
half the width of the initial interval. This is repeated until the
width of the interval is small enough.
Theorem 2: Given c > κ(η),Rsis a concave function of
η. The optimal η∗that maximizes Rsis given by
η∗=
1,c
1 + c+[1 + (LE−LC)b]aln ϵ
1−aln ϵ>0,
η⋆,otherwise,
(26)
where η⋆is the unique root of the following equation
dRs
dη =1
ln 2 c
1 + ηc −κ+ηdκ
dη
1 + ηκ = 0,(27)
with dκ
dη defined in (37).
Proof: Please see Appendix D.
We can efficiently search the optimal η⋆by exploiting the
bisection method due to the concave characteristic of Rswith
respect to η.
Corollary 1: The optimal η⋆monotonically increases with
respect to βD,ϵand γP, and monotonically decreases with
recpect to βE.
Proof: Please see Appendix E.
Corollary 1 indicates that in order to promote the secrecy
rate, more power should be: 1) allocated to the information
signal under a better quality of the destination’s channel (a
larger βD) or a higher proportion of common paths’ power to
the total destination’s channel gain (a larger γP), 2) allocated
to the AN under a better quality of the eavesdropper’s channel
(a larger βE) or a stronger SOP constraint (a smaller ϵ).
We obtain the maximum secrecy rate R∗
s(γ◦)after get-
ting the optimal power allocation ratio η∗. The transmis-
sion region defined in subsection II-C can be expressed as
Υ = {γ◦|c > κ}. The transmitter examines the the constraint
γ◦∈Υfor each instantaneous channel realization of the
destination to decide whether starting communication or not,
i.e., the transmitter emits signals with R∗
s(γ◦)only when
γ◦∈Υwhich guarantees a positive Rs; otherwise, the
transmitter keeps silence and we set Rs(γ◦) = 0. Evidently,
if the optimal η∗that maximizes Rs(γ◦)does not meet the
constraint, any other ηcan not satisfy it either. So we can only
check the transmission constraint by substituting η∗. Since
ΩC∩ΩP= Ø, we find that γC∼Gamma(LC,1) and
γP∼Gamma(LD−LC,1) are independent to each other.
Together with (6), the maximum secrecy throughput of AN
beamforming can be given by
τ∗= Υ
R∗
s(γ◦)fγP(x)fγC(y)dxdy, (28)
where fγC(x) = xLC−1
Γ(LC)e−xand fγP(x) = xLD−LC−1
Γ(LD−LC)e−xare
the PDFs of γCand γP, respectively.
C. Secrecy Performance of MRT Beamforming
Since MRT beamforming is a special case of AN beamform-
ing with η= 1, we can easily analyze the secrecy performance
of MRT beamforming from the results of AN beamforming
and derive several more concise results.
By substituting η= 1 into (19), we obtain the SOP of MRT
beamforming Pso(γ◦) = exp −c−(T−1)
aT . According to (24),
since κ=−aln ϵwhen η= 1, the maximum Rs(γ◦)can be
given by
R∗
s(γ◦) = log 1 + c
1−aln ϵ= log (γC+γP)[1 + ¯c(γC+γP)]
γP+ (1 −¯aln ϵ)γC
,
(29)
with the transmission region Υ = {γ◦|γP>−γC+
−¯aln ϵ
¯c√γC}.
Corollary 2: The maximum secrecy throughput of MRT
beamforming can be given by
τ∗=
LD−LC−1
m=0
1
Γ(LD−LC−m)∞
0
fγC(y)e−ς
×ςLD−LC−1−m[Q3(µ1) + Q3(µ2)−Q3(µ3) + µ4]dy,
(30)
where ς=−γC+−¯aln ϵ
¯c√γC,µ1=1
y+ς,µ2=
¯c
¯cy+¯cς +1 ,µ3=1
(1−¯aln ϵ)y+ς,µ4= log (y+ς)(¯cy +¯cς+1)
(1−¯aln ϵ)y+ς
and Q3(x) = 1
ln 2 m
n=1 1
(m−n)! [(−1)m−n−1
xm−ne1
xEi(−1
x) +
m−n
k=1 (k−1)!(−1
x)m−n−k].
Proof: Please see Appendix F.
We need to mention that when LC= 0,CMRT
E= 0,
we have Pso(γ◦)=0for the MRT beamforming, hence the
outage constraint Pso(γ◦)≤ϵalways holds and the maximum
R∗
s(γ◦) = Rt=CMRT
D. For the MISO channel, the MRT
beamforming can achieve the largest receive SNR ξDat the
destination with the given transmit power P. Any other scheme
has a lower received SNR and hence its reliable rate must be
less than CMRT
D. Thus τ=Eγ◦[Rs(γ◦)] achieved by the
MRT scheme reaches the maximum achievable value. In a
word, the MRT beamforming is the best scheme in terms of
achieving the maximum secrecy throughput under the scenario
of LC= 0.
V. SE CR EC Y PER FO RM AN CE O F PMRT BEAMFORMING
In this section, we analyze the SOP and the secrecy through-
put of the PMRT beamforming. Since ξP
E= 0 and CP
E= 0,
Pso(γ◦)defined in (5) equals to 0. We observe that secrecy
outage never occurs when PMRT beamforming is adopted,
which is a significant property of the PMRT scheme.
Since the SOP constraint Pso(γ◦)≤ϵalways holds, the
maximum achievable value of Rs(γ◦)is formulated as
R∗
s(γ◦) = Rt=CP
D= log(1 + ¯cγP).(31)
The maximum secrecy throughput is given by
τ∗=∞
0
R∗
s(γ◦)fγP(x)dx
=∞
0
log(1 + ¯cx)xLD−LC−1
Γ(LD−LC)e−xdx
=1
ln 2
LD−LC−1
n=1
1
(LD−LC−1−n)!
×(−1)LD−LC−n
¯cLD−LC−1−ne1
¯cEi(−1
¯c)
+
LD−LC−1−n
k=1
(k−1)!(−1
¯c)LD−LC−1−n−k.
(32)
0090-6778 (c) 2016 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/TCOMM.2017.2672661, IEEE
Transactions on Communications
9
The above equation holds for the integration formula [38,
4.337.5]. We observe that τ∗monotonically increases with
¯c, hence τ∗monotonically increases with P,Ntand βD. It
implies that increasing transmit power, increasing the trans-
mitter’s antennas or decreasing the distance between the
transmitter and the destination are beneficial for increasing
the secrecy throughput.
At the low-SNR regime as P→0, we have ¯c→0. Since
lim
y→01+y=ey, the maximum secrecy throughput can be given
by
τ∗=1
ln 2 ∞
0
¯cx xLD−LC−1
Γ(LD−LC)e−xdx =1
ln 2 ¯c(LD−LC).
(33)
We easily observe that τ∗monotonically decreases with
respect to LC. Although we derive this property at the low
SNR regime, we find the property is also correct when SNR
becomes higher from the numerical results. It is because that as
the number of common paths LCdecreases, the beamforming
matrix matches with more destination’s resolvable paths and
the destination’s capacity CDincreases, hence the secrecy
throughput increases.
VI. NU ME RI CA L RES ULTS
In this section, numerical results are presented to verify our
theoretical analysis. The location coordinates of the transmitter
and the destination are (0,0) and (rD,0), respectively. The
transmitter is equipped with a 100-antenna ULA, where the
antenna spacing dis set to be half wavelength. We assume
the angular spread of each cluster is σϕ= 20◦[18], [19],
and the cluster’s central AOD of the destination is ϕD= 0◦.
Therefore the AODs for paths transmitted to the destination
are within the angular range [−10◦,10◦]. Considering an
eavesdropper at a distance rE= 100m from the transmitter,
when the eavesdropper moves, LCvaries. According to the
experimental results obtained in [18], the path loss is modeled
as β= 10−1
10 [ρ1+ρ210 lg(r)], where ris the distance in meters,
ρ1= 61.4and ρ2= 2 are the measurement results for 28 GHz
frequency. The noise power σ2
n=−50 dBm.
A. Secrecy Performance of AN Beamforming
Fig. 2 describes SOP versus the number of common paths
LCfor different P’s. We see that Pso significantly reduces
as LCdecreases or Pincreases. It is because that as LC
becomes larger, more confidential information is leaked to the
eavesdropper’s channel. We also find that AN beamforming
has obvious superiority over MRT beamforming in terms of
achieving lower SOP. Therefore, AN transmission is beneficial
to security. The results of Monte-Carlo simulations match well
with the analytical values.
Fig. 3 illustrates the optimal power allocation that minimize
Pso(γ◦)versus Pfor different Rs’s. “Searched η◦” refers to
the results obtained by exhaustively searching the minimum
of (19) and “theoretical η◦” corresponds to the root of the
cubic equation in (22). It is easy to see that the theoretical η◦
matches well with the searched η◦. Evidently, η◦becomes
lower with an increase in Por a decrease in Rs. The
1 3 5 7 9 11 13 15 17 19
10−5
10−4
10−3
10−2
10−1
100
LC
Secrecy Outage Probability
P= 0dBm,analytica l
P= 0dBm,simulation
P= 5dBm,analytica l
P= 5dBm,simulation
AN
MRT
Fig. 2. SOP versus LCfor different P’s, with Nt= 100,Rs= 5bit/s/Hz,
LD= 20 and rD= 100m.
0 5 10 15 20
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
P(dBm)
η° that Minimizes the SOP
theoretical η◦
searched η◦
Rs= 5
Rs= 3
Rs= 4
Fig. 3. Optimal power allocation ratio η◦that minimizes the SOP versus P
for different Rs’s, with Nt= 100,LC= 11,LD= 20 and rD= 100m.
−10 −5 0 5 10 15 20
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
P(dBm)
η* that Maximizes the Secrecy Throughput
LC= 6
LC= 11
LC= 16
Fig. 4. Optimal power allocation ratio η∗that maximizes the secrecy
throughput versus Pfor different LC’s, with Nt= 100,rD= 100m,
LD= 20 and ϵ= 0.01.
underlying reason is that to support a higher Rs, more power
should be allocated to the information bearing signal.
Fig. 4 shows the optimal η∗that maximizes the secrecy
throughput versus Pfor different LC’s. At the low transmit
power region, η∗keeps 1, which reveals that full power should
0090-6778 (c) 2016 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/TCOMM.2017.2672661, IEEE
Transactions on Communications
10
1 3 5 7 9 11 13 15 17 19
0.2
0.3
0.4
0.5
0.6
0.7
0.8
LC
η* that Maximizes the Secrecy Throughput
ǫ= 0.1
ǫ= 0.01
ǫ= 0.001
Fig. 5. Optimal power allocation ratio η∗that maximizes the secrecy
throughput versus LCfor different ϵ’s, with Nt= 100,P= 15dBm,
LD= 20 and rD= 100m.
−10 −5 0 5 10 15 20
0
1
2
3
4
5
6
7
8
P(dBm)
Secrecy Throughput
MRT
AN
PMRT
Fig. 6. Secrecy throughput versus P, with Nt= 100,rD= 100m, LD=
20,LC= 16 and ϵ= 0.01.
be allocated to information signals to guarantee a reliable
link to the destination. Then η∗drops with the increase of
P, which implies that more power is permitted for the AN
transmission to confuse the eavesdropper on the premise that
the outage constraint is satisfied. In addition, as LCincreases,
η∗decreases. Since the eavesdropper receives more power of
confidential signals in the larger LCscenario, we should shift
more power to AN transmission in order to interfere with the
eavesdropper.
Fig. 5 presents the optimal η∗that maximizes the secrecy
throughput versus LCfor different outage threshold ϵ’s. As
shown in the figure, η∗decreases as ϵdecreases, which
indicates that a larger fraction of power should be allocated
to AN in order to confuse the eavesdropper and achieve a
tolerable SOP under a more rigorous SOP constraint.
B. Comparison of Transmission Schemes
Fig. 6 describes the secrecy throughput versus P. Obvious-
ly, by increasing transmit power P, the secrecy throughput
increases. Furthermore, we obtain an important observation
that to achieve higher secrecy throughput, the PMRT scheme
1 3 5 7 9 11 13 15 17 19
0
1
2
3
4
5
6
7
8
9
LC
Secrecy Throughput
MRT,rD= 100m
MRT,rD= 150m
AN,rD= 100m
AN,rD= 150m
PMRT,rD= 100m
PMRT,rD= 150m
Fig. 7. Secrecy throughput versus LCfor different rD’s, with Nt= 100,
P= 15dBm, LD= 20 and ϵ= 0.1.
is a better choice than the AN scheme at the high transmit
power regime, whereas the AN scheme shows its superiority
over the PMRT scheme. The underlying reason is that at the
low SNR regime, CDis small, hence CDis the dominant
factor contributing to the secrecy throughput. For the AN
scheme, most of the transmit power or even the full power
should be allocated to the information bearing signals in
order to increase CD. Since information bearing signals of
AN beamforming cover all directions of the destination’s
paths while those of PMRT beamforming only cover partial
directions of the destination’s paths, the destination’s capacity
of AN beamforming is larger than that of PMRT beamforming.
Therefore AN beamforming is better. At the high SNR regime,
for the AN scheme, the eavesdropper receives a large amount
of the confidential message’s power, so we should allocate
most of the power to AN to confuse the eavesdropper and only
a little power is allocated to the information bearing signals to
support CD. For PMRT beamforming, full power is utilized
to transmit the confidential signals which leads to a large CD.
Therefore PMRT beamforming is better.
In addition, Fig. 6 shows that as a special case of AN
beamforming, MRT beamforming presents a comparable per-
formance to AN beamforming under the low transmit power
scenario. However, with the increase of the transmit power,
the superiority of AN beamforming becomes more obvious.
Fig. 7 plots the secrecy throughput as a function of LC
for different rD’s. Secrecy throughput becomes higher with
the decrease of rD, which confirms the fact that increasing
βDis helpful to enhance the secrecy throughput performance.
Meanwhile, when LCdecreases, since the beamforming ma-
trix matches with more destination’s resolvable paths in the
PMRT scheme and the eavesdropper receives fewer confi-
dential messages in the AN scheme, the secrecy throughput
becomes higher.
Besides, we obtain an important observation about the supe-
riorities among the three schemes when LCvaries from Fig. 7.
On one hand, under large LCscenarios, AN beamforming has
the best performance among the three schemes. The underlying
reason is that, in order to keep away from the directions of the
0090-6778 (c) 2016 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/TCOMM.2017.2672661, IEEE
Transactions on Communications
11
5 10 15 20
0
2
4
6
8
10
∆φ(◦)
Secrecy Throughput
ASM[27]
MRT
AN
PMRT
Fig. 8. Secrecy throughput versus ∆ϕ, with Nt= 100,P= 15dBm,
rD= 100m and ϵ= 0.1.
eavesdropper’s resolvable paths in the large LCcase, signals
of PMRT beamforming are emitted in very few directions
of the destination’s resolvable paths which leads to a poor
match with the destination’s channel. However, information
bearing signals of AN beamforming cover all directions of
the destination’s paths to reach a high capacity CD. Although
the eavesdropper receives more information, it is disturbed
by the AN. On the other hand, under small LCscenarios,
PMRT beamforming has the best performance among the three
schemes. It is because that PMRT beamforming matches with
most directions of the destination’s paths which leads to a
large CD, meanwhile the eavesdropper can not intercept any
information.
Fig. 8 presents the secrecy throughput versus ∆ϕto com-
pare secrecy performances between transmission schemes in-
vestigated in this paper and the ASM scheme proposed in
[27], where ∆ϕ=ϕE−ϕDis the angular difference between
the cluster’s central AOD of the eavesdropper and that of the
destination. We set the number of antennas in each subset
equal to 30 for the ASM scheme. Obviously, transmission
schemes analyzed in this paper have better performances in
terms of achieving higher secrecy throughput than the ASM
scheme.
VII. CON CL US IO N
This paper provided the secure transmission scheme design
and comprehensive performance analysis under slow fading
channels considering multipath propagation for millimeter
wave systems. In consideration of the specific propagation
features of the millimeter wave, discrete angular domain chan-
nel model was proposed considering spatially resolvable paths
from the perspective of analysis and design of the millimeter
wave system. In the absence of the common path between
the destination and the eavesdropper, the MRT beamforming
was demonstrated as the best scheme. In the presence of
common paths, two transmission schemes were proposed to
improve the secrecy performance. The closed-form SOP was
obtained for both schemes. Furthermore, the secrecy through-
put was analyzed by deriving an accurate integral expression
for the AN scheme and a closed-form formula for the PMRT
scheme, respectively. For the AN beamforming, The optimal
power allocation ratio was derived to achieve the objective of
minimizing the SOP or maximizing the secrecy throughput.
It can be demonstrated from the numerical results that the
secrecy performance of the AN scheme is better than the
PMRT scheme when the transmit power is low or the number
of common paths is large, whereas the PMRT scheme shows
its superiority over the AN scheme.
APP EN DI X A
PROO F OF TH EO RE M 1
We have that the transmitter only emits signals when η >
ηmin =T−1
c. If T−1
c≥1, i.e., γ≤T−1
¯c, no feasible η∈
[0,1] satisfies η > ηmin and the transmission is suspended.
If T−1
c<1, the transmitter emits signals in the range of
η∈(ηmin,1]. Next, we derive the optimal value of ηthat
minimizes Pso(γ◦).
We first prove the convexity of K(η)on η∈(ηmin,1]. From
the expression of K(η), we have d2K(η)
dη2= 6η+ 2ε1>0, i.e.,
K(η)is a convex function of η. Then we determine the sign
of K(η). The values of K(η)at the boundaries η=ηmin
and η= 1 are K(ηmin) = η2
min(ηmin −1) −1
b(LE−LC)and
K(1) = b(LE−LC)[c−(T−1)]−(T−1)
bc(LE−LC). Obviously, K(ηmin)<0
always holds. Then we discuss the optimal value of ηfor the
following two cases.
1) When K(1) ≤0, since K(η)is convex in the range
η∈(ηmin,1],K(η)and dPso
dη are always negative. Hence
Pso monotonically decreases with η, and the minimum Pso is
achieved at η= 1,with the corresponding condition obtained
from K(1) ≤0, which is T−1
¯c< γ ≤(¯a+1)(T−1)
¯a¯c.
2) When K(1) >0, it means K(η)and dPso
dη become first
negative and then positive as ηincreases from ηmin to 1, i.e.,
Pso first decreases and then increases with η, and the optimal
value of ηis the unique root of the cubic equation K(η) = 0.
Solving this equation using Cardano’s formula yields η⋆.
Combining the above two cases completes the proof.
APP EN DI X B
PROO F OF LE MM A 1
With the definition of Pso in (5), we rewritten (23b) as
Pso(γ◦) = 1 −FAN
ξE(2Rt−Rs−1) ≤ϵ. (34)
Since FAN
ξE(x)is a monotonically increasing function, (34)
is equivalent to 2Rt−Rs−1≥F−1
ξE(1 −ϵ). Defining κ(η),
F−1
ξE(1−ϵ)
η, constraint in (23b) can be expressed as
Rs≤Rt−log(1 + ηκ(η)).(35)
From (23a), in order to obtain a higher Rs, we set Rtequal
to its maximum value, which is CAN
D= log(1+ηc). According
to (35), the maximum achievable value of Rscan be given by
Rs(γ◦) = [log(1 + ηc)−log(1 + ηκ(η))]+.(36)
To achieve a positive secrecy rate, γ◦needs to satisfy the
constraint c > κ(η), which means that the transmitter does
not start communication unless this constraint is satisfied.
Therefore, the optimization problem in (23) can be rewritten
as (24).
0090-6778 (c) 2016 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/TCOMM.2017.2672661, IEEE
Transactions on Communications
12
APP EN DI X C
PROO F OF LE MM A 2
Exploiting the derivative rule for implicit functions, we
obtain the first-order derivative of κ(η)
dκ
dη =−∂Q1/∂η
∂Q1/∂κ =ab(LE−LC)κ
b(1 −η)κ+ab(LE−LC)(1 −η) + a.
(37)
Since a > 0,b > 0,c > 0,LE−LC>0and 1−η≥0,
we have dκ(η)
dη >0. Then the second-order derivative of κ(η)
can be given by
d2κ
dη2=1
κdκ
dη 2
+
ab2(LE−LC)κκ+a(LE−LC)−(1 −η)dκ
dη
[b(1 −η)κ+ab(LE−LC)(1 −η) + a]2.
(38)
Substituting (37) into (38), we obtain
κ+a(LE−LC)−(1 −η)dκ
dη =a(LE−LC)
+b(1 −η)κ2+aκ
b(1 −η)κ+ab(LE−LC)(1 −η) + a>0.
(39)
Thus we can derive that
d2κ
dη2>1
κdκ
dη 2
>0.(40)
With dκ
dη >0and d2κ
dη2>0, we complete the proof.
APP EN DI X D
PROO F OF TH EO RE M 2
The first-order and second-order derivatives of Rswith
respect to ηcan be described respectively as
dRs
dη =1
ln 2 c
1 + ηc −κ+ηdκ
dη
1 + ηκ ,(41)
d2Rs
dη2=−1
ln 2 c2
(1 + ηc)2+1
(1 + ηκ)2
×(1 + ηκ)2dκ
dη +ηd2κ
dη2−κ+ηdκ
dη 2
(b)
<−1
ln 2 c2
(1 + ηc)2−κ2
(1 + ηκ)2
=−1
ln 2 c
1 + ηc +κ
1 + ηκ c−κ
(1 + ηc)(1 + ηκ)
(c)
<0,
(42)
where dκ
dη and d2κ
dη2are given by (37) and (38) in Appendix C,
(b)is derived by utilizing (40), (c)holds for c > κ. Thus we
have proven that Rsis a concave function.
Now we investigate the optimal power allocation ratio η∗
that maximizes Rs. From the first-order derivative formula of
Rsin (41), we obtain
dRs
dη |η=0 =1
ln 2 (c−κ)>0,(43)
dRs
dη |η=1 =1
ln 2 c
1 + c+[1 + (LE−LC)b]aln ϵ
1−aln ϵ.
(44)
Rsmonotonically increases through η= 0 according to
(43). Due to the concave property of Rs, when dRs
dη |η=1 >0,
the maximum Rsis achieved at η= 1; otherwise, the maxi-
mum Rsis derived at the unique η⋆which is the zero-crossing
point of dRs
dη . Therefore, the optimal η∗that maximizes Rscan
be given in (26).
APP EN DI X E
PROO F OF CO ROLLARY 1
Substituting (37) into dRs
dη = 0, we can derive that
b(1 −η)κ2+ [abc(LE−LC)η2+ab(LE−LC)
+a−bc(1 −η)]κ−abc(LE−LC)(1 −η)−ac = 0.
(45)
Denote the left side of the equation as Q2, we can obtain
the following formulas
∂Q2
∂η =z1
dκ
dη +z2>0,(46)
∂Q2
∂a =bc(LE−LC)η2κ+b(LE−LC)κ+κ
−bc(LE−LC)(1 −η)−c
(d)
=b
a(1 −η)κ(c−κ)>0,(47)
∂Q2
∂b = (1 −η)κ2+ac(LE−LC)η2κ−c(1 −η)κ
+a(LE−LC)κ−ac(LE−LC)(1 −η)
(e)
=a
b(c−κ)>0,(48)
∂Q2
∂c =ab(LE−LC)η2κ−b(1 −η)κ
−ab(LE−LC)(1 −η)−a
(f)
=1
c[−b(1 −η)κ2−ab(LE−LC)κ−aκ]<0,(49)
where z1= [2b(1−η)κ+abc(LE−LC)η2−bc(1−η)+ab(LE−
LC)+a],z2=−bκ2+bcκ+2abc(LE−LC)ηκ+abc(LE−LC)
and dκ
dη >0. Substituting (45) into z1, we can obtain that
z1=b(1 −η)κ+1
κ[abc(LE−LC)(1 −η) + ac]>0.z2>0
holds for c > κ.(d),(e)and (f)are derived from substituting
(45).
1) γP: Exploiting derivative rule for implicit function-
s with the equation Q2= 0.dη
dγP=−∂Q2/∂γP
∂Q2/∂η =
−z1dκ
da
da
dγP+∂Q2
∂a
da
dγP+∂Q2
∂c
dc
dγP
∂Q2/∂η . Obviously, da
dγP<0,dc
dγP>0,
and we have dκ
da >0from (25). Thus we obtain dη
dγP>0
which implies the optimal η⋆monotonically increases with
respect to γP.
2) βD:dη
dβD=−∂Q2/∂βD
∂Q2/∂η =−
∂Q2
∂c
dc
dβD
∂Q2/∂η , where dc
dβD>
0. Thus we obtain dη
dβD>0which implies the optimal η⋆
monotonically increases with respect to βD.
3) βE:dη
dβE=−∂Q2/∂βE
∂Q2/∂η =−z1dκ
dβE+∂Q2
∂a
da
dβE+∂Q2
∂b
db
dβE
∂Q2/∂η ,
where da
dβE>0,db
dβE>0and we have dκ
dβE>0from
0090-6778 (c) 2016 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/TCOMM.2017.2672661, IEEE
Transactions on Communications
13
(25). Thus we obtain dη
dβE<0which implies the optimal
η⋆monotonically decreases with respect to βE.
4) ϵ:dη
dϵ =−∂Q2/∂ϵ
∂Q2/∂η =−z1dκ
dϵ
∂Q2/∂ η , and we have dκ
dϵ <0
from (25). Thus we obtain dη
dϵ >0which implies the optimal
η⋆monotonically increases with respect to ϵ.
APP EN DI X F
PROO F OF CO ROLLARY 2
By denoting ς=−γC+−¯aln ϵ
¯c√γCand substituting (29)
into (28), we derive the maximum secrecy throughput of MRT
beamforming
τ∗=∞
0∞
ς
R∗
s(γ◦)fγP(x)fγC(y)dxdy
(g)
=∞
0∞
0
1
Γ(LD−LC)log (¯
x+y+ς)[¯
c(¯
x+y+ς) + 1]
¯x+ς+ (1 −¯aln ϵ)y
×(¯x+ς)LD−LC−1e−(¯x+ς)fγC(y)d¯xdy
=∞
0
e−ςfγC(y)
Γ(LD−LC)
LD−LC−1
m=0 LD−LC−1
mςLD−LC−1−m
∞
0log(1 + 1
y+ς¯x) + log(1 + ¯c
¯cy + ¯cς + 1 ¯x)
−log(1 + 1
(1 −¯aln ϵ)y+ς¯x) + log (y+ς)(¯cy + ¯cς + 1)
(1 −¯aln ϵ)y+ς
ׯxme−¯xd¯xdy,
(50)
where (g)holds for the transformation ¯x=x−ς. According
to [38, 4.337.5], the final result shown in (30) is obtained.
REF ER EN CE S
[1] A. D. Wyner, “The wire-tap channel,” Bell Syst. Tech. J., vol. 54, no. 8,
pp. 1355-1387, Oct. 1975.
[2] Y.-W. P. Hong, P.-C. Lan, and C.-C. J. Kuo, “Enhancing physical-layer
secrecy in multiantenna wireless systems: An overview of signal processing
approaches,” IEEE Commun. Mag., vol. 30, no. 5, pp. 29-40, Aug. 2013.
[3] A. Khisti, and G. W. Wornell, “Secure transmission with multiple
antennas I: The MISOME wiretap channel,” IEEE Trans. Inf. Theory, vol.
56, no. 7, pp. 3088-3104, Jul. 2010.
[4] A. Khisti and G. Wornell, “Secure transmission with multiple antennas
II: The MIMOME wiretap channel,” IEEE Trans. Inf. Theory, vol. 56, no.
11, pp. 5515C5532, Nov. 2010.
[5] S. Shafiee, N. Liu, and S. Ulukus, “Towards the secrecy capacity of the
Gaussian MIMO wire-tap channel: The 2-2-1 Channel,” IEEE Trans. Inf.
Theory, vol. 55, no. 9, pp. 4033-4039, Sep. 2009.
[6] H.-M. Wang, T.-X. Zheng, and X.-G. Xia, “Secure MISO wiretap chan-
nels with multi-antenna passive eavesdropper: artificial noise vs. artificial
fast fading,” IEEE Trans. Wireless Commun., vol. 14, no. 1, pp. 94-106,
Jan. 2015.
[7] H.-M. Wang, C. Wang, D. W. K. Ng, M. H. Lee, and J. Xiao, “Artificial
noise assisted secure transmission for distributed antenna systems,” IEEE
Trans. Signal Process., vol. 64. no. 15, pp. 4050-4064, Aug. 2016.
[8] S. Goel, and R. Negi, “Guaranteeing secrecy using artificial noise,” IEEE
Trans. Wireless Commun., vol. 7, no. 6, pp. 2180-2189, Jun. 2008.
[9] X. Zhou, and M. R. McKay, “Secure transmission with artificial noise
over fading channels: Achievable rate and optimal power allocation,” IEEE
Trans. Veh. Technol., vol. 59, no. 8, pp. 3831-3842, Oct. 2010.
[10] Q. Li and W.-K. Ma, “Spatially selective artificial-noise aided trans-
mit optimization for MISO multi-eves secrecy rate maximization,” IEEE
Trans.Signal Process., vol. 61, no. 10, pp. 2704-2717, May 2013.
[11] J. Xiong, K.-K. Wong, D. Ma, and J. Wei, “A closed-form power
allocation for minimizing secrecy outage probability for MISO wiretap
channels via masked beamforming,” IEEE Commum. Lett., vol. 16, no. 9,
pp. 1496-1499, Sep. 2012.
[12] X. Zhang, X. Zhou, and M. R. McKay, “On the design of artificial-
noise-aided secure multi-antenna transmission in slow fading channels,”
IEEE Trans. Veh. Technol., vol. 62, no. 5, pp. 2170-2181, Jun. 2013.
[13] Z. Pi, and F. Khan, “An introduction to millimeter-wave mobile broad-
band systems,” IEEE Commun. Mag., vol. 49, no. 6, pp. 101-107, Jun.
2011.
[14] S. Rangan, T. S. Rappaport, and E. Erkip, “Millimeter-wave cellular
wireless networks: Potentials and challenges,” Proc. of the IEEE, vol. 102,
no. 3, pp. 366-385, Mar. 2014.
[15] J. Brady, N. Behdad, and A. M. Sayeed, “Beamspace MIMO for
millimeter-wave communications: System architecture, modeling, analysis,
and measurements,” IEEE Trans. Antennas Propag., vol. 61, no. 7, pp.
3814-3827, Jul. 2013.
[16] M. Park, and H. Pan, “A spatial diversity technique for IEEE 802.11ad
WLAN in 60 GHz band,” IEEE Commun. Lett., vol. 16, no. 8, pp. 1260-
1262, Aug. 2012.
[17] A. Kammoun, H. Khanfir, Z. Altman, M. Debbah, and M. Kamoun,
“Preliminary results on 3D channel modeling: from theory to standardiza-
tion,” IEEE J. Sel. Areas Commun., vol. 32, no. 6, pp. 1219-1229, Jun.
2014.
[18] M. R. Akdeniz, Y. Liu, M. K. Samimi, S. Sun, S. Rangan, T. S.
Rappaport, and E. Erkip, “Millimeter wave channel modeling and cellular
capacity evaluation,” IEEE J. Sel. Areas Commun., vol. 32, no. 6, pp.
1164-1179, Jun. 2014.
[19] A. Adhikary, E. Al Safadi, M. K. Samimi, R. Wang, G. Caire, T. S.
Rappaport, and A. F. Molisch, “Joint spatial division and multiplexing
for mm-wave channels,” IEEE J. Sel. Areas Commun., vol. 32, no. 6, pp.
1239-1255, Jun. 2014.
[20] A. Alkhateeb, O. El Ayach, G. Leus, and R. W. Heath, “Channel
estimation and hybrid precoding for millimeter wave cellular systems,”
IEEE J. Sel. Topics Signal Process., vol.8, no. 5, pp. 831-846, Oct. 2014.
[21] S. Hur, T. Kim, D. J. Love, J. V. Krogmeier, T. A. Thomas, and A.
Ghosh, “Millimeter wave beamforming for wireless backhaul and access
in small cell networks,” IEEE Trans. Commun., vol. 61, no. 10, pp. 4391-
4403, Oct. 2013.
[22] O. El Ayach, S. Rajagopal, S. Abu-Surra, Z. Pi, and R. W. Heath,
“Spatially sparse precoding in millimeter wave MIMO systems,” IEEE
Trans. Wireless Commun., vol. 13, no. 3, pp. 1499-1513, Mar. 2014.
[23] Y. Lee, C. Wang, and Y. Huang, “A hybrid RF/Baseband precoding
processor based on parallel-index-selection matrix-inversion-bypass simul-
taneous orthogonal matching pursuit for millimeter wave MIMO systems,”
IEEE Trans. Signal Process., vol. 63, no. 2, pp. 305-317, Jan. 2015.
[24] J. Singh, and S. Ramakrishna, “On the feasibility of beamforming in
millimeter wave communication systems with multiple antenna arrays,” in
Proc. IEEE GLOBECOM, Austin, TX, USA, Dec. 2014, pp. 3802-3808.
[25] A. L. Swindlehurst, E. Ayanoglu, P. Heydari, and F. Capolino,
“Millimeter-wave massive MIMO: The next wireless revolution?” IEEE
Commun. Mag., vol. 52, no. 9, pp. 56-62, Sep. 2014.
[26] T. L. Marzetta, “Noncooperative cellular wireless with unlimited num-
bers of base station antennas,” IEEE Trans. Wireless Commun., vol. 9, no.
11, pp. 3590-3600, Nov. 2010.
[27] J. Zhu, R. Schober, and V. K. Bhargava, “Secure transmission in
multicell massive MIMO systems,” IEEE Trans. Wireless Commun., vol.
13, no. 9, pp. 4766-4781, Sep. 2014.
[28] N. Yang, L. Wang, G. Geraci, M. Elkashlan, J. Yuan, and M. Di Renzo,
“Safeguarding 5G wireless communication networks using physical layer
security,” IEEE Commun. Mag., vol. 53, no. 4, pp. 20-27, Apr. 2015.
[29] N. Valliappan, A. Lozano, and R. W. Heath, “Antenna subset modu-
lation for secure millimeter-wave wireless communication,” IEEE Trans.
Commun., vol. 61, no. 8, pp. 3231-3245, Aug. 2013.
[30] L. Wang, M. Elkashlan, T. Q. Duong, and R. W. Heath, “Secure
communication in cellular networks: The benefits of millimeter wave
mobile broadband,” in Proc. IEEE SPAWC, Toronto, ON, Canada, Jun.
2014, pp. 115-119.
[31] C. Wang, and H.-M. Wang, “Physical layer security in millimeter wave
cellular networks,” IEEE Trans. Wireless Commun., vol. 15, no. 8, pp.
5569-5585, Aug. 2016.
[32] Y. Ju, H.-M. Wang, T.-X. Zheng, and Q. Yin, “Secure transmission with
artificial noise in millimeter wave systems,” in Proc. IEEE WCNC, Doha,
Qatar, Apr. 2016, pp. 1-6.
[33] Y. Ju, H.-M. Wang, T.-X. Zheng, Y. Zhang, Q. Yang, and Q. Yin,
“Secrecy throughput maximization for millimeter wave systems with
artificial noise,” in Proc. IEEE PIMRC, Valencia, Spain, Sep. 2016, pp.
1-6.
0090-6778 (c) 2016 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/TCOMM.2017.2672661, IEEE
Transactions on Communications
14
[34] A. Ghosh, T. A. Thomas, M. C. Cudak, and R. Ratasuk, “Millimeter-
wave enhanced local area systems: A high-data-rate approach for future
wireless networks,” IEEE J. Sel. Areas Commun., vol. 32, no. 6, pp. 1152-
1163, Jun. 2014.
[35] D. Tse, and P. Viswanath, Fundamentals of Wireless Communication.
Cambridge, U. K.: Cambridge Univ. Press, 2005.
[36] A. M. Sayeed, “Deconstructing multiantenna fading channels,” IEEE
Trans. Signal Process., vol. 50, no. 10, pp. 2563-2579, Oct. 2002.
[37] H. Yin, D. Gesbert, M. Filippou, and Y. Liu, “A coordinated approach to
channel estimation in large-scale multiple-antenna systems,” IEEE J. Sel.
Areas Commun., vol. 31, no. 2, pp. 264-273, Feb. 2013.
[38] I. S. Gradshteyn, I. M. Ryzhik, A. Jeffrey, D. Zwillinger, and S.
Technica, Table of Integrals, Series, and Products, 7th ed. New York:
Academic Press, 2007.
Ying Ju received the B.S. and M.S. degrees in
the School of Electronic Information Engineering
from Tianjin University, Tianjin, China, in 2008
and 2010, respectively. She is currently pursuing
the Ph.D. degree in the School of Electronic and
Information Engineering at Xi’an Jiaotong Univer-
sity, Xi’an, China. She is also an engineer with the
Department of Shaanxi Monitoring Station, State
Radio Monitoring Center, Xi’an, China. Her research
interests include physical layer security of wireless
communications, millimeter wave communication
systems, and stochastic geometry.
Hui-Ming Wang (S’07-M’10-SM’16) received the
B.S. and Ph.D. degrees, both with first class honors
in Electrical Engineering from Xi’an Jiaotong Uni-
versity, Xi’an, China, in 2004 and 2010, respectively.
He is currently a Full Professor with the Department
of Information and Communications Engineering,
Xi’an Jiaotong University, and also with the Ministry
of Education Key Lab for Intelligent Networks and
Network Security, China. From 2007 to 2008, and
2009 to 2010, he was a Visiting Scholar at the
Department of Electrical and Computer Engineering,
University of Delaware, USA. His research interests include cooperative
communication systems, physical-layer security of wireless communications,
MIMO and space-timing coding.
Dr. Wang received the National Excellent Doctoral Dissertation Award in
China in 2012, a Best Paper Award of International Conference on Wireless
Communications and Signal Processing, 2011, and a Best Paper Award of
IEEE/CIC International Conference on Communications in China, 2014. He
is currently an Associate Editor for the IEEE ACCESS. He also served as a
Symposium Chair of Wireless Communications and Networking in ChinaCom
2015, a Technical Program Committee Chair of the Workshop on physical
layer security in IEEE Globecom 2016, and TPC members of various IEEE
sponsored conferences, including the IEEE Globecom, ICC, WCNC, VTC,
and PIMRC, etc. He is a co-author of the book “Physical Layer Security in
Random Cellular Networks” published by Springer in 2016.
Tong-Xing Zheng (S’14-M’16) received the B.S.
and Ph.D. degrees in the School of Electronic and
Information Engineering from Xi’an Jiaotong Uni-
versity, Xi’an, China, in 2010 and 2016, respectively.
He is currently a Lecturer with the Department
of Information and Communications Engineering,
Xi’an Jiaotong University, and also with the Ministry
of Education Key Lab for Intelligent Networks and
Network Security, China.
He has co-authored the book Physical Layer Secu-
rity in Random Cellular Networks (Springer, 2016).
His current research interests include the physical-layer security of wireless
communications, heterogeneous cellular networks, full-duplex communica-
tions, cooperative communication systems, and stochastic geometry.
0090-6778 (c) 2016 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/TCOMM.2017.2672661, IEEE
Transactions on Communications
15
Qinye Yin received the B.S., M.S., and Ph.D. de-
grees in communication and electronic systems from
Xi’an Jiaotong University, Xi’an, China, in 1982,
1985, and 1989, respectively.
He is currently a Professor of Information En-
gineering Institute, Xi’an Jiaotong University. His
research interests focus on the joint time-frequency
analysis and synthesis, the theory and applications of
wireless sensor networks, multiple antenna MIMO
broadband communication systems (including smart
antenna systems), parameter estimation, and array
signal processing.