See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/343516533

Robust Beamforming Designs in Secure MIMO SWIPT IoT Networks with a

Nonlinear Channel Model

Article  in  IEEE Internet of Things Journal · August 2020

DOI: 10.1109/JIOT.2020.3014933

CITATIONS READS
17 97

5 authors, including:

Zhengyu Zhu Ning Wang

Zhengzhou University University of British Columbia - Vancouver
87 PUBLICATIONS   883 CITATIONS    70 PUBLICATIONS   950 CITATIONS   

SEE PROFILE SEE PROFILE

Wanming Hao Inkyu Lee

Kyushu University Korea University
85 PUBLICATIONS   885 CITATIONS    349 PUBLICATIONS   6,245 CITATIONS   

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Zhengyu Zhu on 07 July 2021.

The user has requested enhancement of the downloaded file.

1702
Robust Beamforming Designs in Secure MIMO
SWIPT IoT Networks With a Nonlinear
Channel Model
Zhengyu Zhu, Member, IEEE, Ning Wang , Member, IEEE, Wanming Hao, Zhongyong Wang ,
and Inkyu Lee , Fellow, IEEE
Abstract—In this article, we study a robust beamform-
ing design for multiuser multiple-input–multiple-output secrecy
networks with simultaneous wireless information and power
transfer (SWIPT). In this system, an access point, multiple
Internet-of-Things (IoT) devices under the nonlinear energy har-
vesting (EH) model with a help of one cooperative jammer (CJ).
We employ artificial noise (AN) generation to facilitate efficient
wireless energy transfer and secure transmission. To achieve EH
fairness, we aim to maximize the minimum harvested energy
among users subject to secrecy rate constraint and total transmit
power constraint in the presence of channel estimation errors.
By incorporating a norm-bounded channel uncertainty model,
the original robust problem is transformed into a two-layer
optimization problem, where the inner layer problem is refor-
mulated as semidefinite programming (SDP) and the outer layer
problem is solved by a one-dimensional (1-D) line search algo-
rithm. In addition, in order to reduce computational complexity,
we propose an algorithm based on sequential parametric convex
approximation (SPCA). Finally, simulation results show that the
proposed SPCA method achieves the same performance as the
two-layer algorithm with much lower complexity.
Index Terms—Internet of Things, nonlinear energy harvest-
ing, physical-layer security, robust beamforming, simultaneous
wireless information and power transfer.
Manuscript received March 5, 2020; revised May 27, 2020 and July 8,
2020; accepted July 29, 2020. Date of publication August 7, 2020; date
of current version January 22, 2021. This work was supported in part by
the National Natural Science Foundation of China under Grant 61801434,
Grant 61801435, Grant 61771431, and Grant U1736107; in part by the
Scientific and Technological Key Project of Henan Province under Grant
192102310178; in part by the Science and Technology Innovation Project
of Zhengzhou under Grant 2019CXZX0037; in part by the National Key
Research and Development Program of China under Grant 2019QY0302 and
Grant 2019YFB1803200; and in part by the National Research Foundation
through the Ministry of Science, ICT, and Future Planning (MSIP), Korean
Government under Grant 2017R1A2B3012316. (Corresponding author:
Inkyu Lee.)
Zhengyu Zhu and Zhongyong Wang are with the School of Information
Engineering, Zhengzhou University, Zhengzhou 450001, China (e-mail:
iezyzhu@zzu.edu.cn; iezywang@zzu.edu.cn).
Ning Wang is with the School of Information Engineering, Zhengzhou
University, Zhengzhou 450001, China, and also with the Department of
Electrical and Computer Engineering, McMaster University, Hamilton, ON
L8S 4K1, Canada (e-mail: ienwang@zzu.edu.cn).
Wanming Hao is with the Henan Institute of Advanced Technology and
the School of Information Engineering, Zhengzhou University, Zhengzhou
450001, China (e-mail: iewmhao@zzu.edu.cn).
Inkyu Lee is with the School of Electrical Engineering, Korea University,
Seoul 136-713, South Korea (e-mail: inkyu@korea.ac.kr).
Digital Object Identifier 10.1109/JIOT.2020.3014933
2327-4662 
c 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: Zhengzhou University. Downloaded on February 05,2021 at 02:31:35 UTC from IEEE Xplore. Restrictions apply.
IEEE INTERNET OF THINGS JOURNAL, VOL. 8, NO. 3, FEBRUARY 1, 2021
I. I NTRODUCTION
HE FIFTH-GENERATION (5G) wireless systems are
T expected to meet a continuously increasing demand for
wireless applications, such as wide radio coverage and high
data rate [1]. Due to the advancement of wireless sensor
networks (WSNs) and Internet of Things (IoT), the energy-
constrained 5G network needs support millions of intelligent
terminals [2], [3]. Since no wire connection is normally
assumed for IoT devices, an energy-constrained environment
becomes an important issue for IoT systems [4]. Energy har-
vesting (EH) techniques have been introduced as an efficient
solution that allows IoT devices to capture sufficient and con-
tinuous energy [5], [6]. Since the energy from the nature,
such as solar power and wind energy cannot be provided
reliably, the radio-frequency (RF) signal can be utilized as
an alternative to carry the information and energy simultane-
ously [7]–[9]. One main research approach in wireless power
transfer (WPT) is the simultaneous wireless information and
power transfer (SWIPT) technique [10], which has attracted a
lot of interests for providing power supplies for wireless IoT
networks [11]–[16].
On the other hand, secrecy transmission has extracted
attentions in communication systems [17], [18]. Specifically,
physical-layer security (PLS) has been recognized as an impor-
tant issue for SWIPT systems since the wireless information
is more vulnerable to eavesdropping [19]–[22]. In a two-
way cognitive radio IoT network aided with SWIPT, a secure
beamforming design algorithm has been proposed by max-
imizing the secrecy sum rate [21]. Also, in the presence
of multiple passive eavesdroppers, the security and energy
coverage performance framework for the downlink millimeter-
wave (mmWave) SWIPT unmanned-aerial-vehicles (UAVs)
networks was developed [22].
Another aspect that we address in this article is the impact
of imperfect channel state information (CSI). In practice, it
is not always possible to obtain perfect CSI at the trans-
mitter due to channel estimation errors [23], [24]. Secure
communication with SWIPT would be more challenging in
the presence of the imperfect CSI [25]–[33]. Assuming sta-
tistical channel uncertainty models, a robust secure beam-
forming design has been presented in secure MISO SWIPT
systems [25]. In addition, some robust optimization tech-
niques have been constructed for the secrecy SWIPT systems
under imperfect channel estimation in [34]–[36]. Also, a
ZHU et al.: ROBUST BEAMFORMING DESIGNS IN SECURE MIMO SWIPT IoT NETWORKS WITH NONLINEAR CHANNEL MODEL
robust beamforming design with low complexity was intro-
duced to efficiently strike the balance between the information
decoding (ID) and EH [34]. Considering the nonorthogonal
multiple access (NOMA) SWIPT scheme with CSI errors, the
resource allocation designs were examined from the aspects
of the weighted sum-rate maximization and the total power
consumption minimization [35].
In addition, artificial noise (AN) methods have been used
by embedding the transmit beamforming to confuse eaves-
droppers [17]. In secrecy SWIPT systems, the AN plays the
roles of both carrying an energy signal for WPT and pro-
tecting the secrecy information transmission [26], [27], [37].
In [27], an EH scheme for an AN-aided multiuser multiple-
input–multiple-output (MU-MIMO) secrecy transmission was
investigated. Considering heterogeneous networks supported
by SWIPT, an AN-based coordinated beamforming design was
proposed by applying the successive convex approximation
(SCA) technique [37].
Also, to further increase the secrecy performance, a jam-
ming node has been introduced in the secrecy networks, which
prevents the eavesdroppers from intercepting the intended mes-
sages [19], [20], [28]–[30]. In a downlink wiretap channel
SWIPT system, a cooperative jammer (CJ) has been illustrated
as an effective means to provide secret wireless communica-
tions [19]. In the downlink multiuser orthogonal frequency-
division multiplexing (OFDM) SWIPT system, the secrecy
performance with jamming from the SWIPT devices has been
analyzed in the presence of multiple eavesdroppers [20].
When information receivers (IRs) and EH receivers (ERs)
are placed in a same cell, the ERs are normally assumed to
be closer to the transmitter compared with IRs due to its low-
power sensitivity level. This gives rise to a new information
security issue in the SWIPT systems. In such a situation, the
ERs have a possibility of eavesdropping the information sent
to the IRs, and thus can become potential eavesdroppers [27].
In this article, we investigate a secure transmission design
problem in an MU-MIMO secrecy IoT system with SWIPT. In
this system, a multiantenna transmitter supports single-antenna
co-located receiver (CR) IoT nodes, and multiantenna EH IoT
nodes in the presence of a multiantenna CJ. The CR node
employs a power splitting (PS) scheme to split the received
signals into the ID and the EH part. Unlike the work in [20]
and [35], our objective is to maximize the minimum harvested
energy of both EH nodes and CR nodes subject to secrecy
rate constraints and the total transmit power constraint by
incorporating the norm-bounded channel uncertainty model.
Moreover, we seek a transmission strategy to jointly design
the AN-aided beamforming matrix, the AN covariance matrix,
the jamming covariance matrix, and the PS ratio. To the best
of our knowledge, there are no prior works that investigated
the joint designs for the AN, CJ, and PS ratio.
We summarize the main contributions of this article as
follows.
1) Due to its difficulty to directly solve the original
optimization problem, we propose a method which
decouples the original one into a two-layer problem. To
remove the impact of channel uncertainty, semi-infinite
constraints are converted into linear matrix inequalities
Authorized licensed use limited to: Zhengzhou University. Downloaded on February 05,2021 at 02:31:35 UTC from IEEE Xplore. Restrictions apply.
1703
Fig. 1. System model of secure SWIPT IoT systems with cooperative
jamming for MU-MIMO downlink.
(LMIs). As a result, the inner layer of the worst case
robust problem is recast to a semidefinite programming
(SDP). In addition, the outer layer becomes a single
variable optimization problem, where a simple 1-D line
search algorithm is applied. Furthermore, the tightness
of the SDP relaxation is verified.
2) To reduce the computational complexity of the two-
layer method, an iterative algorithm based on sequen-
tial parametric convex approximation (SPCA) is also
addressed, where the original problem is transformed to
a second-order cone programming (SOCP) problem that
can be directly solved by CVX. We demonstrate that
the proposed SPCA scheme achieves almost the same
performance as the two-layer method.
The remainder of this article is organized as follows.
Section II presents the system model. Section III provides the
problem formulation and robust design methods. We compare
the computational complexities of the robust design methods in
Section IV. Section V illustrates the simulation results. Finally,
we conclude this article in Section VI.
Notation: Vectors and matrices are denoted by bold lower-
case and uppercase letters, respectively. (·)T and (·)H represent
transpose and Hermitian transpose, respectively. Also, ⊗
defines the Kronecker product and [x]+ stands for max{x, 0}.
For a vector x, x indicates the Euclidean norm and diag(x)
denotes a diagonal matrix whose diagonal element vector is x.
CM×L and HM×L describe the space of M × L complex matri-
ces and Hermitian matrices, respectively. H+ equals the set of
positive semidefinite Hermitian matrices, and R+ denotes the
set of all nonnegative real numbers. For a matrix A, A  0
means that A is positive semidefinite, and AF , tr(A), |A|,
and rank(A) stand for the Frobenius norm, trace, determinant,
and rank, respectively. vec(A) stacks the elements of A in
a column vector. 0M×L is a null matrix of size M × L. E{·}
describes the expectation, and {·} equals the real part of a
complex number.
II. S YSTEM M ODEL
A. Network Model
We consider a secure downlink MU-MIMO SWIPT-based
IoT system as shown in Fig. 1, which consists of one
1704 IEEE INTERNET OF THINGS JOURNAL, VOL. 8, NO. 3, FEBRUARY 1, 2021

transmitter, one CJ, K CR nodes, and L EH nodes. Here, the where np,k ∼ CN (0, σp,k
2 ) is the noise at the kth CR node [10].

CR node employs the PS scheme to decode information and Let us define Fk = fk fH

k as the transmit covariance matrix,
harvest power simultaneously. It is assumed that the transmit- Z = zzH as the AN covariance matrix, and Q = qqH as the
ter, CJ, and the EH node are equipped with NT , NJ , and NE CJ covariance matrix. Hence, the achieved secrecy rate can be
antennas, respectively, while CR nodes have a single antenna. calculated by
We denote hs,k ∈ CNT as the channel vector between the trans-
 
mitter and the kth CR node, He,l ∈ CNT ×NE as the channel     +
matrix between the transmitter and the lth EH node, gs,k ∈ CNJ R̂s,k = Cs,k ρs,k , Fk , Z, Q − max Cl,k ρs,k , Fk , Z, Q (4)
l
as the channel vector between CJ and the kth CR node, and
Ge,l ∈ CNJ ×NE as the channel matrix between CJ and the lth where Cs,k and Cl,k are defined in (5), shown at the bottom
EH node, respectively. of the page.

B. Transmission Protocol C. Nonlinear EH Model

To achieve reliable transmission, the transmitter adopts
transmit beamforming with AN, which acts as interference
to the EH nodes and provides energy to the CR nodes and EH
nodes. Then, the transmitter sends the message xk to the kth
CR node as
xk = fk sk + z
where fk ∈ CNT indicates the linear beamforming vector for
the kth CR node at the transmitter, sk ∈ C represents the
information-bearing signal intended for the kth CR node with
E{|sk |2 } = 1, and z ∈ CNT is the energy-carrying AN vector,
which can also be composed by multiple energy beams.
The received signal at the kth CR node and the lth EH node
can be expressed as

K 1+exp(ai bi )
ys,k = hH
s,k fk sk + hH
s,k z + gs,k qsJ + ns,k , k = 1, . . . , K
H
As mentioned previously, most existing works on the
SWIPT system adopted the linear EH model, in which the
harvested energy is assumed to linearly increase with respect
to the input received power. Thus, the harvested energy is
given by  = ηPin , where Pin represents the input power
at the receiver, and η ∈ [0, 1] is the energy conversion effi-
(1)
ciency [25]–[33]. Recently, it was shown in [38]–[41] that
the practical EH circuits exhibit nonlinear EH behavior due
to the saturation of the rectifier, which consists of nonlin-
ear components, such as diodes and capacitors. To avoid the
mismatch caused by the linear EH model, we employ the
logistic-function-based nonlinear EH model [38], as
Mi Mi
1+exp(−ai (Pin −bi )) − 1+exp(ai bi )
 prac
(Pin ) = (6)
1− 1

k=1
where Mi is a constant expressing the maximum harvested
power at the ith CR/ER node when EH circuit reaches satu-
K
ye,l = HH fk sk + HH ration. Parameters ai and bi are also constants in connection
e,l z + Ge,l qsJ + ne,l , l = 1, . . . , L
H
e,l
k=1 with the detailed circuit specifications, as defined in [38]–[41].
(2) Meanwhile, as for the kth CR node, the (1−ρs,k ) part of the
received signal is utilized for EH. Thus, the received RF power
where q ∈ CNJ indicates the CJ vector, sJ equals the coop- at the kth CR node and the lth EH node can be written as
erative jamming signal introduced by CJ with E{|sJ |2 } = 1,   K  
and ns,k ∼ CN (0, σs,k 2 ) and n
e,l ∼ CN (0, σe,l I) stand for the
2   H 
additive Gaussian noise at the kth CR node and the lth EH Es,k = 1 − ρs,k hs,k Fk + Z hs,k + gs,k Qgs,k + σs,k
H 2

k=1
node, respectively.   K  
The received signal at the kth CR node is divided into the ID   
Ee,l = tr HH e,l QGe,l + NE σe,l .
Fk + Z He,l + tr GH 2
and EH parts by the PS ratio ρs,k ∈ (0, 1]. Thus, the received e,l
k=1
signal for ID at the kth CR node can be given as
(7)
√
s,k =
yID ρs,k ys,k + np,k
  Using the nonlinear EH model, the harvested power at the kth
√ K
= ρs,k hH fk sk + hH CR node and the lth EH node is expressed as
s,k z + gs,k qsJ + ns,k + np,k
H
s,k
k=1 prac   prac  
(3) Es,k = prac Es,k ∀k, Ee,l = prac Ee,l ∀l. (8)

⎛ ⎞
  ρs,k hH
s,k Fk hs,k
Cs,k ρs,k , Fk , Z, Q = log2 ⎝1 +   ⎠
ρs,k σs,k
2 + hH Wh
j=k s,k j s,k + hH Zh
s,k s,k + g H Qg
s,k s,k + σ 2
p,k
  −1 
   
Cl,k ρs,k , Fk , Z, Q = log2 I + HHe,l ZHe,l + Ge,l QGe,l + σe,l I
H 2
HH 
e,l Fk He,l  (5)

Authorized licensed use limited to: Zhengzhou University. Downloaded on February 05,2021 at 02:31:35 UTC from IEEE Xplore. Restrictions apply.
ZHU et al.: ROBUST BEAMFORMING DESIGNS IN SECURE MIMO SWIPT IoT NETWORKS WITH NONLINEAR CHANNEL MODEL 1705

III. P ROBLEM F ORMULATION AND ROBUST B. Two-Layer Optimization Approach

D ESIGN M ETHODS In this section, we consider a two-layer approach to solve
In this article, our goal is to jointly optimize the worst problem (10). The outer layer problem is a one-variable
case formulation assuming imperfect CSI. Adopting the norm- optimization problem, in which the optimal solution can be
bounded channel uncertainty model [27], the actual channels obtained via 1-D search, i.e., golden search method, while
hs,k , He,l , gs,k , and Ge,l can be given as the inner layer can be reformulated into an SDP. In order to
address problem (10), we introduce slack variables Ēs , Ēe , and
hs,k = h̄s,k + es,k ∀k, He,l = H̄e,l + Ee,l ∀l t as
gs,k = ḡs,k + ẽs,k ∀k, Ge,l = Ḡe,l + Ẽe,l ∀l (9)
max Ēs + Ēe (11a)
where h̄s,k , ḡs,k , H̄e,l , and Ḡe,l denote the estimated channel ρs,k ,{Fk },Z,Q,Ēs ,Ēe ,t
prac
at the transmitter and CJ, and es,k , ẽs,k , Ee,l , and Ẽe,l are the s.t. min min Es,k ≥ Ēs ∀k (11b)
es,k ≤εs,k ,ẽs,k ≤ε̃s,k k
channel errors bounded as es,k 2 ≤ εs,k , ẽs,k 2 ≤ ε̃s,k , prac
Ee,l F ≤ εe,l , and Ẽe,l F ≤ ε̃e,l . min
  min Ee,l ≥ Ēe ∀l (11c)
Ee,l F ≤εe,l ,Ẽe,l  ≤ε̃e,l l
F

A. Problem Formulation min Cs,k + log t ≥ R̄s ∀k (11d)

es,k ≤εs,k ,ẽs,k ≤ε̃s,k
We aim to maximize the minimum harvested power among
1
all CR nodes and EH nodes subject to the secrecy rate con- max
  max Cl,k ≤ log ∀l
straint, the total transmit power constraint and the CJ power Ee,l F ≤εe,l ,Ẽe,l  ≤ε̃e,l l
F
t
constraint. By taking the above channel model into account, (10c) and (10d). (11e)
the robust max–min problem can be rewritten as (10), shown
at the bottom of the page, where R̄s stands for a given secrecy As problem (11) is intractable due to (11b)–(11e), the two-
rate threshold, and PT and PJ denote the available power bud- layer reformulation is considered. First, in the inner loop,
get at the transmitter and CJ, respectively. Problem (10) is we solve problem (11) for a given t by introducing another
nonconvex due to the secrecy rate constraint and the objective auxiliary variable χs,k and ωe,l , which is relaxed as (12),
function, and thus cannot be solved directly. shown at the bottom of the page, where gs,k = (ḡs,k +

⎛ ⎞

max ⎝ min
prac
Es,k + min  Ee,l ⎠
prac
(10a)
ρs,k ,{Fk },Z,Q es,k ≤εs,k ,ẽs,k ≤ε̃s,k E ≤ε ,

Ẽ

 e,l F e,l  e,l F e,l
≤ε̃

s.t. min Cs,k − max   Cl,k ≥ R̄s ∀k (10b)

es,k ≤εs,k ,ẽs,k ≤ε̃s,k Ee,l F ≤εe,l ,Ẽe,l  ≤ε̃e,l
F


K
tr(Fk ) + tr(Z) ≤ PT , tr(Q) ≤ PJ (10c)
k=1
1 ≥ ρs,k > 0, Fk  0, Z  0, Q  0, rank(Fk ) = 1 (10d)

f (t) = max Ēs + Ēe (12a)

ρs,k ,{Fk },Z,Q,Ēs ,Ēe ,t,{χs,k },{ωe,l }
  K  
  H 
s.t. 1 − ρs,k hs,k Fk + Z hs,k + gs,k + σs,k
2
≥ χs,k ∀k (12b)
k=1
  K  
 
tr HH
e,l Fk + Z He,l + tr Ĝe,l + NE σe,l
2
≥ ωe,l ∀l (12c)
k=1
⎛ ⎛ ⎞⎞
1  σp,k
2
hH
s,k
⎝ Fk − ⎝ Wj + Z⎠⎠hs,k − gs,k ≥ σs,k
2
+ ∀k (12d)
2R̄s
−1 ρs,k
t j=k
 
1
− 1 HH e,l ZHe,l + Ĝe,l + σ 2
e,l I  HH
e,l Fk He,l ∀l (12e)
t
Ms,k  
   ≥ Ēs 1 − s,k + Ms,k s,k ∀k (12f)
1 + exp −as,k χs,k − bs,k
Me,l  
   ≥ Ēe 1 − e,l + Me,l e,l ∀l
1 + exp −ae,l ωe,l − be,l
(10c) and (10d) (12g)

Authorized licensed use limited to: Zhengzhou University. Downloaded on February 05,2021 at 02:31:35 UTC from IEEE Xplore. Restrictions apply.
1706 IEEE INTERNET OF THINGS JOURNAL, VOL. 8, NO. 3, FEBRUARY 1, 2021

ẽs,k )H Q(ḡs,k + ẽs,k ), Ĝe,l = (Ḡe,l + Ẽe,l )H Q(Ḡe,l + Ẽe,l ), where ts,k ≥ 0 is a slack variable and Js,k = h̄H
s,k Bh̄s,k + gs,k +
s,k = 1/(1 + exp(as,k bs,k )), e,l = 1/(1 + exp(ae,l be,l )), σe,l
2 . Based on Lemma 1, it follows:
and f (t) is defined as the optimal value of problem (12), ⎡ ⎤
ts,k
which is a function over t. The function f (t) does not have B+ 2 I 0NT ×1 Bh̄s,k
εs,k
a closed-form solution, but is numerically tractable. ⎢ √ ⎥
⎣ 01×NT 1 − ρs,k χs,k ⎦  0. (15)
Remark 1: It is noted that the LMI constraint (12e) is √
h̄H
s,k B χs,k Js,k − ts,k
obtained from [42, Proposition 1] based on the assumption
rank(Fk ) ≤ 1, which will be shown later. By employing Lemma 2 again, (15) is rewritten as
It is observed that problem (12) is still nonconvex due
 t 
to (12b)–(12e). Now, we first reformulate constraint (12b). The B + s,k2 I B̂s,k
εs,k
following lemma is required to convert (12b) into a quadratic 0 (16)
matrix inequality (QMI). B̂H
s,k B̃s,k
Lemma 1 (Schur Complement [44]): Let X be a complex
where λs,k ≥ 0 is a slack variable
Hermitian matrix as
  ⎡ √ ⎤
1 − ρs,k χs,k 01×NJ
B1 B2 √
⎢ χs,k ⎥
X=X = H
. B̃s,k = ⎣ B − λs,k ḡH
s,k Q
BH
2 B3 k
λs,k
⎦
0NJ ×1 Qḡs,k Q + 2 INJ
ε̃s,k
−1
Then, we have X  0 if and only if B1 − BH 2 B3 B2  0 with
−1
B3  0, or B3 − BH 2 B1 B2  0 with B1  0. with Bk = h̄H s,k Bh̄s,k + ḡs,k Qḡs,k + σe,l − ts,k and B̂s,k =
H 2
K
Let us define B  k=1 Fk + Z. By utilizing Lemma 1 [ 0NT ×(NJ +1) Bh̄s,k ].
to (12b), the robust problem (12) can be converted into 
Defining Ck  [1/(2R̄s /t − 1)]Fk − ( j=k Wj + Z),
constraint (12d) can be similarly expressed from Lemma 1 as
max Ēs + Ēe (13a)
ρs,k ,Fk ,Z,Q,Ēs ,Ēe  
 √  ρs,k σp,k
1 − ρs,k χs,k  0 ∀k. (17)
s.t. √  0 ∀k (13b) σp,k hH s,k Ck hs,k − gs,k − σs,k
2
χs,k hHs,k Bhs,k + gs,k + σs,k
2

(10c), (10d) and (12c)−(12g). Applying Lemma 2 to (17) yields

⎡ ⎤
Problem (13) has semi-infinite constraints (12c)–(12e) ρs,k σp,k 01×NT
⎢ σp,k Cs,k − μs,k h̄H ⎥
and (13b). We can see that (12f) and (12g) are convex con- ⎣ s,k Ck ⎦  0 (18)
μs,k
straints. In order to make (13b) tractable, we exploit the 0NT ×1 Ck h̄s,k Ck + 2 I
εs,k
following lemma to convert constraint (13b) into LMIs.
Lemma 2 ([46, Th. 3.5]): If Di  0(i = 1, 2), then the QMIs where μs,k ≥ 0 is a slack variable and Cs,k = h̄H
s,k Ck h̄s,k −
for complex matrices Ak (k = 1, . . . , 6) and X gs,k −σs,k
2 . From Lemmas 1 and 2, (18) can be reformulated as
   
A1 A2 + A3 X μs,k
0 Ck + 2 I Ĉs,k
(A2 + A3 X)H A4 + XH A5 + AH 5 X + X A6 X
H εs,k
0 (19)
I − X Di X  0
H ĈH
s,k C̃s,k

where βs,k ≥ 0 is a slack variable

are equivalent to the following LMI:
⎡ ⎤
⎡ ⎤ ⎡ ⎤ ρs,k σp,k 01×NJ
A1 A2 A3 0 0 0 ⎢ σp,k ⎥
⎣ AH ⎦ − λ1 ⎣ 0 I C̃s,k = ⎣ Q − βs,k −ḡH
s,k Q
2 A4 AH 5 0 ⎦ s,k
βs,k
⎦
AH A5 A6 0 0 −D1 0NJ ×1 −Qḡs,k −Q + 2 INJ
ε̃s,k
3 ⎡ ⎤
0 0 0
− λ2 ⎣ 0 I 0 ⎦0 with Qs,k = h̄H s,k Ck h̄s,k − ḡs,k Qḡs,k − σs,k − μs,k and Ĉs,k =
H 2

0 0 −D2 [ 0NT ×(NJ +1) Ck h̄s,k ].

Next, in order to relax constraint (12e) as a tractable con-
where λi ≥ 0(i = 1, 2). vex LMI, we introduce the following lemma to obtain the
Then, denoting X = es,k , D1 = 1/εs,k 2 I, D = 0, A =
2 1 equivalent transformation of the worst case constraint (12e).
√
1 − ρs,k , A2 = χs,k , A3 = 01×NT , A4 = h̄H s,k B h̄s,k + gs,k + Lemma 3 [42]: Let h(J) = JH K1 J + JH K2 + KH 2 J + K3 ,
2 − t , A = h̄H B, and A = B, constraint (13b) can be
σe,l s,k 5 s,k 6 satisfying h(J)  0 for ∀J ∈ {J|Tr(JH K4 J) ≤ 1} with F4  0.
equivalently modified as Then, the following LMI holds:
⎡ √ ⎤    
1 − ρs,k χs,k 01×NT K3 KH 2 −α I 0
√ 0
⎢ χs,k Js,k − ts,k h̄H ⎥ K2 K1 0 −K4
⎣ s,k B ⎦  0 (14)
ts,k
0NT ×1 Bh̄s,k B+ 2 I where α ≥ 0.
εs,k

Authorized licensed use limited to: Zhengzhou University. Downloaded on February 05,2021 at 02:31:35 UTC from IEEE Xplore. Restrictions apply.
ZHU et al.: ROBUST BEAMFORMING DESIGNS IN SECURE MIMO SWIPT IoT NETWORKS WITH NONLINEAR CHANNEL MODEL 1707

From Lemma 3, constraint (12e) can be recast as be reformulated as

⎡  ⎤
ζ̄k,l I + H̄H Rk H̄e,l + 1
− 1 Ĝe,l H̄HR
k
max Ēs + Ēe (27a)
⎣ e,l t e,l
⎦  0 (20)
ζ
Rk H̄e,l Rk + k,l 2 I
s.t. (10c), (10d), (12f), (12g), (16), (19), (22), (26)
εe,l
ts,k ≥ 0, λs,k ≥ 0, μs,k ≥ 0, βs,k ≥ 0, ζk,l ≥ 0 (27b)
where ζk,l ≥ 0 is a slack variable, ζ̄k,l = ([1/t] − 1)σe,l
2 −ζ
k,l ηk,l ≥ 0, αe,l ≥ 0, γe,l ≥ 0, χs,k ≥ 0, ωe,l ≥ 0 (27c)
and Rk  ([1/t] − 1)Z − Fk . By using Lemma 1, we can
where = {ρs,k , {Fk }, Z, Q, Ēe , χs,k , ωe,l }. Problem (27) is
rewrite (20) as √
⎡ ⎤ still nonconvex due to χs,k in (16). Thus, introducing ps,k 
ζ √
Rk + k,l 2 I Rk H̄e,l χs,k as a slack variable, problem (27) is then recast as
⎣ ε ⎦0
e,l  (21)
H̄H
e,l R k Fk,l + 1
t − 1 Ĝe,l max p2s,k + Ēe (28a)
,ps,k
 t

where Fk,l = (([1/t]−1)σe,l2 −ζ )I+ H̄H R H̄ . By adopting
k,l e,l k e,l B + εs,k I B̂s,k
Lemma 2, (21) can be converted as s.t. s,k 0 (28b)
B̂s,k B̄s,k
⎡ ζ ⎤
Rk + k,l (10c), (10d), (12f), (12g), (19), (22)
2 I
εe,l
Rk H̄e,l 0NT ×NJ
⎢ ⎥ (26), (27b), (27c)
⎢ H̄H Rk ⎥
⎣ e,l Fk,l + ḠH
e,l Q̂Ḡe,l − ηk,l I e,l Q̂ ⎦  0
ḠH
ηk,l
0NJ ×NT Q̂Ḡe,l Q̂ + 2 I where
ε̃e,l ⎡ ⎤
(22) 1 − ρs,k ps,k 01×NJ
⎢ ps,k Bs,k − λs,k ḡH
s,k Q ⎦.
⎥
B̄s,k =⎣
where ηk,l ≥ 0 is a slack variable λs,k
 and Q̂ = ([1/t] − 1)Q. 0NJ ×1 Qḡs,k Q+ 2 I
ε̃s,k
Let us define L  I ⊗ ( K k=1 Fk + Z), M  I ⊗ Q,
h̄e,l  vec(H̄e,l ), ee,l  vec(Ee,l ), ḡe,l  vec(Ḡe,l ), and Problem (28) is still nonconvex due to the objective func-
ẽe,l  vec(Ẽe,l ). Applying vec(V1 V2 V3 ) = (VT3 ⊗V1 )vec(V2 ), tion p2s,k . We propose an iterative algorithm by using an
constraint (12c) can be equivalently written as SCA method to solve problem (28). The first-order Taylor
 H   expansion of p2s,k around the point p̃s,k is expressed as
h̄e,l + ee,l L h̄e,l + ee,l + Me,l + NE σe,l
2
≥ ωe,l ∀l (23) p̃2s,k + 2p̃s,k (ps,k − p̃s,k ). Thus, we tackle the objective function
where Me,l  (ḡe,l + ẽe,l )H M(ḡe,l + ẽe,l ). In order to of problem (28) by applying its Taylor expansion as
make (12c) tractable, we exploit the S-procedure to transform max p̃2s,k + 2p̃s,k (ps,k − p̃s,k ) + Ēe (29)
constraint (23) into LMIs. For completeness, the S-procedure ,ps,k
is presented in Lemma 4. s.t. (10c), (10d), (12f), (12g), (19)
Lemma 4 (S-Procedure [44, Appendix B.2]): Let a function (22), (26), (27b), (27c), (28b)
fn (x) for x ∈ CN×1 (n = 1, 2) be defined as
  Given p̃s,k , problem (29) becomes a convex SDP by ignoring
fn (x) = xH Vn x + 2 bH
n x + cn (24) the constraint rank(Fk ) = 1, and can be solved by convex
optimization softwares, such as CVX [45]. Based on the SCA
where Vn ∈ HN×N , bn ∈ CN×1 , and cn ∈ R. Then, fn (x) ≤ 0 method, an approximation with the current solution p̃s,k can be
holds if and only if there exists θ ≥ 0 such that updated iteratively, which implies that the inner loop problem
   
V b V b is optimally solved.
θ H1 1 − H2 2  0 Then, we consider the outer loop problem to obtain the
b1 c1 b2 c2
optimal value of t, which is given as
provided that there is a point x which satisfies fn (x) < 0. 
By employing the S-procedure, constraint (23) is min f (t) (30)
t
expressed as s.t. tmin ≤ t ≤ tmax
 
αe,l I + L Lh̄e,l where tmax and tmin stand for an upper bound and a lower
 0 ∀l (25)
e,l L Qe,l + Me,l − αe,l εe,l
h̄H 2
bound, respectively. We can solve problem (30) via a 1-D line
search method, and compute the optimal value of t. According
where αe,l ≥ 0 is a slack variable and Qe,l = h̄H e,l Lh̄e,l + to Remark 1, we prove the tightness of problem (29) in the
NE σe,l − ωe,l . Then, from Lemma 2, we get
2
following theorem.
⎡ ⎤ Theorem 1: For the feasible problem (11) with a given t > 0,
αe,l I + L Lh̄e,l 0NT NE ×NE NJ there exists an optimal solution to (29) such that rank(Fk ) = 1
⎢ h̄H Ve,l − γe,l ḡH ⎥
⎣ e,l L e,l M ⎦  0 ∀l (26) for k = 1, . . . , K.
γe,l
0NE NJ ×NT NE Mḡe,l M+ 2 I Proof: See the Appendix.
ε̃e,l
A key step for the line search method is to determine the
where γe,l ≥ 0 is a slack variable and Ve,l = Qe,l + ḡH
e,l Mḡe,l − lower and upper bounds of t. It is easily verified that an
αe,l εe,l
2 . According to (12)–(26), the original problem (11) can
upper bound can be set as tmax = 1 and a lower bound

Authorized licensed use limited to: Zhengzhou University. Downloaded on February 05,2021 at 02:31:35 UTC from IEEE Xplore. Restrictions apply.
1708 IEEE INTERNET OF THINGS JOURNAL, VOL. 8, NO. 3, FEBRUARY 1, 2021

is calculated as tmin = 1/log2 (1 + [(Phs,k 2 )/(σs,k

2 + σ 2 )])
p,k Let us define H̄s,k  h̄s,k h̄H s,k , Ḡs,k  ḡs,k ḡs,k , Ĥe,l  H̄e,l H̄e,l ,
H H

by using the method in [27]. Since t lies in the interval Ĝe,l  Ḡe,l Ḡe,l , Hs ,k  H̄s,k + s,k , Gs ,k  Ḡs,k + s,k ,
H
[1/log2 (1 + [(Phs,k 2 )/(σs,k
2 + σ 2 )]), 1], problem (30) can
p,k He ,l  Ĥe,l + e,l , and Ge ,l  Ĝe,l + e,l where s,k =
be solved by performing a 1-D line search over t and choosing
s,k + es,k h̄s,k + es,k es,k , s,k = ḡs,k ẽs,k + ẽs,k ḡs,k + ẽs,k ẽs,k ,
h̄s,k eH H H H H H
the minimum f (t) as an optimal solution of (29). Note that the
e,l = H̄e,l Ee,l + Ee,l H̄e,l + Ee,l Ee,l , and e,l = Ḡe,l Ẽe,l +
H H H H
optimal beamforming solution w∗k can be computed based on
e,l + Ẽe,l Ẽe,l , which stand for the CSI uncertainties.
Ẽe,l ḠH H
W∗k = w∗k w∗H
k . It is worth noting that the convex problem (29)
is not equivalent to the original problem (11), since the solution Then, the inequalities in (34b) and (34c) can be rearranged,
to (29) is suboptimal of (11) due to the SCA method. respectively, as
 σp,k
2

C. Low-Complexity SPCA Algorithm σs,k

2
+ j Hs ,k wj + z Hs ,k z + q Gs ,k q +
wH H H
ρs,k
j=k
In the previous section, we have addressed the max–min
problem by recasting the original problem as an SDP. To fHk Hs ,k fk
≤ ∀k (36a)
reduce the computational complexity of the two-layer algo- r1 − 1
rithm, we consider another reformulation of problem (10) σe,l
2
+ zH He ,l z + wH
k He ,l wk + q Ge ,l q
H

based on the SPCA method [48]. The optimization framework 2 + zH H

σe,l e ,l z + q Ge ,l q
H
can also be obtained as a convex form by incorporating chan- ≤ ∀l. (36b)
nel uncertainties. We consider the following lemma to provide r2
a convex approximation form for Cl,k in constraint (10b). It is straightforward to show that
       
Lemma 5 [43]: For any square matrix A  0, it holds true s,k  ≤ h̄s,k eH  + es,k h̄H  + es,k eH 
F s,k F s,k F s,k F
that  
= εs,k
2
+ 2εs,k h̄s,k   ξs,k (37)
|I + A| ≥ 1 + tr(A) (31)        
s,k  ≤ ḡs,k ẽ  + ẽs,k ḡ  + ẽs,k ẽ 
H H H
F s,k F s,k F s,k F
where the equality holds if and only if rank(A) ≤ 1.  
 
= ε̃s,k + 2ε̃s,k ḡs,k  ξ̃s,k
2
(38)
First, by applying Lemma 5, the robust secrecy rate con-        
e,l  ≤ H̄e,l EH  + Ee,l H̄H  + Ee,l EH 
straint (10b) can be relaxed as F e,l F
 
e,l F e,l F
⎛  ⎞  
= εe,l + 2εe,l H̄e,l F  ξe,l
2
(39)
tr HH F H
e,l k e,l        
Cs,k − log2 ⎝1 +  ⎠ ≥ R̄s    
e,l  ≤ Ḡe,l Ẽ  + Ẽe,l Ḡ  + Ẽe,l Ẽ 
H H  H 
F e,l e,l e,l
σe,l + tr He,l ZHe,l + Ge,l QGe,l
2 H H F  F F
 
= ε̃e,l + 2ε̃e,l Ḡe,l F  ξ̃e,l .
2
(40)
(32)
It is noted that s,k , s,k , e,l , and e,l are norm-bounded
which can be rewritten as (33), shown at the bottom of the
matrices.
page, where r1 > 0 and r2 > 0 are slack variables. Then, (33)
According to the results in [47], we can minimize con-
can be further simplified as
straint (34b) by minimizing the right-hand side (RHS) of (36a)
r1 r2 ≥ 2R̄s (34a) while maximizing its left-hand side (LHS). Then, (36a)
hH and (36b) can be approximately reformulated, respectively, as
s,k Fk hs,k 
 σp,k
2
max σs,k
2
+ j Hs ,k wj + z Hs ,k z
wH H
σs,k
2 +
j=k s,k Wj hs,k + hs,k Zhs,k + gs,k Qgs,k +
hH H H
ρs,k s,k F ≤ξs,k ,s,k F ≤ξ̃s,k j=k
≥ r1 − 1 ∀l (34b) σp,k
 2
fH
k Hs ,k fk
2 + tr HH ZH + GH QG
σe,l + qH Gs ,k q + ≤ min ∀k (41)
e,l e,l e,l e,l ρs,k s,k F ≤ξs,k r1 − 1
 ≥ r2 ∀k. (34c)
σe,l
2 + tr HH (Z + F )H + GH QG
e,l k e,l e,l e,l max σe,l
2
+ zH He ,l z
e,l F ≤ξe,l ,e,l F ≤ξ̃e,l
The equivalent form of (34a) is 2R̄s +2 +(r1 −r2 )2 ≤ (r1 +r2 )2 , + wH k He ,l wk + q Ge ,l q ≤
H
which is transformed into a second-order cone (SOC) form as
  σe,l
2 + zH H
e ,l z + q Ge ,l q
H
  min , ∀l. (42)
 2R̄s +2 r1 − r2  ≤ r1 + r2 . (35) e,l F ≤ξe,l ,e,l F ≤ξ̃e,l r2

log(r1 r2 ) ≥ R̄s (33a)

ρs,k hH
s,k Fk hs,k
1+   ≥ r1 ∀k (33b)
ρs,k σs,k
2 + hH Wh
j=k s,k j s,k + hH Zh
s,k s,k + gH Qg
s,k s,k + σ 2
p,k

H
tr He,l Fk He,l 1
1+  ≤ ∀l (33c)
σe,l + tr He,l ZHe,l + Ge,l QGe,l
2 H H r 2

Authorized licensed use limited to: Zhengzhou University. Downloaded on February 05,2021 at 02:31:35 UTC from IEEE Xplore. Restrictions apply.
ZHU et al.: ROBUST BEAMFORMING DESIGNS IN SECURE MIMO SWIPT IoT NETWORKS WITH NONLINEAR CHANNEL MODEL 1709

Then, a loose approximation method [47] can be utilized to By adopting the above Taylor expansion, for the points
minimize the RHS of (41) and (42) as (w̃k , r̃1 ), (z̃, r̃2 ), and (q̃, r̃2 ), (46) and (47) are transformed
into convex forms, respectively, as
fH
k Hs ,k fk fH H̄ξ ,k fk
min ≥ k s ∀k  σp,k
2
s,k F ≤ξs,k r1 − 1 r1 − 1
σs,k
2
+ j Hξs ,k wj + z Hξs ,k z + q Gξ̃s ,k q +
wH H H
σe,l
2 + zH H
e ,l z + q Ge ,l q
H ρs,k
j=k
min  
e,l F ≤ξe,l ,e,l F ≤ξ̃e,l r2 ≤ FH̄ξ ,k ,1 wk , r1 , w̃k , r̃1 (50a)
s

σe,l
2 + zH Ĥξe ,l z + qH Ĝξ̃e ,l q σe,l
2
+z H
Hξe ,l z + wH
k Hξe ,l wk + q Gξ̃e ,l q
H
≥ (43)  
r2 2 r2  
≤ σe,l
2
− 2 + FĤ ,0 z, r2 , z̃, r̃2
r̃2 r̃2 ξe ,l
where H̄ξs ,k = H̄s,k − ξs,k INT , Ĥξe ,l = Ĥe,l − ξe,l INT , and  
Ĝξ̃e ,l = Ĝe,l − ξ̃e,l INJ . Applying a similar technique to the + FĜ ,0 q, r2 , q̃, r̃2 ∀k, l. (50b)
ξ̃e ,l
LHS of (41) and (42), we obtain
 In addition, it can be easily checked that the EH con-
max σs,k
2
+ j Hs ,k wj + z Hs ,k z
wH H
straints (11b) and (11c) are nonconvex because of CSI
s,k F ≤ξs,k ,s,k F ≤ξ̃s,k j=k uncertainties. In order to approximate (11b) and (11c) as a
σp,k
2  convex one, we first explore a loose approximation method.
+ qH Gs ,k q + ≤ σs,k
2
+ j Hξs ,k wj + z Hξs ,k z
wH H
Following (6), the inverse function of prac (Pin ) can be
ρs,k
j=k written as
σp,k
2  ab 
  1 e (Pmax − prac )
+ qH Gξ̃s ,k q + ∀k (44) Pin prac  b − ln . (51)
ρs,k a eab prac + Pmax
max σe,l
2
+ zH He ,l z
e,l F ≤ξe,l ,e,l F ≤ξ̃e,l Then, applying (51), (11b) and (11c) can be transformed
to be
+ wH
k He ,l wk + q Ge ,l q ≤ σe,l
H 2

K
+ zH Hξe ,l z + wH
k Hξe ,l wk + q Gξ̃e ,l q ∀l
H
(45)
j H̄ξs ,k wj + z H̄ξs ,k z + q Ḡξ̃s ,k q
wH H H

j=1
where Hξs ,k = H̄s,k + ξs,k INT , Gξ̃s ,k = Ḡs,k + ξ̃s,k INJ , Hξe ,l =  
Pin Ēs
Ĥe,l + ξe,l INT , and Gξ̃e ,l = Ĝe,l + ξ̃e,l INJ . ≥ − σs,k
2
∀k (52a)
According to (41)–(45), (36a) and (36b) can be converted, 1 − ρs,k
respectively, as K
zH Ĥξe ,l z + wHj Ĥξe ,l wj + q Ĝξ̃e ,l q
H

 σp,k
2 j=1
σs,k
2
+ j Hξs ,k wj + z Hξs ,k z + q Gξ̃s ,k q +
wH H H  
ρs,k ≥ Pin Ēe − NE σe,l
2
∀l (52b)
j=k
fH
k H̄ξs ,k fk where Ḡξ̃s ,k = Ḡs,k + ξ̃s,k I. Substituting wj  w̃j + wj ,
≤ ∀k (46)
r1 − 1 z  z̃ + z, and q  q̃ + q into the LHS of (52a), we get
σe,l
2
+ zH Hξe ,l z + wHk Hξe ,l wk + q Gξ̃e ,l q
H

K

j H̄ξs ,k wj + z H̄ξs ,k z + q Ḡξ̃s ,k q

wH H H
σe,l
2 + zH Ĥξe ,l z + qH Ĝξ̃e ,l q
≤ ∀k, l. (47) j=1
r2
K 
 ! "
It is observed that (46) and (47) are nonconvex, while the RHS ≥ j H̄ξs ,k w̃j + 2 w̃j H̄ξs ,k wj
w̃H H
+ z̃H H̄ξs ,k z̃
of both (46) and (47) has a Quadratic-over-Linear (QoL) form, j=1
which is convex [44]. ! "
Following the idea of the constrained convex procedure [48], + q̃H Ḡξ̃s ,k q̃ + 2 z̃H H̄ξs ,k z + q̃H Ḡξ̃s ,k q (53)
the QoL functions can be replaced by their first-order expan-
where the inequality is obtained by dropping the quadratic
sions, which turn the problem into a convex programming
terms wH j H̄ξs ,k wj , z H̄ξs ,k z and q Ḡξ̃s ,k q.
H H
form. Specifically, defining a function
Similarly, in the LHS of (52b), it follows:
wH Aw 
K
fA,a (w, t) = (48) zH Ĥξe ,l z + j Ĥξe ,l wj + q Ĝξ̃e ,l q
wH H
t−a
j=1
for A  0 and t ≥ a, at a certain point (w̃, t̃), the first-order ! "
Taylor expansion of (48) is expressed as ≥ z̃ Ĥξe ,l z̃ + q̃H Ĝξ̃e ,l q̃ + 2 z̃H Ĥξe ,l z + q̃H Ĝξ̃e ,l q
H

K 
 ! "
  2{w̃H Aw} w̃H Aw̃
FA,a w, t, w̃, t̃ = − 2 (t − a). (49) + j Ĥξe ,l w̃j + 2 w̃j Ĥξe ,l wj
w̃H H
. (54)
t̃ − a t̃ − a j=1

Authorized licensed use limited to: Zhengzhou University. Downloaded on February 05,2021 at 02:31:35 UTC from IEEE Xplore. Restrictions apply.
1710 IEEE INTERNET OF THINGS JOURNAL, VOL. 8, NO. 3, FEBRUARY 1, 2021

TABLE I
C OMPLEXITY A NALYSIS OF D IFFERENT A LGORITHMS

Therefore, an approximate expression of (52b) can be D. Optimality Analysis

derived as For the two-layer method, several auxiliary various are used,
and then, the original problem is transformed into a convex
z̃H Ĥξe ,l z̃ + q̃H Ĝξ̃e ,l q̃ + 2{z̃H Ĥξe ,l z + q̃H Ĝξ̃e ,l q} optimization problem (29). The solutions of problem (10) can
K 
 ! " be obtained by iteratively updating problem (29). Since (29)
 
+ w̃H
j Ĥξe ,l w̃j + 2 w̃H
j Ĥξe ,l wj ≥ Pin Ēe − NE σe,l
2
. is a convex optimization problem, the obtained solution is
j=1 optimal at each iteration. Therefore, iteratively updating the
(55) variables will increase or maintain the value of the objective
function [49]. With the constraint of the maximum trans-
It is noted that (52a) is still nonconvex in its current form mit power, the objective value (29) become a nondecreasing
since it involves coupled Ēs and 1 − ρs,k . By defining sequence with an upper bound, and thus, the solution is at
K
as,k  j=1 (w̃j H̄ξs ,k w̃j + 2{w̃j H̄ξs ,k wj }) + z̃ H̄ξs ,k z̃
H H H
least local optimal. Meanwhile, the proposed two-layer method
+2{z̃H H̄ξs ,k z + q̃H Ḡξ̃s ,k q} + q̃H Ḡξ̃s ,k q̃ + σs,k
2 , (52a) can converges to a stationary solution. By applying the similar pro-
be rewritten as cedures, we can see that the proposed SOCP-SPCA method
    converges to a stationary solution that is at least local optimal.
as,k 1 − ρs,k ≥ Pin Ēs ∀k. (56)
IV. C OMPUTATIONAL C OMPLEXITY
Based on a first-order Taylor approximation of Pin (Ēs ), (56) In this section, we evaluate the computational complex-
can be recast as the SOC constraint ity of the proposed robust methods. As will be shown in
# $ Section V, the proposed SPCA algorithm exhibits almost the
 2es as,k + ρs,k − 1  ≤ as,k − ρs,k + 1 (57) same performance as the two-layer approach with substantially
reduced complexity. We compare the complexity of the algo-
where es = Pin (Ẽs ) + Pin (Ẽs )(Ēs − Ẽs ) and Pin is defined as rithms through analyses similar to that in [25] and [50]. The
  complexity of the proposed algorithms are shown in Table I,
  Pmax 1 + eab where we denote n, T = log2 [(tmax − tmin )/η], and Q as the
Pin Ẽs =   . (58)
a Pmax + Ẽs eab Pmax − Ẽs number of decision variables, the two-layer search size, and
the SPCA iteration number, respectively.
In addition, (10c) can be reformulated as two SOC forms The two-layer method in problem (29) involves K + 1 LMI
# T $  constraints of size NT , one LMI constraint of size NJ , 2K LMI
w · · · wTK zT  ≤ PT (59a)
1
 T  constraints of size NT + NJ + 2, KL LMI constraints of size
q  ≤ PJ . (59b) NT +NJ +NE , L LMI constraints of size NT NE +NJ NE +1, and
2KL + 6K + 3L + 3 linear constraints. In contrast, the SOCP-
Now, we propose a new algorithm based on the SPCA SPCA algorithm in problem (60) has one SOC constraint of
method [48], which iteratively optimizes {F̃k , q̃, z̃, r̃1 , r̃2 , Ẽs }. dimension 2, K SOC constraints of dimension KNT + 1, KL
(n) (n) (n) (n)
Let us denote  (n) = {w̃k , q̃(n) , z̃(n) , r̃1 , r̃2 , Ẽs } as the SOC constraints of dimension 2NT +NJ +1, K SOC constraints
optimal solution obtained at the nth iteration. At the (n + 1)th of dimension (K + 1)NT + NJ + 2, one SOC constraint of
iteration, given  (n) , problem (10) is converted into the convex dimension (K + 1)NT , one SOC constraint of dimension NJ ,
form as and 2K + 2L + 4 linear constraints.
For example, for a system with K = 2, L = 2, M = 1,
max Ēs + Ēe (60) NT = 4, NJ = 4, NE = 2, D = 15, and Q = 5,
ρs,k ,{fk },z,q,Ēs ,Ēe ,r1 ,r2
the complexity of the SDP with two-layer method and the
s.t. (35), (50a), (50b), (55), (57) SOCP-SPCA method with one CJ equals O(1.04 × 109 ) and
(59a), (59b) O(4.40×105 ), respectively. Thus, the computational complex-
0 < ρs,k ≤ 1, z = z − z̃(n) ity of the SOCP-SPCA method is only 0.4% compared to the
q = q − q̃(n) , wk = wk − w̃(n)
k ∀k. SDP with two-layer method.

Problem (60) is an SOCP problem [44], which can be solved V. S IMULATION R ESULTS
by employing the CVX tools [45]. Note that the convex In this section, we provide the simulation results to val-
problem (60) is not equivalent to the orignal problem (11). idate the performance of our proposed scheme. The system

Authorized licensed use limited to: Zhengzhou University. Downloaded on February 05,2021 at 02:31:35 UTC from IEEE Xplore. Restrictions apply.
ZHU et al.: ROBUST BEAMFORMING DESIGNS IN SECURE MIMO SWIPT IoT NETWORKS WITH NONLINEAR CHANNEL MODEL
TABLE II
S YSTEM PARAMETERS
Fig. 2. Average harvested energy with respect to the iteration numbers for
the SPCA method.
parameters that we have considered for the simulations are
summarized in Table II, where Tx means the transmitter.
The estimated channels hs,k , He,l ,gs,k , and Ge,l are, respec-
tively, modeled as hs,k = H(ds,k )hI , He,l = H(de,l )HI ,
gs,k = H(fs,k )gI , and Ge,l = H(fe,l )GI , where H(d) =
(c/4π fc )(1/d)(κ/2) , hI ∼ CN (0, I), HI ∼ CN (0, I), gs,k ∼
CN (0, I), and Ge,l ∼ CN (0, I). We fix the channel error
bound for the deterministic model as εs = εs,k = ε̃s,k ∀k
and εe = εe,l = ε̃e,l ∀k. For the nonlinear EH model, we
set M = Ms,k = Me,l = 3.9 mW, a = as,k = ae,l = 1500,
and b = bs,k = be,l = 0.0022 according to the measurement
data results [39]. In our simulations, we compare the no-AN
scheme which is obtained by setting Qm = 0 ∀m [21], no-CJ
scheme which is computed by Z = 0 [27], and the nonrobust
method which assumes no uncertainty in the CSI [27].
Fig. 2 illustrates the convergence of the SOCP-SPCA
method with respect to the iteration numbers for PT =
40 dBm, PJ = 40 dBm, R̄s = 0.5 b/s/Hz, and ε = 0.01.
We can see that the convergence of the SOCP-SPCA method
is achieved for all cases within just five iterations.
Fig. 3 shows the average harvested energy in terms of differ-
ent target secrecy rates with PT = 30 dBm and PJ = 30 dBm.
Here, “Two-layer” and “SOCP-SPCA” denote our proposed
schemes, while “no-AN,” “no-CJ,” and “no-robust” represent
the benchmark designs. It is observed that the harvested power
Authorized licensed use limited to: Zhengzhou University. Downloaded on February 05,2021 at 02:31:35 UTC from IEEE Xplore. Restrictions apply.
1711
Fig. 3. Average harvested energy with respect to the target secrecy rate.
Fig. 4. Average harvested energy with respect to the power budget at CJ.
of all schemes declines with the increased secrecy rate target.
In addition, the SOCP-SPCA algorithm can achieve almost the
same performance as the two-layer method with much lower
complexity. When  = 0.01, the average harvested power of
the SOCP-SPCA algorithm reaches 99.85% of the two-layer
algorithm. Compared with the no-AN scheme and the no-CJ
scheme, the harvested power of the SOCP-SPCA algorithm is
9 and 8 dB higher, respectively. Moreover, we can check that
the proposed algorithm outperforms the nonrobust scheme,
and the performance gap increases as the target secrecy rate
becomes large.
Fig. 4 depicts the average harvested energy in terms of the
power budget at CJ with PT = 10 dBm and R̄s = 0.5 b/s/Hz.
It is easily observed that the achieved harvested power grows
with PJ . Also the perfect CSI case, the two-layer algorithm,
and the SOCP-SPCA algorithm exhibit the same slope. Also,
the average harvested power of the SOCP-SPCA algorithm
reaches 99.91% of the two-layer algorithm regardless of .
Moreover, we can check that as PJ increases, the performance
gap between the proposed SOCP-SPCA algorithm and the no-
AN scheme becomes larger and a performance loss of the
nonrobust scheme grows.
1712 IEEE INTERNET OF THINGS JOURNAL, VOL. 8, NO. 3, FEBRUARY 1, 2021

harvested power of the no-AN scheme shows lower growth

with respect to NJ .

VI. C ONCLUSION
In this article, we have studied robust secure beamforming
designs for a PS MIMO SWIPT IoT network under a nonlinear
EH model with cooperative jamming, where norm-bounded
channel uncertainties are present. The formulated optimization
problem has been transformed into a tractable two-layer form.
The inner layer problem has been recast as a sequence of SDPs
by converting all nonconvex constraints into LMIs. Then the
optimal solution to the outer layer problem can be obtained
by conducting a 1-D line search. To reduce the computational
complexity due to the 1-D search technique, an SPCA-based
iterative algorithm has been proposed. Simulation results have
Fig. 5. CDF of the target secrecy rate at the CR nodes.
shown that the proposed robust design outperforms conven-
tional schemes reported in the literature. In particular, it has
been illustrated that the proposed SPCA algorithm can achieve
almost the same performance as the two-layer search method
with significantly reduced complexity. The proposed robust
secrecy design framework and solution approach have poten-
tials to be extended to more complex SWIPT systems by
incorporating new security strategies.

A PPENDIX
P ROOF OF T HEOREM 1
First, we reformulate the constraints of problem (29).
Equation (19) can be rewritten as
s,k + H
s,k Ck s,k − J s,k Qs,k J  0 ∀k
H H
(61)

where s,k = [ INT 0NT ×(NJ +1) hs,k ], J =

Fig. 6. Average harvested energy with respect to the number of transmit [ 0(NJ +2)×NT INJ +2 ], s,k = [ INJ 0NJ ×1 gs,k ], and
antennas at CJ.
s,k = diag{(μs,k /εs,k2 )I , F̂ } with
NT s,k
⎡ ⎤
ρs,k σp,k 01×NJ
⎢ 01×NJ ⎥
F̂s,k = ⎣ σp,k −σs2 − μs,k − βs,k
Fig. 5 plots the cumulative density function (CDF) of the
⎦.
target secrecy rate with PT = 40 dBm, PJ = 30 dBm, 0NJ ×1 0NJ ×1
βs,k
2 INJ ε̃s,k
 = 0.01, and R̄s = 1 b/s/Hz. We can see that the two-layer
algorithm and the SOCP-SPCA algorithm always satisfy the
Also, we can recast (22) as
secrecy rate target (R̄s = 1 b/s/Hz) regardless of M, while the
nonrobust scheme achieves only 40% of the predefined target. k,l + H
e,l Rk e,l + e,l E Q̂e,l E  0
H H
∀k, l (62)
Finally, the role of the number of transmit antennas NJ at
CJ in the average harvested power performance is depicted where e,l = [ INT He,l 0NT ×NJ ], e,l = [ Ge,l INJ ],
in Fig. 6 for different channel error bound values with PT = E = [ 0(NE +NJ )×NT INE +NJ ], k,l = ([1/t] − 1)σe,l 2 −ζ −
k,l
40 dBm, PJ = 40 dBm, and R̄s = 0.5 b/s/Hz. From this ηk,l , and k,l = diag{(ζk,l /εe,l )INT , k,l INE , (ηk,l /ε̃e,l
2 2 )I }.
NJ
figure, we observe that the two-layer and SOCP-SPCA algo- Then, (26) can be given by
rithms have the same performance in terms of NJ . Also, the
SOCP-SPCA algorithms outperforms the no-AN scheme, the ˜H
e,l +  ˜ ˜H H ˜
e,l Le,l + e,l T Me,l T  0 ∀l (63)
no-CJ scheme and the nonrobust scheme. Increasing NJ from
where  ˜ e,l = [ IN N h ˜
e,l 0NT NE ×NE NJ ], e,l =
2 to 6 improves the average harvested power for the proposed T E
algorithms by about 4 dB. Due to the low EH efficiency of the [ ge,l INE NJ ], Ñl = NE σe,l − Ēe,l − αe,l εe,l − γe,l ,
2 2
nonlinear EH models, these results are as expected. Although T = [ 0(NE NJ +1)×NT NE INE NJ +1 ], and e,l =
a similar trend happens in the both proposed algorithms for diag{αe,l INT NE , Ñl , (γe,l /ε̃e,l
2 )I }.
NE NJ
ε = 0.01, it is noted that they both result in an average increase Furthermore, we modify (28b) as
of about 4.3 dB in terms of the average harvested power for EH
as compared to the no-CJ scenario. In addition, the average ˜ s,k + H
 s,k Bs,k + J s,k Qs,k J  0 ∀k
H H
(64)

Authorized licensed use limited to: Zhengzhou University. Downloaded on February 05,2021 at 02:31:35 UTC from IEEE Xplore. Restrictions apply.
ZHU et al.: ROBUST BEAMFORMING DESIGNS IN SECURE MIMO SWIPT IoT NETWORKS WITH NONLINEAR CHANNEL MODEL 1713

where B̃k = σe,l

2 − t
s,k − λs,k and ˜ s,k = Fk Mk = 0, ZN = 0, QP = 0, Mk  0, N  0
diag{(ts,k /εs,k )INT , Ks,k } with
2
P  0, Ss,k  0, Tk,l  0, Ue,l  0, Vs,k  0 (65c)
⎡ ⎤  
1 − ρs,k ps 01×NJ s,k + H s,k Ck s,k − J s,k Qs,k J Ss,k = 0.
H H
(65d)
⎢ 01×NJ ⎥
Ks,k = ⎣ ps B̃k ⎦. Subtracting (65b) from (65a) generates
λs,k
0NJ ×1 0NJ ×1 2 INJ ε̃s,k t
Mk − N + s,k Ss,k H
s,k + s,k Ss,k s,k
H
We consider the Lagrange dual function of (29) as 2 s −t
R̄
  L L  
L {Fk }, Q, Z, {Mk }, N, P, Ss,k , Tk,l , Ue,l , Vs,k , γ , ν − e,l Tk,l e,l −
H 1
− 1 e,l Tk,l He,l

K t
√ l=1 l=1
= τ ps,k + Ēe + tr(Fk Mk ) + tr(ZN) + tr(QP)
− s,k Vs,k H
s,k = 0. (66)
k=1

K
   Then, it follows:
+ tr Ss,k s,k + H
s,k Ck s,k − J s,k Qs,k J
H H
2R̄s
k=1 Mk + s,k Ss,k Hs,k

K 
L 2R̄s − t
 
+ ν(tr(Q) − PJ ) + tr Tk,l k,l + H
e,l Rk e,l 1
L

k=1 l=1 =N+ e,l Tk,l H

e,l + s,k Vs,k s,k .
H
(67)
 K  t
 l=1
+ H
e,l E Q̂e,l E
H
+γ tr(Fk ) + tr(Z) − PT It is observedthat the RHS of (67) is positive semidef-
k=1 inite, i.e., N + Ll=1 (1/t)e,l Tk,l H
e,l + s,k Vs,k s,k  0.
H


K   Premultiplying both sides of the above equality by Fk and
+ tr Vs,k ˜ s,k + H
s,k Bs,k +  
H H
J s,k Q s,k  J applying the inverse yield
k=1
 K   2R̄s

L
  L 
NE  (j,j) Fk s,k Ss,k Hs,k
+ tr Ue,l e,l + tr Fk + Z De,l 2R̄s − t
 −1
l=1 l=1 j=1 k=1  L
1

L 
NJ  × N+ e,l Tk,l e,l + s,k Vs,k s,k
H H
= Fk .
(j,j) t
+ tr QEe,l l=1
l=1 j=1 Hence, we have the rank relation as
⎛
where Mk ∈ HNT
+ , N ∈ HN NT NT
+ , P ∈ H+ , Sk ∈ H+ , Tk,l ∈ H+ ,
T NT
2R̄s
NT NT
Ue,l ∈ H+ , Vs,k ∈ H+ , γ ∈ R+ , and ν ∈ R+ are the dual rank(Fk ) = rank⎝ Fk s,k Ss,k Hs,k
2R̄s − t
variables of Fk , Q, P, (61)–(64), and (10c), respectively, and
(j,j)
De,l ∈ HN
(j,j) NT  −1 ⎞
+ and Ee,l ∈ H+ denote a block submatrix of
T
 L
1
 ˜H
˜ e,l Ue,l  ˜ H ˜H × N+ e,l Tk,l He,l + s,k Vs,k s,k
H ⎠
e,l and e,l T Ue,l T e,l , respectively, as t
⎡ (1,1) (1,N )
⎤ l=1
De,l ··· De,l E
⎢ . .. ⎥ ≤ rank(s,k Ss,k H
s,k ). (68)
 ˜H
˜ e,l Ue,l  ⎢ . .. ⎥
e,l = ⎣ . . . ⎦
(N ,1) (N ,N ) Now, we will compute the rank of the matrix s,k Ss,k H
s,k .
De,lE · · · De,lE E To this end, we consider the following equalities as:
⎡ (1,1) ⎤
Ee,l ···
(1,N )
Ee,l J # $
⎢ . ⎥ I 0 H s,k = I
˜ e,l T Ue,l H ˜H ⎢ . .. .. ⎥. # $
 T e,l = ⎣ . . . ⎦ μs,k  # $
(NJ ,1) (NJ ,NJ )
I 0 s,k = 2 s,k − 0 hs,k
Ee,l · · · Ee,l εs,k
# $ H
The related Karush–Kuhn–Tucker (KKT) conditions are I 0 J = 0. (69)
listed as Premultiplying [I 0] and post-multiplying H s,k to (65d)
∂L t generate
= Mk + s,k Ss,k H
s,k + γ I
∂Fk 2 −t
R̄s # $ 
I 0 s,k + H s,k Ck s,k − J s,k Qs,k J Ss,k s,k = 0
H H H
L L NE
(j,j)
− e,l Tk,l H
e,l + De,l = 0 (65a) (70)
l=1 l=1 j=1
and thus we get
L 
   
∂L 1
= N − s,k Ss,k s,k + γ I +
H
− 1 e,l Tk,l H μs,k μs,k # $
∂Z t e,l
I + Ck s,k Ss,k H
s,k = 2 0 hs,k Ss,k H
s,k .
l=1 εs,k
2 εs,k

L 
NE
(j,j)
+ s,k Vs,k H
s,k + De,l = 0 (65b) It is easily verified that (μs,k /εs,k
2 )I + C  0 and it is
k
l=1 j=1 nonsingular. Thus, multiplying this matrix will not change the

Authorized licensed use limited to: Zhengzhou University. Downloaded on February 05,2021 at 02:31:35 UTC from IEEE Xplore. Restrictions apply.
1714
rank of the resulting matrix, and we obtain
  
μs,k
rank(s,k Ss,k s,k ) = rank
H
I + Ck s,k Ss,k s,k
H
εs,k
2
 
μs,k # $ 7, no. 3, pp. 1884–1897, Mar. 2020.
= rank 2 0 hs,k Ss,k s,k H
εs,k
# $
≤ rank 0 hs,k ≤ 1. (71)
This completes the proof.
R EFERENCES
[1] C. X. Wang et al., “Cellular architecture and key technologies for 5G
wireless communication networks,” IEEE Commun. Mag., vol. 52, no. 2,
pp. 122–130, Feb. 2014.
[2] L. Atzori, A. Iera, and G. Morabito, “The Internet of Things: A survey,”
Comput. Netw., vol. 54, no. 15, pp. 2787–2805, Oct. 2010.
[3] A. Al-Fuqaha, M. Guizani, M. Mohammadi, M. Aledhari, and
M. Ayyash, “Internet of Things: A survey on enabling technologies, pro-
tocols, and applications,” IEEE Commun. Surveys Tuts., vol. 17, no. 4,
pp. 2347–2376, 4th Quart., 2015.
[4] Z. Chu et al., “Resource allocation for secure wireless powered inte-
grated multicast and unicast services with full duplex self-energy
recycling,” IEEE Trans. Wireless Commun., vol. 18, no. 1, pp. 620–636,
Jan. 2019.
[5] P. Kamalinejad et al., “Wireless energy harvesting for the Internet of
Things,” IEEE Commun. Mag., vol. 53, no. 6, pp. 102–108, Jun. 2015.
[6] Z. Chu, F. Zhou, Z. Zhu, Q. Hu, and P. Xiao, “Wireless powered sen-
sor networks for Internet of Things: Maximum throughput and optimal
power allocation,” IEEE Internet Things J., vol. 5, no. 1, pp. 310–321,
Feb. 2018.
[7] P. Grover and A. Sahai, “Shannon meets Tesla: Wireless information
and power transfer,” in Proc. IEEE ISIT, pp. 2363–2367, Jun. 2010.
[8] Z. Chu, F. Zhou, Z. Zhu, M. Sun, and N. Al-Dhahir, “Energy beamform-
ing design and user cooperation for wireless powered communication
networks,” IEEE Wireless Commun. Lett., vol. 6, no. 6, pp. 750–753,
Dec. 2017.
[9] H. Ko and S. Pack, “OB-DETA: Observation-based directional energy
transmission algorithm in energy harvesting networks,” J. Commun.
Netw., vol. 21, no. 2, pp. 168–176, Apr. 2019.
[10] R. Zhang and C. Ho, “MIMO broadcasting for simultaneous wire-
less information and power transfer,” IEEE Trans. Wireless Commun.,
vol. 12, no. 5, pp. 1989–2001, May 2013.
[11] J. Tang, D. K. C. So, N. Zhao, A. Shojaeifard, and K.-K. Wong,
“Energy efficiency optimization with SWIPT in MIMO broadcast chan-
nels for Internet of Things,” IEEE Internet Things J., vol. 5, no. 4,
pp. 2605–2619, Aug. 2018.
[12] Y. Huang, M. Liu, and Y. Liu, “Energy-efficient SWIPT in IoT dis-
tributed antenna systems,” IEEE Internet Things J., vol. 5, no. 4,
pp. 2646–2656, Aug. 2018.
[13] Y. Hu, Y. Zhu, M. C. Gursoy, and A. Schmeink, “SWIPT-enabled relay-
ing in IoT networks operating with finite blocklength codes,” IEEE J.
Sel. Areas Commun., vol. 37, no. 1, pp. 74–88, Jan. 2019.
[14] D. S. Gurjar, H. H. Nguyen, and H. D. Tuan, “Wireless information and
power transfer for IoT applications in overlay cognitive radio networks,”
IEEE Internet Things J., vol. 6, no. 2, pp. 3257–3270, Apr. 2019.
[15] D. K. P. Asiedu, H. Lee, and K. Lee, “Simultaneous wireless information
and power transfer for decode-and-forward multihop relay systems in
energy-constrained IoT networks,” IEEE Internet Things J., vol. 6, no. 6,
pp. 9413–9426, Dec. 2019.
[16] D. Mishra, G. C. Alexandropoulos, and S. De, “Energy sustainable IoT
with individual QoS constraints through MISO SWIPT multicasting,”
IEEE Internet Things J., vol. 5, no. 4, pp. 2856–2867, Aug. 2018.
[17] S. Goel and R. Negi, “Guaranteeing secrecy using artificial noise,” IEEE
Trans. Wireless Commun., vol. 7, no. 6, pp. 2180–2189, Jun. 2008.
[18] Z. Chu, Z. Zhu, M. Johnston, and S. Le goff, “Simultaneous wireless
information power transfer for MISO secrecy channel,” IEEE Trans. Veh.
Technol., vol. 65, no. 9, pp. 6913–6925, Sep. 2016.
[19] P. Lubega, T. Ssettumba, H. Nabuuma, and J. Serugunda, “A secure
energy efficient multi-user selection scheme for SWIPT wireless IoT
networks in the presence of cooperative jamming,” in Proc. 27th
Telecommun. Forum, 2019, pp. 1–4.
[20] D. Xu and H. Zhu, “Secure transmission for SWIPT IoT systems
with full-duplex IoT devices,” IEEE Internet Things J., vol. 6, no. 6,
pp. 10915–10933, Dec. 2019.
Authorized licensed use limited to: Zhengzhou University. Downloaded on February 05,2021 at 02:31:35 UTC from IEEE Xplore. Restrictions apply.
IEEE INTERNET OF THINGS JOURNAL, VOL. 8, NO. 3, FEBRUARY 1, 2021
[21] Z. Deng, Q. Li, Q. Zhang, L. Yang, and J. Qin, “Beamforming design
for physical layer security in a two-way cognitive radio IoT network
with SWIPT,” IEEE Internet Things J., vol. 6, no. 6, pp. 10786–10798,
Dec. 2019.
[22] X. Sun, W. Yang, and Y. Cai, “Secure communication in NOMA assisted
millimeter wave SWIPT UAV networks,” IEEE Internet Things J., vol.
[23] Z. Chu, K. Cumanan, Z. Ding, M. Johnston, and S. Le goff, “Robust
outage secrecy rate optimizations for a MIMO secrecy channel,” IEEE
Wireless Commun. Lett., vol. 4, no. 1, pp. 86–89, Feb. 2015.
[24] Z. Chu, Z. Zhu, W. Xiang, and J. Hussein, “Robust beamforming
and power splitting design in MISO SWIPT downlink system,” IET
Commun., vol. 10, no. 6, pp. 691–698, Apr. 2016.
[25] Z. Zhu, Z. Chu, Z. Wang, and I. Lee, “Outage constrained robust beam-
forming for secure broadcasting systems with energy harvesting,” IEEE
Trans. Wireless Commun., vol. 15, no. 11, pp. 7610–7620, Nov. 2016.
[26] Z. Chu, Z. Zhu, and J. Hussein, “Robust optimization for AN-aided
transmission and power splitting for secure MISO SWIPT system,” IEEE
Commun. Lett., vol. 20, no. 8, pp. 1571–1574, Aug. 2016.
[27] Z. Zhu, Z. Chu, N. Wang, S. Huang, Z. Wang, and I. Lee, “Beamforming
and power splitting designs for AN-aided secure multi-user MIMO
SWIPT systems,” IEEE Trans. Inf. Forensics Security, vol. 12, no. 12,
pp. 2861–2874, Dec. 2017.
[28] Z. Chu, H. Nguyen, T. Le, M. Karamanoglu, E. Ever, and A.
Yazici, “Secure wireless powered and cooperative jamming D2D
communications,” IEEE Trans. Green Commun. Netw., vol. 2, no. 1,
pp. 1–13, Mar. 2018.
[29] Z. Chu, K. Cumanan, Z. Ding, M. Johnston, and S. Le goff, “Secrecy rate
optimizations for a MIMO secrecy channel with a cooperative jammer,”
IEEE Trans. Veh. Technol., vol. 64, no. 5, pp. 1833–1847, May 2015.
[30] J. Moon, H. Lee, C. Song, and I. Lee, “Secrecy performance
optimization for wireless powered communication networks with an
energy harvesting jammer,” IEEE Trans. Commun., vol. 65, no. 2,
pp. 764–774, Feb. 2017.
[31] Z. Chu, T. A. Le, H. X. Nguyen, M. Karamanoglu, and Z. Zhu, “Robust
design for MISO SWIPT system with artificial noise and cooperative
jamming,” in Proc. IEEE GLOBECOM, Dec. 2017, pp. 1–6.
[32] Z. Zhu, Z. Chu, N. Wang, Z. Wang, and I. Lee, “Energy harvesting fair-
ness in AN-aided secure MU-MIMO SWIPT systems with cooperative
jammer,” in Proc. IEEE ICC, May 2018, pp. 1–6.
[33] F. Zhou, Z. Li, J. Cheng, Q. Li, and J. Si, “Robust AN-aided beam-
forming and power splitting design for secure MISO cognitive radio with
SWIPT,” IEEE Trans. Wireless Commun., vol. 16, no. 4, pp. 2450–2464,
Apr. 2017.
[34] Q. Qi, X. Chen, D. W. K. Ng, C. Zhongl, and Z. Zhang, “Robust beam-
forming design for SWIPT in cellular Internet of Things,” in Proc. IEEE
ICCC, Aug. 2019, pp. 523–528.
[35] Q. Qi, X. Chen, and D. W. K. Ng, “Robust beamforming for NOMA-
based cellular massive IoT with SWIPT,” IEEE Trans. Signal Process.,
vol. 68, pp. 211–224, Jan. 2020.
[36] R. Jiang, K. Xiong, P. Fan, Y. Zhang, and Z. Zhong, “Power
minimization in SWIPT networks with coexisting power-splitting
and time-switching users under nonlinear EH model,” IEEE Internet
Things J., vol. 6, no. 5, pp. 8853–8869, Oct. 2019.
[37] X. Hu, B. Li, K. Huang, Z. Fei, and K. Wong, “Secrecy energy efficiency
in wireless powered heterogeneous networks: A distributed ADMM
approach,” IEEE Access, vol. 6, pp. 20609–20624, 2018.
[38] E. Boshkovska, D. W. K. Ng, N. Zlatanov, and R. Schober, “Practical
non-linear energy harvesting model and resource allocation for SWIPT
systems,” IEEE Commun. Lett., vol. 19, no. 12, pp. 2082–2085, Dec.
2015.
[39] K. Xiong, B. Wang, and K. J. R. Liu, “Rate-energy region of SWIPT
for MIMO broadcasting under nonlinear energy harvesting model,” IEEE
Trans. Wireless Commun., vol. 16, no. 8, pp. 5147–5161, Aug. 2017.
[40] E. Boshkovska, D. W. K. Ng, N. Zlatanov, A. Koelpin, and R. Schober,
“Robust resource allocation for MIMO wireless powered communication
networks based on a non-linear EH model,” IEEE Trans. Commun.,
vol. 65, no. 5, pp. 1984–1999, May 2017.
[41] Y. Lu, K. Xiong, P. Fan, Z. Ding, Z. Zhong, and K. B. Letaief,
“Global energy efficiency in secure MISO SWIPT systems with non-
linear power-splitting EH model,” IEEE J. Sel. Areas Commun., vol. 37,
no. 1, pp. 216–232, Jan. 2019.
[42] 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.
[43] K. B. Petersern and M. S. Pedersern, The Matrix Cookbook.
Cambridge, MA, USA: MIT Press, Nov. 2008. [Online]. Available:
http://matrixcookbook.com
ZHU et al.: ROBUST BEAMFORMING DESIGNS IN SECURE MIMO SWIPT IoT NETWORKS WITH NONLINEAR CHANNEL MODEL
[44] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge, U.K.:
Cambridge Univ. Press, 2004.
[45] M. Grant and S. Boyd. (Sep. 2012). CVX: MATLAB Software
for Disciplined Convex Programming. [Online]. Available:
http://cvxr.com/cvx
[46] Z.-Q. Luo, J. F. Sturm, and S. Zhang, “Multivariate nonnegative
quadratic mappings,” SIAM J. Optim., vol. 14, no. 4, pp. 1140–1162,
Jul. 2004.
[47] M. Bengtsson and B. Ottersten, “Optimum and suboptimum transmit
beamforming,” in Handbook of Antennas in Wireless Communications,
L. C. Godara, Ed. Boca Raton, FL, USA: CRC Press, Aug. 2001.
[48] A. S. Vishwanathan, A. J. Smola, and S. V. N. Vishwanathan, “Kernel
methods for missing variables,” in Proc. Int. Workshop Artif. Intell. Stat.,
Jan. 2005, pp. 325–332.
[49] Q. Zhang, Q. Li, and J. Qin, “Robust beamforming for nonorthogonal
multiple-access systems in MISO channels,” IEEE Trans. Veh. Technol.,
vol. 65, no. 12, pp. 10231–10236, Dec. 2016.
[50] K.-Y. Wang, A. M.-C. So, T.-H. Chang, W.-K. Ma, and C.-Y. Chi,
“Outage constrained robust transmit optimization for multiuser MISO
downlinks: Tractable approximations by conic optimization,” IEEE
Trans. Signal Process., vol. 62, no. 21, pp. 5690–5705, Nov. 2014.
Zhengyu Zhu (Member, IEEE) received the B.S.
degree from Henan University, Kaifeng, China,
in 2010, and the Ph.D. degree from Zhengzhou
University, Zhengzhou, China, in 2017.
He is currently with the School of Information
Engineering, Zhengzhou University. His research
interests include information theory and signal
processing for wireless communications, such as
MIMO wireless network, physical layer security,
wireless cooperative networks, convex optimization
techniques, and energy harvesting communication
systems.
Dr. Zhu has served on the Technical Program Committees of international
conferences, including IEEE GLOBECOM, IEEE ICC, IEEE WCNC, IEEE
VTC, and ICCC.
Ning Wang (Member, IEEE) received the B.E.
degree in communication engineering from Tianjin
University, Tianjin, China, in 2004, the M.A.Sc.
degree in electrical engineering from the University
of British Columbia, Vancouver, BC, Canada, in
2010, and the Ph.D. degree in electrical engineering
from the University of Victoria, Victoria, BC,
Canada, in 2013.
From 2004 to 2008, he was with the China
Information Technology Design and Consulting
Institute, Zhengzhou, China, as a Mobile
Communication System Engineer, specializing in planning and design of
large-scale commercial mobile communication networks, network traffic
analysis, and radio network optimization. He was a Postdoctoral Research
Fellow with the Department of Electrical and Computer Engineering,
University of British Columbia from 2013 to 2015. Since 2015, he has
been with the School of Information Engineering, Zhengzhou University,
Zhengzhou, where he is currently an Associate Professor. He also holds
adjunct appointments with the Department of Electrical and Computer
Engineering, McMaster University, Hamilton, ON, Canada. His research
interests include resource allocation and security designs of future cellular
networks, channel modeling for wireless communications, statistical signal
processing, and cooperative wireless communications.
Dr. Wang was on the Finalist of the Governor General’s Gold Medal for
outstanding graduating doctoral student with the University of Victoria in
2013. He has served on the technical program committees of international
conferences, including IEEE GLOBECOM, IEEE ICC, IEEE WCNC, IEEE
VTC, and CyberC.
Authorized licensed use limited to: Zhengzhou University. Downloaded on February 05,2021 at 02:31:35 UTC from IEEE Xplore. Restrictions apply.
View publication stats
1715
Wanming Hao received the Ph.D. degree from the
School of Electrical and Electronic Engineering,
Kyushu University, Fukuoka, Japan, in 2018.
He has worked as a Research Fellow with the
5G Innovation Center, Institute of Communication
Systems, University of Surrey, Guildford, U.K. He
is currently a Research Fellow with the School
of Information Engineering, Zhengzhou University,
Zhengzhou, China. His research interests include
millimeter wave, edge caching, NOMA, SWIPT, and
cloud radio access networks.
Dr. Hao has served as the IEEE ICC Technical Program Committee Member
in 2019 and 2020.
Zhongyong Wang received the B.S. and M.S.
degrees in automatic control from Harbin
Shipbuilding Engineering Institute, Harbin, China,
in 1986 and 1988, respectively, and the Ph.D.
degree in automatic control theory and application
from Xi’an Jiaotong University, Xi’an, China, in
1998.
Since 1988, he has been a Lecturer with the
Department of Electronics, Zhengzhou University,
Zhengzhou, China, where he was an Associate
Professor from 1999 to 2002 and promoted to a
Professor with the Department of Communication Engineering in 2002. His
general fields of interests cover numerous aspects within embedded systems,
signal processing, and communication theory.
Inkyu Lee (Fellow, IEEE) received the B.S. degree
(Hons.) in control and instrumentation engineering
from Seoul National University, Seoul, South Korea,
in 1990, and the M.S. and Ph.D. degrees in electri-
cal engineering from Stanford University, Stanford,
CA, USA, in 1992 and 1995, respectively.
From 1995 to 2001, he was a member of
the Technical Staff with Bell Laboratories, Lucent
Technologies, Murray Hill, NJ, USA, where he stud-
ied high-speed wireless system designs. From 2001
to 2002, he was a Distinguished Member of the
Technical Staff with Agere Systems (formerly, the Microelectronics Group,
Lucent Technologies), Murray Hill. Since 2002, he has been with Korea
University, Seoul, where he is currently a Department Head of the School
of Electrical Engineering. In 2009, he was a Visiting Professor with the
University of Southern California, Los Angeles, CA, USA. He has authored or
coauthored more than 180 journal papers in IEEE publications and has 30 U.S.
patents granted or pending. His research interests include digital communi-
cations, signal processing, and coding techniques applied for next-generation
wireless systems.
Dr. Lee was a recipient of the IT Young Engineer Award at the IEEE/IEEK
Joint Award in 2006, the Best Paper Award at the Asia–Pacific Conference
on Communications in 2006, the IEEE Vehicular Technology Conference in
2009, the IEEE International Symposium on Intelligent Signal Processing
and Communication Systems in 2013, the Best Research Award from the
Korean Institute of Communications and Information Sciences in 2011, the
Best Young Engineer Award from the National Academy of Engineering of
Korea in 2013, and the Korea Engineering Award from the National Research
Foundation of Korea in 2017. He has served as an Associate Editor for
the IEEE T RANSACTIONS ON C OMMUNICATIONS from 2001 to 2011 and
the IEEE T RANSACTIONS ON W IRELESS C OMMUNICATIONS from 2007 to
2011. In addition, he was a Chief Guest Editor for the IEEE J OURNAL ON
S ELECTED A REAS IN C OMMUNICATIONS (Special Issue on 4G Wireless
Systems) in 2006. He currently serves as the Co-Editor-in-Chief for the
Journal of Communications and Networks. He has been elected as a member
of the National Academy of Engineering of Korea in 2015, and is an IEEE
Distinguished Lecturer.