Int. J. Electron. Commun.

(AEÜ) 128 (2021) 153495

Contents lists available at ScienceDirect

International Journal of Electronics and Communications

journal homepage: www.elsevier.com/locate/aeue

Regular paper

Secure cooperative spectrum sharing in full-duplex multi-antenna cognitive

radio networks with jamming
Zahra Mobini
Faculty of Engineering, Shahrekord University, Shahrekord 115, Iran

A R T I C L E I N F O A B S T R A C T

Keywords: We consider a secure overlay cognitive radio network with an eavesdropper wherein a multi-antenna secondary
Average secrecy rate transmitter performs transmission in primary spectrum, on the condition that it helps primary system to perform
Secrecy outage probability secure and reliable transmission via cooperative relaying and jamming. To improve secrecy performance of
Jamming
primary network, we propose full-duplex jammer protocol and zero-forcing based beamforming design, which
Full-duplex (FD)
completely cancels the interferences at the primary and secondary users and simultaneously avoids the leakage of
Beamforming
confidential information to eavesdropper. Moreover, we present new expressions for the average secrecy rate and
a lower bound for the secrecy outage probability. Furthermore, an asymptotic analysis in the high signal-to-noise
ratio regime is carried out to obtain closed-form average secrecy rate. Our analytical findings reveal that by
exploiting beamforming and full-duplex at the secondary transmitter secrecy outage probability can be signifi­
cantly reduced and a diversity order of min (NR − 1, NT − 2) can be achieved where NR and NT are the number of
received and transmit antennas at secondary transmitter. Simulation results also demonstrate that as compared
to the half-duplex scenario without jamming, the proposed cooperative FD overlay CR scheme with jamming can
improve the average secrecy rate up to 224%.

1. Introduction exposed to multiple internal and/or external malicious threats which

Cognitive radio (CR) is foreseen as one of the promising technologies
of spectrum-constrained fifth generation (5G) wireless networks [1].
The key idea of CR is to allow licensed users, known as primary users
(PUs), and unlicensed users, known as secondary users (SUs), coexist
and share the same spectrum while the PUs have the higher priorities in
using the spectrum [2]. From different possible implementation spec­
trum sharing strategies, the overlay and underlay methods are the most
popular ones. In the underlay approach, the secondary user is allowed to
use the spectrum of the primary user when the interference from the
secondary user is less than the interference level which the primary user
can tolerate. Hence, the transmission power of the secondary user is
constrained not to exceed the interference level. In the overlay
approach, the secondary user uses the same spectrum concurrently with
the primary user while maintaining or improving the transmission of the
primary user by applying sophisticated signal processing and coding [3].
However, in practice, there are still many challenges ahead for CR
networks, including network coverage and security of the confidential
information signals [4,5]. In particular, due to the open and dynamic
characteristics of spectrum sharing CR networks, legitimate users are
make security issues much more emergent and prominent [6]. Among
four different phases of cognitive cycle, the sensing (observe) and acting
(communication) phases are more important from security view point
since they are most prone to attacks. Cognitive networks has special
challenges in each cycle for underlay and overlay modes. For example
among several, in the underlay mode, to protect the PU from harmful
interference, the communication requirements of SU are limited by the
regulators. One main goal of malicious attackers is trying to make failure
on the CR network by creating a situation not allowed by the regulator
which poses more challenge for security of successful implementation of
CR in underlay mode compared to its overlay counterpart [7]. As
another example, as it will be discussed later in the next section, in the
cooperation based overlay CR systems where SUs act as a relay to for­
ward the PUs’ information signal, the transmission protocol performs
into two transmission phases and eavesdropper can overhear the infor­
mation signal from two phases. Hence, these transmissions are more
vulnerable to the eavesdropping attack than conventional non-
cooperative underlay communications.
To ensure security, wireless physical-layer security methods can be
exploited to notably enhance the secrecy rate which is defined as the

E-mail address: z.mobini@sku.ac.ir.

https://doi.org/10.1016/j.aeue.2020.153495
Received 24 May 2020; Accepted 7 October 2020
Available online 16 October 2020
1434-8411/© 2020 Elsevier GmbH. All rights reserved.
Z. Mobini
difference between the instantaneous rate of the legitimate link and that
of the wiretap link. If the secrecy rate falls below zero, the eavesdropper
can intercept confidential information. To this end some recent efforts
were devoted to improving the wireless secrecy rate by using multiple-
input multiple-output (MIMO) and beamforming [8–12]. The physical
layer security of MIMO underlay cognitive radio systems with multiple-
antennas SUs, PUs, and eavesdropper were studied in [13] wherein the
impact of system parameters, including number of transmit/receive
antennas, channel qualities, and fading parameter on the secrecy per­
formance were thoroughly investigated.
Furthermore, cooperative relaying is emerging as a promising mean
of improving the reliability, capacity, and security of wireless networks
[14,15]. Particularly, the relay nodes may act as conventional cooper­
ative nodes to help the transmitter to send the information signal, or
may act as the jammer by transmitting jamming signal to deteriorate the
received signal of the potential eavesdroppers [11,12,14]. In [14], two
different cooperation strategies, namely relay-jammer and cluster-
beamforming, were proposed. In the former, two individual SUs act as
a relay and a friendly jammer to enhance the PU’s secrecy. In the latter,
PU cooperates with a cluster of SUs to improve the secrecy using
collaborative beamforming. Secure cooperative communications for PUs
in orthogonal frequency-division multiple-access (OFDMA) CR networks
in the presence of a set of passive eavesdroppers have been studied in
[16]. Instantaneous and ergodic resource allocation problems for the
relay-based cooperative CR network to maximize the secrecy rate of the
SU subject to the minimum required PU’s secrecy rate was studied in
[17]. The secrecy performance of dual-hop multi-antenna spectrum
sharing relaying systems under the presence of an eavesdropper has
been investigated in [8]. The authors in [9] presented a downlink
cascaded beamforming scheme to ensure the secure transmission for a
two-cell MIMO CR network. Beamforming optimization for the secure
primary transmission using the multi-antenna secondary user in a CR
network was studied in [10]. In [18] joint secondary user scheduling,
power, and time allocation schemes to maximize the secondary network
ergodic rate under the primary network secrecy constraint for cooper­
ative cognitive wireless powered communication network were inves­
tigated. A secure CR network with cooperative jamming wherein
multiple SUs interfere with multiple eavesdroppers to protect the PU and
gain transmission opportunities were investigated in [19], and the
problem of optimizing the resource allocation for maximizing the SUs
ergodic transmission rate under PU secrecy outage and SUs transmission
power constraints was solved.
On the other hand, full-duplex (FD) technique has recently received
significant research interest, because of its great potential to double the
spectral efficiency of traditional HD relaying by allowing concurrent
transmission and reception in the same frequency band [20]. Self-
interference (SI) problem due to signal leakage from the output of the
transceiver to the input, is considered as one of the major bottleneck in
practical implementation of FD. Nevertheless, many effective and
practical SI suppression methods such as passive SI suppression, analog
and digital baseband cancellation techniques, have been developed
today [21–23]. Specifically, spatial suppression techniques such as null-
space projection [21] and multiple-antenna techniques [24,25] help us
to use FD relays with cooperative relaying and cooperative jamming
[26] in secure wireless networks. However, to the best of our knowledge,
the performance of the spectrum-sharing overlay CR network with
multi-antenna full-duplex relaying and jamming has not been well un­
derstood. Recent work in [27] investigated the dual-hop randomize-and-
forward (RaF) underlay cognitive wiretap networks over Rayleigh
fading channels, in which the RaF relay is considered both as half-duplex
and full-duplex operations. The authors in proposed [28] a collaboration
interference transmission scheme for cognitive full-duplex wireless
wiretap networks by using antenna selection and beamforming technics
to improve the performance of the network. The considered CR system
models and the proposed beamforming designs in [27,28] are
completely different from our paper. Also, [27] impose a simplifying
2
AEUE - International Journal of Electronics and Communications 128 (2021) 153495
assumption that the self-interference (SI) is completely nulled out at the
full-duplex relay. Moreover, the influence of SI is not taken into
consideration in the beamforming design.
Motivated by all above, in this paper we develop a novel cooperative
FD multi-antenna spectrum sharing overlay CR scheme with jamming
that achieves high reliability and also high secrecy performance against
an eavesdropper. More specifically, we focus on the CR communication
scenario with one pair of PUs, one pair of SUs, and one passive eaves­
dropper wherein the SUs perform transmission in the primary spectrum,
on the condition that they help the PUs to perform secure and reliable
transmission. FD multi-antenna secondary transmitter (STx) acts as a
cooperative jammer and cooperative relay in two transmission stages,
respectively. In the stage of cooperative jamming, thanks to FD opera­
tion, the STx receives the information signal of the primary network
while simultaneously sends a jamming signal to confound the eaves­
dropper. In the stage of cooperative relaying, STx superposes its own
information signal over that of the primary transmitter (PTx), then
amplifies and forwarders to both secondary and primary receivers.
Furthermore, at each transmission stage, we design beamforming vec­
tors at the STx that benefit the PUs and/or SUs and hurt the eaves­
dropper. The main contributions of this work are as follows:
• In the first transmission phase, we propose beamforming design at
the STx such that the jamming signal to the eavesdropper and hence
the security level is maximized, while the SI signal at the STx is
completely cancelled. Moreover, in the second transmission phase
beamforming vectors are designed such that not only the interfering
signals from the STx and PTx at the PRx and SRx are respectively
cancelled but also the PTx’s signal relayed by STx is completely
nulled out at the eavesdropper. Accordingly, the proposed scheme is
not interference limited (i.e., there is no interference at the primary
and secondary networks from the STx and PTx transmissions,
respectively) and does not have any primary information leakage in
the relaying path.
• In order to highlight the system behavior and provide important
insights into the performance, closed-form expressions for the
average secrecy rates and secrecy outage probability lower bound
are presented which shows a diversity order of min (NR − 1, NT − 2)
can be achieved where NR and NT are the number of received and
transmit antennas. These results reveal the effects of key system
parameters such as the number of STx antennas and the transmission
powers on the system performance.
• Our findings reveal that the proposed cooperative FD multi-antenna
overlay CR scheme with jamming can achieve, up to 224%(480%)
average secrecy gains compared to its HD counterpart without jam­
ming and with conventional beamforming (with ZF beamforming).
In addition, the additional transmit antenna significantly enhances
the average secrecy performance, while the average secrecy perfor­
mance is less sensitive to the number of receive antennas at the STx.
Notation: We use bold upper case letters to denote matrices, bold
lower case letters to denote vectors. The superscripts (⋅)† and (⋅)− 1 stand
for conjugate transpose, and matrix inverse, respectively; the Euclidean
norm of the vector is denoted by ‖⋅‖; Pr(⋅) denotes the probability; fX (⋅)
and FX (⋅) denote the probability density function (pdf), and cumulative
distribution function (cdf) of the random variable (RV) X, respectively;
and 𝒞𝒩 (0, σ2 ) denotes a zero-mean circularly symmetric complex
Gaussian RV X with variance σ2 . We also use the notation X ∼ χ 22K to
denote a chi-square distributed RV X with 2K degrees-of-freedom. Γ(a) is
the Gamma function [29, Eq. (8.310.1)]; Γ(a, x) and γ(a, x) are upper
and lower incomplete Gamma functions, respectively [29, Eq. (8.350)];
Ei (x) is the exponential integral function [30, Eq. (5.1.2)]; Ψ(a, b, x)
denotes Kummer confluent hypergeometric function [29, Eq. (9.210.1)];
2 F1 (a, b; c; z) is Gauss’ Hypergeometric function [29, Eq. (9.111)]; ψ (⋅) is
the digamma function [29, Eq. (8.360)].
Z. Mobini AEUE - International Journal of Electronics and Communications 128 (2021) 153495

2. System model 𝒞NT ×1 and wtS ∈ 𝒞NT ×1 in the next section. Therefore, the received signal
at the PRx can be written as
We consider a secure cooperative FD multi-antenna CR network, (√̅̅̅ [ ] √̅̅̅̅̅̅̅̅̅̅̅ [ ]) [ ]
consisting of a primary transmitter, PTx, a primary receiver, PRx, a yPR,2 = Gh†SP α wtP GS yST,1 n + 1 − α wtS xS n + nPR,2 n
secondary transmitter base station, STx, a secondary receiver, SRx, and √̅̅̅̅̅̅̅̅ [ ] √̅̅̅̅̅̅̅̅ [ ]
= GGS αPP h†SP wtP w†r hPS xP n + GGS αPS h†SP wtP w†r HSI wt xJ n
one eavsedropper, E, as shown in Fig. 1. In this scenario, the PUs allow
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ [ ] [ ] [ ]
the SUs to access their spectrum bands on the condition that base station √̅̅̅
+G (1 − α)h†SP wtS xS n + GGS αh†SP wtP w†r nST,1 n + nPR,2 n ,
STx has to relay the confidential message of the primary network.
Moreover, the STx operates in the FD mode. Accordingly, it receives the (4)
information signal from PTx and simultanously sends a jamming signal
with
to combat the eavesdropping and ensures secure information trans­
mission of the primary network. For the FD operation, the STx has two 1
GS = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
) , (5)
sets of antennas, i.e., NR receiveing antennas and NT transmitting an­ (
wr PP hPS h†PS + σ2ST I wr
†

tennas. Moreover PTx, PRx, SRx, and E all have a single antenna, and
operate in an HD mode.
and
In order to effectively share the spectrum among the PUs and SUs,
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
the transmission protocol is divided into two phases. In the first phase, √
√ PS
the PTx broadcasts its information signal to STx and PRx. At the same G=√ ( ( ) ), (6)
Trace αwtP w†tP + 1 − α wtS w†tS
time, to avoid the eavesdropper from overhearing the confidential pri­
mary network’s information, STx sends a jamming signal to mislead the
eavesdropper using the transmit beamforming vector wt ∈ 𝒞NT ×1 . where nPR,2 ∼ 𝒞𝒩 (0, σ 2PR ) is the AWGN at PRx in the second phase and α
Accordingly, the received signal at the STx can be written as is the power allocation ratio of the PTx’s signal to the total STx transmit
[ ] √̅̅̅̅̅ [ ] [ ] power PS , 0⩽α ≤ 1. Moreover, the received signals at the SRx and E in
√̅̅̅̅̅̅
yST,1 = PP w†r hPS xP n + PS w†r HSI wt xJ n + w†r nST,1 n , (1) the second phase can be respectively, expressed as
(√̅̅̅̅̅̅̅̅̅̅̅ [ ] √̅̅̅ [ ]) [ ]
ySR = Gh†S 1 − α wtS xS n + α wtP GS yST,1 n + nSR n
where PP denotes the PTx transmit power, xP [n] is the PTx information
{ } √̅̅̅̅̅̅̅̅̅̅̅̅̅̅ [ ] √̅̅̅̅̅̅̅̅ [ ]
symbol with E xP [n]x†P [n] = 1, xJ [n] is the jamming signal satisfying = G (1 − α)h†S wtS xS n + GGS αPP h†S wtP w†r hPS xP n
{ } [ ] [ ] [ ]
E xJ [n]x†J [n] = 1, wr ∈ 𝒞NR ×1 is the receive beamformer, hPS ∈ 𝒞NR ×1 is √̅̅̅̅̅̅̅̅ √̅̅̅
+GGS αPS h†S wtP w†r HSI wt xJ n + GGS αh†S wtP w†r nST,1 n + nSR n ,
the channel between the PTx and STx and its entries follow i.i.d., 𝒞𝒩 (0,
λps ), HSI denotes the NR × NT residual SI channel which is modeled as (7)
identically independent distributed (i.i.d) 𝒞𝒩 (0, σ 2RR ) RVs [31,21], and
and
nST,1 [n] denotes the additive white Gaussian noise (AWGN) at the STx (√̅̅̅ [ ] √̅̅̅̅̅̅̅̅̅̅̅ [ ]) [ ]
{ }
with E nST,1 n†ST,1 = σ 2ST I. Moreover, the received signals at the PRx yE,2 = Gh†SE α wtP GS yST,1 n + 1 − α wtS xS n + nPR,2 n
√̅̅̅̅̅̅̅̅ [ ] √̅̅̅̅̅̅̅̅ [ ]
during the first phase can be expressed as = GGS αPP h†SE wtP w†r hPS xP n + GGS αPS h†SE wtP w†r HSI wt xJ n
√̅̅̅̅̅̅ [ ] √̅̅̅̅̅ [ ] [ ] √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ [ ] √̅̅̅ [ ] [ ]
yPR,1 = PP hP xP n + PS h†SP xJ n + nPR,1 n , (2) +G (1 − α)h†SE wtS xS n + GGS αh†SE wtP w†r nST,1 n + nE,2 n ,
(8)
where hP ∼ 𝒞𝒩 (0, 1) denotes the channel coefficient of the PTx-PRx
link, hSP ∈ 𝒞NT ×1 is the channel coefficient of the link between STx where hS ∈ 𝒞NT ×1 is the channel coefficients of the links from STx to SRx,
and PTx, and nPR,1 ∼ 𝒞𝒩 (0, σ2PR ) is the AWGN at the PTx. On the other nSR ∼ 𝒞𝒩 (0, σ2SR ) and nE,1 ∼ 𝒞𝒩 (0, σ 2E ) denote the AWGN at the SRx and
hand, the received signal at E in the first phase is given by E, respectively.
√̅̅̅̅̅̅ [ ] √̅̅̅̅̅ [ ] [ ] The channel coefficients hP and hPE corresponding to the PTx-PRx
yE,1 = PP hPE xP n + PS h†SE wt xJ n + nE,1 n , (3) link and PTx-E link are assumed to be i.i.d. complex Gaussian RVs
with zero-mean and variance ΩP and ΩPE , respectively. The entries of
where hPE ∼ 𝒞𝒩 (0, 1) and hSE ∈ 𝒞NT ×1 denote the channel coefficients of hPS , hSE , hSP and hS are i.i.d. complex Gaussian RVs with zero-mean and
the links between PTx and E and between STx and E, respectively, and variance ΩPS , ΩSE , ΩSP and ΩS , respectively.
nE,1 ∼ 𝒞𝒩 (0, σ2E ) is the AWGN at the E. The PRx uses the maximal-ratio combining (MRC) method to the
During the second phase, the PTx remains silent and the STx multi­ received signals in (2) and in (4) form the first and second transmission
plies its received signal from the first phase, i.e., yST,1 , by the normali­ phases, respectively. Further, we assume that the jamming signal in our
zation constant GS . Then, it amplifies the resulted normalized signal system is known a priori at the PRx and hence, PRx can eliminate it from
with the amplification gain G and broadcasts the superposition of this the received signal in the first phase, yPR,1 . It is worth to mention that
signal and its own non-confidential information signal to the PRx and this is a widely adopted assumption in the physical-layer security with
STx. To this end, STx employs two different transmit beamforming jammer [32–34], wherein the jamming signals are produced by using
vectors denoted by wtP and wtS to the normalized information PTx’s the pseudo-random codes which are not known at the eavesdroppers but
signal and STx’s signal, respectively. We will discuss the design of wtP ∈ available at the legitimate users and hence can be effectively removed.
Accordingly, the received signal-to-interference-plus-noise ratio (SINR)

3
Z. Mobini AEUE - International Journal of Electronics and Communications 128 (2021) 153495

at the PRx can be written as notable that the suboptimal ZF method has known in the literature [24]
as a practical and simple scheme which has a lower complexity in

⃒ ⃒
PP ⃒hP |2 G2 G2S αPP ⃒h†SP wtP w†r hPS |2
γ PR = + ( )⃒
⃒
⃒
⃒
⃒
⃒ , (9)
σ2PR G2 1 − α ⃒h†SP wtS |2 + G2 G2S ασ 2ST ⃒h†SP wtP w†r nST,1 |2 + G2 G2S αPS ⃒h†SP wtP w†r HSI wt |2 + σ 2PR

and the received SINR at the SRx can be written as comparison with other interference cancelation methods. The signifi­
cant performance improvements of the proposed beamforming designs

( )⃒
⃒
G2 1 − α ⃒h†S wtS |2
γ SR = ⃒ ⃒ ⃒ (10)
G2 G2S αPP ⃒h†S wtP w†r hPS |2 + G2 G2S ασ 2ST ⃒h†S wtP w†r nST,1 |2 + G2 G2S αPS ⃒h†S wtP w†r HSI wt |2 + σ 2SR

will be shown in Section 4. In the first transmission phase, the primary

Finally, the received SINR at the E in the first and second trans­ signal is transmitted from the PTx to the PRx and STx which can be
mission phases can be respectively written as eavesdropped by E. At the same time the FD STx transmits the jamming
⃒ signal to disrupt the eavesdropping while receiving the PTx’s signal.
PP ⃒hPE |2 Hence, to maximize the received SINR at STx, we fix the MRC beam­
γ E,1 = ⃒ † , (11)
P S ⃒ h w t |2 + σ 2 E forming vector as wMRC = ‖hhPS at the STx. Further, we mitigate the
r
SE
PS ‖

harmful SI by projecting the STx transmit signal to the null space of the
and

⃒
G2 G2S αPP ⃒h†SE wtP w†r hPS |2
γ E,2 = ( )⃒
⃒
⃒
⃒
⃒
⃒ . (12)
G2 1 − α ⃒h†SE wtS |2 + G2 G2S ασ 2ST ⃒h†SE wtP w†r nST,1 |2 + G2 G2S αPS ⃒h†SE wtP w†r HSI wt |2 + σ 2E

received signal at the STx 1 input [24]. Hence, the optimal transmit
beamforming vector wt which minimizes the received SINR at E is ob­
Considering the MRC, the overall SINR at E is given by tained by solving the following problem:
⃒ ⃒ ⃒
PP ⃒hPE |2 max ⃒h†SE wt ⃒
γE = ⃒ † (13) ‖wt ‖=1
(14)
PS ⃒h wt |2 + σ 2
SE E
s.t. h†PS HSI wt = 0.

⃒
G2 G2S αPP ⃒h†SE wtP w†r hPS |2
+ ( )⃒ ⃒ ⃒ .
⃒ ⃒ ⃒
G2 1 − α ⃒h†SE wtS |2 + G2 G2S ασ 2ST ⃒h†SE wtP w†r nST,1 |2 + G2 G2S αPS ⃒h†SE wtP w†r HSI wt |2 + σ 2E

The optimum transmit beamforming vector wt from (14) is derived as

H†SI hPS h†PS HSI
[24], wZF
t
AhSE
= ‖Ah , where A = INT − .
SE ‖ ‖h†PS HSI ‖2
In the second transmission phase, the STx amplifies the received
3. Beamforming design signal in the first transmission phase and broadcasts the superposition of
this signal and its own information signal to the PRx and STx which also
From (9), (10), and (13) we observe that the received SNR/SINRs at can be overheard by E. In this phase, we choose the beamforming vector
the PRx, STx, SRx, and E are the functions of receive and/or transmit wtP to lie in the null space of the equivalent channel of STx to SRx and
beamforming vectors in the first and second transmission phases.
Therefore, to have an effective spectral sharing among the PUs and the
SUs as well as to cancel the SI effect and further enhance the security of
the primary network, in the sequel we present suboptimal receive and 1
To employ ZF method, we make the common assumption that the STx has
transmit beamforming design based on zero forcing (ZF) principle. It is NT > 1 transmit antennas.

4
Z. Mobini AEUE - International Journal of Electronics and Communications 128 (2021) 153495

Fig. 1. System model for the proposed cooperative FD multi-antenna CR with jamming.

STx to E. That is HS wtP = 0, where HS = [h†S ; h†SE ] is the 2 × NT equiva­ the overheard SINR at the eavesdropper E (from the first transmission
lent channel matrix. In this case, the interference caused by the PTx’s phase) can be expressed by
signal to the SRx is cancelled out. Also, the PTx’s information signal ⃦
PS (1 − α)⃦
relayed by STx is completely nulled out at the E and hence there will be γ SR = ⃦h† wt ‖2
⃦ S S (19)
σ2
no PUs information leakage to E. Accordingly, the optimal wtP which
maximizes the received SINR at the PRx can be written as and
⃒ † ⃒ ⃒
max ⃒h w t ⃒,
SP P PP ⃒hPE |2
‖ tP ‖
w =1 (15) γE = ⃦ 2
, (20)
PS ⃦hSE wZF
t ‖ +σ
2
s.t. HS wtP = 0.
⃦ 2
Based on projection matrix theory [35], the solution of the optimization where ⃦hSE wZF
t ‖ can be further simplified as [36]

problem (15) can be written as wZF tP = ‖ChSP ‖, where C =

ChSP ⃦
⃦
⃦
⃦̃ 2
⃦hSE wZF
2 † ⊥ ̂† ̂ (21)
− 1 t ‖ = hSE Δ hSE = h SE diag(0, 1, ⋯, 1) h SE = ⃦hSE ‖ ,
INT − H†S (HS H†S ) HS .
In addition, based on the ZF criterion wtS is designed such that the where ̂h SE = ΦhSE with Φ is an unitary matrix, and h ̃SE is a (NT − 1) × 1
interference caused by the signals of STx to the PRx is suppressed.
vector, consisting of the (NT − 1) last elements of h
̂ SE .
Hence, the optimal wtS which maximizes the received SINR at the SRx
can be expressed as Remark 1. From (21) we observe that the received SNR at the SRx for
⃒ † ⃒ the proposed overlay CR network with ZF beamforming does not have
max ⃒h w t ⃒,
S S any interference term due to PTx’s transmission, and hence will poten­
‖wtS ‖=1 (16)
tially lead to a better performance for the SUs network in compare with
s.t. h†SP wtS = 0.
conventional overlay CR networks.

The solution of the optimization problem (16) can be derived as wZF

tS =
It is notable that in the above beamforming designs similar to
− 1 † [37,38] we assume that the channel state information (CSI) of the STx to
‖DhSE ‖, where D = INT − hSP (hSP hSP ) hSP .
DhSE †
E link is available.2 This assumption is valid for the scenarios wherein
By substituting wMRC
r t ,wtP , and wtS
,wZF ZF
into (9), after some algebraic
ZF
the eavesdropper is one of the legitimate users and is trusted on service
manipulation, the received SINR at the PRx can be expressed as level and performs true CSI feedback. However, it is data-level malicious
and acts as a passive eavesdropper to intercept the primary network
γ PR = γ0 + γ R (17)
confidential information for its own purpose [40]. Further, the
⃒ assumption of perfect knowledge of the eavesdropper’s CSI at the STx
where γ 0 = ρp ⃒hP |2 with ρp = PσP2 and
⃒
enables us to develop fundamental understanding of how jamming and
γ1 γ2
γR = , (18)
γ1 + γ2 + 1
2
⃦ ⃦ When the explicit cooperation between the legitimate nodes and the
with γ1 = ρp ⃦hPS ‖2 , and γ2 = ρs α⃦ChSP ‖2 with ρs = Pσ2S .
eavesdropper is not available, perfect knowledge of the eavesdropped’s chan­
In addition, upon substituting wMRC
r t , wtP , and wtS into (10) and
, wZF ZF ZF
nels is difficult to obtain at the legitimate nodes. For these scenarios the robust
(13) and after some algebraic manipulation, the received SNR at SRx and secure designs can be applied to ensure achieving the security and robustness
[39].

5
Z. Mobini AEUE - International Journal of Electronics and Communications 128 (2021) 153495

beamforming can enhance security in cooperative FD multi-antenna CR

setting by characterizing secrecy rate performance. Nevertheless, for the
Proof. See Appendix A. □
case when CSI of the E is unavailable, we have the following Remark.
We now present the cdf of the received SINR at the E, FγE (x), in the
Remark 2. When the CSI of channels related to E is unavailable, in the
following lemma.
first transmission phase wr and wt can be chosen based on MRC and
random beamforming designs, respectively. In addition, in the second Lemma 2. The cdf of the received SINR at the E, γ E , is given by
transmission phase, the transmit beamforming vectors wtP and wtS can be ⎛ ⎞
( )− (NT − 1)
designed using ZF principles. In particular, wtP is chosen such that γ
(26)
− x
⃒ † ⃒ ⃒ ⃒ FγE ⎝x⎠ = 1 − e γ4 1 + 3 x ,
⃒h wt ⃒ is maximized subject to ⃒h† wt ⃒ = 0. Similarly, wt is chosen γ4
SP P S P S
⃒ † ⃒ ⃒ ⃒
such that ⃒hS wtS ⃒ is maximized subject to ⃒h†SP wtS ⃒ = 0. The mathemat­
ical analysis and derivations for these beamforming designs are where γ 3 = ρs ΩSE and γ4 = ρp ΩPE . Moreover, the pdf of γE is given by
straightforward based on the derived results in this paper and is omitted ⎛ ⎞
to avoid clutter. ( )NT − 1 − γx ( )− (NT − 1)
[( ) ( )− 1 ]
⎜ ⎟ γ4 e 4 γ4 γ4
fγE ⎜ ⎟
⎝x⎠ = +x NT − 1 γ4 +x +1 .
γ3 γ4 γ3 γ3
4. Performance analysis

(27)
In this section, we investigate the secrecy performance of the primary
network for the proposed cooperative FD multi-antenna CR system with
jamming and ZF beamforming in terms of two important secrecy criteria
namely average secrecy rate and secrecy outage probability. The derived Proof. See Appendix B. □
results will highlight the behavior of the system and provide important
insights into the performance. 4.2. Secrecy rate

4.1. Preliminaries To evaluate the security, instantaneous secrecy rate is one of the
important secrecy performance criterion defined as [42]
In this subsection, the pdf and the cdf of the SINRs of the main and Cs = ⌊CP − CE ⌋+ , (28)
the eavesdropper’s channels are derived, which will facilitate the
ensuing secrecy analysis. where ⌊x⌋+ = max(x, 0). Also, CP and CE are the overall rate at the
Let us derive the cdf of the received SINR at the PRx, γPR , where primary network and eavesdropping over the two transmission phases
γ PR = γ0 + γR . We note that γ0 is an exponential RV and its cdf is given ) )
by and given by CP = 12 log(1 +γPR and CE = 12 log(1 + γ E , respectively.
⎛ ⎞ Therefore, PTx can transmit confidential messages to the PRx at a rate Cs
to guarantee perfect secrecy. In the delay tolerant transmissions, how­
x
(22)
−
Fγ0 ⎝x⎠ = 1 − e γ0 , x⩾0
ever, the codeword length is large enough to experience all possible
realizations of the channels. As such, average secrecy rate is an appro­
where γ 0 = PPσΩ2 P . Moreover, the cdf of γR can be found in [41, Proposition priate performance criterion which is defined as the instantaneous se­
1]. crecy rate, Cs , averaged over γ PR and γE and mathematically can be
Although, we have the distribution of γ0 and γ R , finding the exact expressed as [43,42]
distribution of γPR seems difficult to obtain, since the distribution of ∫ ∫
1 ∞ ∞
( ) ( )
γ 0 +γR is intractable. To overcome this issue, we first apply a widely used Cs = Cs fγPR x1 fγE x2 dx1 dx2 . (29)
2 0
tight upper bound to γ R as γR ⩽γRL = min(γ1 , γ 2 ). Therefore, γ PR can be
0

upper bounded as γ PR ⩽γPRL = γ0 + γRL . We have the following key result The average secrecy rate can be rewritten as [44]
for the cdf of γPRL . ∫ (∫ ∞ [ (
1 ∞ 1 + x1
) ]+ ( ) ) ( )
Cs = log fγE x2 dx2 fγPR x1 dx1
Lemma 1. The cdf of γPRL is given by 2 0 0 1 + x2
∫ ∞ ( ∫ x1 ( ) ( ) ) ( ) (30)
( ) ( ) ( ) 1 + x1
1 ∑
NT− 2
1 ∑
N R− 1 (a)
= log fγ x2 dx2 fγPR x1 dx1 ,
FγPRL x = NR ( ) Ψ ηk , γ 2 + NT − 1 ( ) Ψ θk , γ 1 , 0 0 1 + x2 E
γ 1 Γ NR k=0 γ2 Γ NT − 1 k=0
(23) where (a) follows from the definition of Cs and the condition Cs ⩾0, i.e.,
( )
where ηk = NR +k, θk = NT +k − 1, and log 1+x
1+x2 ⩾0, which implies that x2 ⩽x1 .
1

⎛ ⎞
Using the similar steps as in [45], the average secrecy rate in (30) can
γ(λ, μ1 x) − x γ(λ, μ2 x)
Ψ⎝λ, γ⎠ = − e γ0 (24) be re-expressed as
k!γ k μλ1 k!γk μλ2 [ ( )]
∫∞
1 FγE (x)
Cs = 1 − FγPR x dx. (31)
PS 2ln(2) 0 1 + x
with μ1 = γ1 +γ1 and μ2 = γ1 +γ1 − 1
γ0 with γ1 = PσP2 ΩPS , and γ 2 = σ2 αΩSP .
1 2 1 2

Moreover, the pdf of γPRL is given by From (31) we observe that Cs depends on the cdf of γ PR and γE .
⎛ ⎞ Therefore, by substituting (23) and (26) into (31), the exact average
x ( ) secrecy rate can be derived in integral form. We notice that an exact
⎜ ⎟ e− γ 0 1 ∑
N T− 2
γ(ηk , μ2 x) 1 ∑
NR− 1
γ(θk , μ2 x)
⎜ ⎟
fγPRL ⎝x⎠ = ( ) + ( ) . evaluation of Cs is tedious, if not impossible, to obtain in closed-form.
γ 0 γ N1 R Γ NR k=0 k!γ k2 μη2k γN2 T − 1 Γ NT − 1 k=0 k!γ k1 μθ2k
However, the result can be efficiently calculated numerically using
Matlab or Mathematica.
(25)
In order to explicitly examine the performance in the high SNR
regime, we proceed to derive the asymptotic average secrecy rate.

6
Z. Mobini AEUE - International Journal of Electronics and Communications 128 (2021) 153495

Specifically, in the asymptotic scenario, we assume that both the PTx

and STX have the large enough transmit power (i.e., ρp →∞ and ρs →∞).
Proposition 1. When ρp →∞ and ρs →∞ and assuming that ρs
ρp
asymptotic average secrecy rate of the system is given by
Proof. The proof is straightforward using definition of high SNR slope
= κ, the and the high SNR power offset in (34) and (35), respectively, and is thus
omitted. □

( ) ( )
∑ η +n
∞ 1 1 N T− 2
1 ∑ ∞
( − 1)n a0k Γ(ηk + n)
Cs = log2 a0 ρp + ( ) ψ ηk + n
2ln(2) aN1 R Γ NR k=0 k!ak2 n=0 n!bn2
( )
∑R− 1
1 1 N
1 ∑ ∞
( − 1)n a0θk +n Γ(θk + n) (32)
+ ( ) ψ θ k + n
2ln(2) aN2 T − 1 Γ NT − 1 k=0 k!ak1 n=0 n!bn2
( )
1 a3
+ F1 NT − 1, 1; NT , 1 − ,
2(NT − 1)ln(2)2 a4

From (36), we conclude that the number of receive and transmit

where a0 = ΩP , a1 = ΩPS , a2 = καΩSP , a3 = κΩSE , a4 = ΩPE . antennas at the FD STx and eavesdropper’s channel have no impact on
the high SNR slope. Moreover, by invoking (37), we find that the high
Proof. See Appendix C. □
SNR power offset is independent of ρp . Furthermore, the contribution of
We would like to emphasize that our exact expression in (32) is given the eavesdropper’s channel to ℒ∞ is characterized by Ξ∞ 2 . Specifically,
in closed-form as it involves finite summations of exponentials, Gamma Ξ∞2 increases with NT , and as such the average secrecy capacity
functions, power values, and standard exponential integral functions. increases.
Proposition 1 clearly shows that the average secrecy rate with the
proposed beamforming design is independent of the SI strength. Now, to
obtain additional insights on the secrecy performance, based on (32), we 4.3. Secrecy outage probability
derive two key performance indicators that determine the average se­
crecy rate at high SNR, namely the high SNR slope and the high SNR The secrecy outage probability is defined as the probability of the
power offset [46–48]. The asymptotic average secrecy rate in (32) can achievable secrecy capacity, Cs , being lower than a predetermined se­
be conveniently reexpressed as [47] crecy rate, Rs . Mathematically, it can be represented as [42]
( ( ) ) ( )
(33)
∞ Pout = Pr(Cs < Rs )
Cs = 𝒮 ∞ log2 ρp − ℒ∞ + o 1 . ∫∞ ( ( ) ) ( )
(38)
= FγPR r 1 + x − 1 fγE x dx,
Here, the two key parameters are 𝒮 ∞ , which denotes the high-SNR 0
slope in bits/s/Hz/(3 dB) given by
∞ where r = 22Rs . Substituting (23) and (27) into (38), the secrecy outage
Cs probability of the primary network can be found. We highlight that
𝒮 ∞ = lim ( ) (34)
ρp →∞ log ρ
2 p although (38) does not seem to admit a closed-form solution, it can be
efficiently evaluated numerically. In the sequel, we derive the lower
and ℒ∞ , which represents the zero-order term or high-SNR power offset bound of secrecy outage probability
in 3 dB units given by
( ( ) ) Proposition 2. The secrecy outage probability of the system with proposed
beamforming scheme can be lower bounded as
∞
Cs
ℒ∞ = lim log2 ρp − . (35)
ρp →∞ 𝒮∞
( ) 1 ∑
N T− 2
1
PLout r = ( ) (ℋ(ηk , μ1 , ζ1 ) − ℋ(ηk , μ2 , ζ2 ))
We have the following key results. γN1 R Γ NR k=0 k!γ k2
Corollary 3. The high SNR slope and the high SNR power offset of the
1 ∑
N R− 1
1
proposed beamforming scheme are given by + NT − 1 ( ) (ℋ(θk , μ1 , ζ1 ) − ℋ(θk , μ2 , ζ2 )),
γ2 Γ NT − 1 k=0 k!γk1
𝒮∞ = 1 (36)
(39)
and
where ζ1 = γ1 , ζ2 = γ1 + γr , and
4 4 0

( ) ( )
∑ η +n
1 1 N T− 2
1 ∑∞
( − 1)n a0k Γ(ηk + n)
ℒ∞ = − log2 a0 − ( ) ψ η + n
2ln(2) aN1 R Γ NR k=0 k!ak2 n=0 n!bn2 k

( ) (37)
∑R− 1
1 1 N
1 ∑ ∞
( − 1)n a0θk +n Γ(θk + n)
− ( ) ψ θk + n − Ξ∞
2 ,
2ln(2) aN2 T − 1 Γ NT − 1 k=0 k!ak1 n=0 n!bn2

7
Z. Mobini AEUE - International Journal of Electronics and Communications 128 (2021) 153495

[ ( )
λ− 1 (
∑ ) (
)m (( ( ))
1 γ4 γ
ℋ(λ, μ, ζ) = λ
1− Γ λ rμNT − 1 Ψ m + 1, m + 2 − NT ; 4 ζ + rμ
μ γ3 γ3
(40)
m=0
( ( )) ) ]
1 γ
+ Ψ m + 1, m + 3 − NT ; 4 ζ + rμ .
γ3 γ3

mode and cannot send any jamming signal to E in the first transmission
phase due to its HD nature. The HD-ZF and HD-MRC/MRT schemes are
Proof. See Appendix D. □
outlined as follows. In the HD-ZF scheme during the first time slot, the
From Proposition 2 we observe that the secrecy outage probability PTx transmits its signal to the PRx and STx where STx utilizes the MRC
for CR system shows an outage floor at high values of PTx’s transmission linear processing scheme at its receiver side. In the second time slot, the
power. This is expected because with high values of PTx transmit power STx superposes its own information signal over that of the PTx, then
the overheard signal at eavesdropper from the direct link will be
maximal which reduces the secrecy outage performance.
Remark 3. By inspecting (39), we observe that in the high-SNR 6
FDJ-ZF
regime, assuming that γ 1 →∞, γ2 = κγ1 , γ0 = μγ1 when γ4
γ3 is fixed, the FDJ-ZF, unknown EAVs CSI
proposed cooperative FD multi-antenna CR scheme achieves a diversity 5 HD-ZF
order of min(NR − 1, NT − 2). HD-MRC/MRT

Average secrecy rate (bps/Hz)

=30(dB)
s
4 =15(dB)
5. Numerical results and discussion s

Here, numerical results are presented to demonstrate the perfor­

mance of the proposed secure cooperative FD multi-antenna CR scheme 3
with jamming and ZF beamforming, which is called FDJ-ZF, in the
presence of an eavesdropper, highlight the impact of key system pa­
rameters on its performance and also validate the derived analytical 2
expressions. Unless otherwise stated, the noise variances at E, PRx, and
SRx are set to 1, α = 0.5, ΩP = 0.25, ΩP E = 0.5 ΩPS = 0.5, ΩSE = 0.5,
1
ΩSP = 0.5 and ΩS = 0.5.

5.1. Benchmarks 0
0 5 10 15 20 25 30
To illustrate the secrecy advantages of our proposed FDJ-ZF scheme, p
(dB)
we consider two common cooperative overlay schemes adopted in the
literature [3] called HD-ZF and HD-MRC/maximum ratio transmission Fig. 3. Average secrecy rate of FDJ-ZF, HD-ZF, and HD-MRC/MRT schemes
versus PP for two values of PS (NR = 4 and NT = 6).
(MRT) as the benchmarks. In these schemes STx operates in the HD

3 3
FDJ-ZF FDJ-ZF
HD-ZF HD-ZF
2.5 HD-MRC/MRT 2.5 [NR =5, NT=5]
Average secrecy rate (bps/Hz)

Average secrecy rate (bps/Hz)

[NR =5, NT=3]

[N =3, N =5]
R T
2 2

1.5 1.5

1 1

0.5 0.5

0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
s
(dB) s
(dB)

Fig. 2. Average secrecy rate of FDJ-ZF, HD-ZF, and HD-MRC/MRT schemes Fig. 4. Average secrecy rate of FDJ-ZF, HD-ZF, and HD-MRC/MRT schemes
versus PS (NR = 4, NT = 6, and PP = 10 dB). versus PS for different antenna configurations at STx (PP = 10 dB).

8
Z. Mobini AEUE - International Journal of Electronics and Communications 128 (2021) 153495

5
and
NR =8, NT=4 ⃒ ⃒
4.5 ⃒ † HD ⃒
NR =4, NT=4 ⃦ max⃦ ⃒hS wtS ⃒
⃦wHD ⃦=1
NR =4, NT=3 t (42)
4
S
Average secrecy rate (bps/Hz)

N =4, N =2 s.t. h†SP wHD

tS = 0,
R T
3.5
Asymptotic
respectively. Using projection matrix theory [35], the weights which
3 satisfy the conditions in (41) and (42), are given by

2.5 Ξ⊥
P hSP Ξ⊥
S hS
wHD
tP = ⊥ and wHD
tS = , (43)
‖ΞP hSP ‖ ‖Ξ⊥
S hS ‖
2
− 1 − 1
1.5 where Ξ⊥ P = INT − hS (hS hS ) hS and ΞS = INT − hSP (hSP hSP ) hSP . Note
† † ⊥ † †

that in contrast to the FDJ-ZF scheme, the relayed PTx’s information

1 signal can be overheard by the E in the HD-ZF scheme.
The transmission protocol in the HD-MRC/MRT scheme is the same
0.5 as in the HD-ZF scheme, except that the transmit beamforming vectors in
the second transmission phase are designed based on the MRT scheme, i.
0
0 5 10 15 20 25 30 e., wHD
tP = ‖hSP ‖ and wtS = ‖hS ‖. It is notable that MRT transmit beam­
hSP HD hS

s
= p
(dB) forming in the second transmission phase does not allow interference-
free transmission for the primary and secondary systems since cancel­
Fig. 5. Average secrecy rate of FDJ-ZF scheme versus ρs = ρp for different ling the interferences from the PTx’s and STx’s transmissions on the PRx
antenna configurations. and SRx, respectively, are not possible.
Fig. 2 illustrates the average secrecy rate of FDJ-ZF, HD-ZF, and HD-
MRC/MRT schemes versus STx transmission power, ρs , for NR = 4,NT =
100 6, and ρp = 10 dB. We observe that the proposed FDJ-ZF outperforms all
[N =3, N =3] other schemes for all values of the ρs . For example, at ρs = 10 dB, FDJ-
R T
HD can achieve 224% and 480% average secrecy gains compared to HD-
[NR =4, NT=4]
ZF and HD-MRC/MRT schemes, respectively. This is intuitive since in
Secrecy outage probability

[NR =5, NT=5] the proposed FDJ-ZF the jamming signal transmitted from the FD STx
-2 Exact, Simulation degrades the quality of the eavesdropping channel which accordingly
10 enhances the secrecy rate. Also, FDJ-ZF completely avoids information
Lower bound, Analytical
leakage to E in the second phase. In addition, the average secrecy rate of
FDJ-ZF, HD-ZFJ and HD-MRC/MRT schemes converge to finite limits at
high ρs . More specifically, with high ρs , the FDJ-ZF almost attains the
average secrecy rate of 2.6 bps/Hz, which is about 1.6 times than that of
10-4 HD-ZF and 6 times than that of HD-MRC/MRT.
Fig. 3 compares the average secrecy rate of FDJ-ZF, HD-ZF, and HD-
MRC/MRT schemes versus ρp for two different values of ρs with NR = 4
and NT = 6. The average secrecy rate of the FDJ-ZF scheme improves
with increasing ρp . In the HD-ZF and HD-MRC/MRT schemes however,
the secrecy rate first increases with the ρp , and then decreases when ρp
10-6
0 10 20 30 40 increases beyond a certain value. The main reason is that large trans­
mission power ρp increases the received SINR at the PRx and STx which
s
= p
(dB)
accordingly enhances the secrecy rate. However, it also improves the
achievable rate of the eavesdropper’s channel (in contrast to FDJ-ZF, in
Fig. 6. Secrecy outage probability of FDJ-ZF scheme versus ρs = ρp for different
the HD-ZF and HD-MRC/MRT schemes, the STx cannnot send a jamming
antenna configurations at STx (Rs = 1 bps/Hz).
signal to interfere with the reception of E due to its HD operation).
Morover, we see that the performance gaps between the FDJ-ZF and two
amplifies and forwards to both SUs and PUs networks with transmit
other HD schemes increases with increasing ρs . This is expected because
tP and wtS , respectively. The transmit weight
beamforming vectors wHD HD
increasing ρs improves the received SINR at the PRx from relaying path
vectors wHDtP and wtS are designed such that the interference due to PTx’s
HD
and also increases the jamming signal strength which accordingly en­
signal at the SRx and the interference due to STx’s signal at the PRx are hances the secrecy rate. Fig. 3 also shows that the analytical result for
mitigated. In particular, according to ZF principles, the transmit weight average secrecy rate tightly matches simulation result. In Fig. 3 we also
vector wHDtP is chosen to lie in the orthogonal space of hS such that present the result of FDJ-ZF for the case when CSI of the eavesdropper is
⃒ ⃒ unavailable and beamforming vectors are designed based on Remark 2.
h†S wHD
tP = 0 and ⃒h†SP wHD
tP
⃒ is maximized. Similarly, wHD is chosen in the
t
⃒ † S HD ⃒ It is observed that with unknown eavesdroppe’s CSI the PU secrecy
orthogonal space of hSP such that h†SP wHD
tS = 0 and hS wtS
⃒ ⃒ is maximized.
performance is degraded compared with the perfect case. But, FDJ-ZF,
The problems of designing the transmit weights, wHD
tP and wtS at the STx
HD
unknowen eavesdropper’s CSI scheme still significantly outperforms
can thus be formulated as HD-ZF and HD-MRC/MRT schemes with perfect CSI.
⃒ † HD ⃒
⃒h w ⃒ Fig. 4 shows the average secrecy rate of FDJ-ZF, HD-ZF, and HD-
max SP tP
MRC/MRT schemes with different antenna configurations. For the
‖wtP ‖=1 (41)
HD

FDJ-ZF scheme, we see that the additional transmit antenna could

s.t. h†S wHD
tP = 0,
improve the achievable rate at the PTx and degrade the quality of the

9
Z. Mobini AEUE - International Journal of Electronics and Communications 128 (2021) 153495

eavesdropping channel and hence enhance the average secrecy perfor­
mance. However, the average secrecy performance of FDJ-ZF is less
sensitive to NR specially for low to average values of ρs , since the quality
of STx to PRx channel is more critical for the average secrecy rate per­
formance than the PTx to STx channel. The above observations show the
existence of different design choices when performance-complexity
tradeoff is of interest. Therefore, the beamforming design and antenna
configurations have to be carefully decided.
Fig. 5 presents the average secrecy rate of the FDJ-ZF with different
antenna configurations and for ρs = ρp . The exact and asymptotic
average secrecy rate results are obtained from (31) and Proposition 1,
respectively. It is evident that the exact curves closely match with Monte
Carlo simulations and the asymptotic curves well approximate the exact
ones in the medium-to-high SNR regime. We observe that the curves for
different transmit/receive antennas have the same secrecy slope, which
is presented by (36). Fig. 5 also shows that the average secrecy rate
increases with increasing the number of transmit antennas NT . This
result is in accordance with (37) shows that Ξ∞ 2 increases with NT .
Fig. 6 depicts the secrecy outage probability of FDJ-ZF scheme for
different antenna configurations where the lower bound results are
based on Proposition 2. It is observe that the analytical approximations
in Proposition 2 is sufficiently accurate, and become almost exact in the
high-SNR regime. We see that our proposed scheme achieves a diversity
order of min(NR − 1, NT − 2), which is consistent with the analytical re­
sults derived in the previous section.
6. Conclusions
In this paper, we have investigated the secrecy-enhancing design for
cooperative overlay CR network in the presence of eavesdropper and
with the help of FD multi-antenna STx who acts as the relay and jammer.
To improve the average secrecy of the primary network, deal with the SI,
and prevent information leakage from the relaying path we have pro­
posed FD cooperative jamming and ZF beamforming. The ZF beam­
forming design also enables cancelling PTx and STx interferences to the
SRx and PRx, respectively. Closed-form expression for the average se­
crecy rate, lower bound on the secrecy outage performance along with
high-SNR approximations were also presented. We showed that ZF
beamforming and FD jamming significantly improve the average secrecy
rate and secrecy outage performance of the primary network. However,
secrecy performance gains of the proposed scheme over HD counterparts
and conventional beamforming design highly depend on the system
parameters including the number of antennas at the STx and the trans­
mission powers.
As for future work, it would be interesting to extend these results to
multiple eavesdroppers scenario with both non-colluding and colluding
eavesdroppers as well as to investigate the secrecy performance of
various transmission schemes with multi-antenna SUs/PUs and robust
secure beamforming.
Declaration of Competing Interest
The authors declare that they have no known competing financial
interests or personal relationships that could have appeared to influence
the work reported in this paper.
***
***

Appendix A. Proof of Lemma 1

By using the order statistics, the cdf of γ PRL is expressed as

( )
FγPRL x = Pr(γ0 + γRL ⩽x)
∫x ( ) ( ) (44)
= Fγ0 x − y fγRL y dy.
0

Therefore, we need to derive the pdf of γRL . We will start by finding the cdf of γRL , which by invoking the order statics, can be written as
( ) ( ) ( ) ( ) ( )
FγRL x = Fγ1 x + Fγ2 x − Fγ1 x Fγ2 x . (45)

It can be observed that γ1 ∼ χ 22NR , with cdf given by

⎛ ⎞ ( )
Γ NR , γx
⎜ ⎟ 1
Fγ 1 ⎜ ⎟
⎝x⎠ = 1 − . (46)
Γ(NR )

Moreover, according to [36], γ2 ∼ χ 22(NT − 1) with its cdf given by

⎛ ⎞ ( )
⎜ ⎟ Γ NT − 1, γx2
Fγ 2 ⎜ ⎟
⎝x⎠ = 1 − . (47)
Γ(NT − 1)

Therefore, by substituting (46) and (47) into (45), the cdf of γRL can be expressed as
⎛ ⎞ ( ) ( )
⎜ ⎟ Γ NR , γx1 Γ NT − 1, γx2
Fγ R L ⎜ ⎟
⎝x⎠ = 1 − . (48)
Γ(NR ) Γ(NT − 1)

10
Z. Mobini AEUE - International Journal of Electronics and Communications 128 (2021) 153495

By taking the derivative of FγRL (x) with respect to x, the pdf of γ RL is given by
( ) − x ( )
xNR − 1 e γ1 x
fγRL x = NR ( ) ( ) Γ NT − 1,
γ 1 Γ NR Γ NT − 1 γ2
(49)
− x ( )
xNT − 2 e γ2 x
+ NT − 1
( ) ( ) Γ NR , ,
γ2 Γ NT − 1 Γ NR γ1

where we have used [29, Eq. (8.356.4)]. Now by substituting (22) and (49) into (44) and applying the series expansion of Γ(a, x) [29, Eq. (8.352.4)] we
get
⎛ ⎞
( ) N∑T− 2
∫x
1 1 NR +k− 1 ⎝ − x−γ y
FγPRL x = NR ( ) y 1 − e 0 ⎠e− μ1 y dy
γ1 Γ NR k=0 k!γk2 0

⎛ ⎞ (50)
∑
NR− 1
∫x
1 1 NT +k− 2 ⎝ −
x− y
⎠e− μ1 y
+ NT − 1
( ) y 1− e γ0
dy.
γ2 Γ NT − 1 k=0 k!γk1 0

To this end, by using the integral identity [29, Eq. (3.351.1)], the desired result in (23) is obtained. Furthermore, by takeing derivative with respect to
x, after some simple mathematical manipulation the pdf of γPRL is obtained as (25).

Appendix B. Proof of Lemma 2

⃦ 2
⃒ ⃒2 ⃦
Let us define X = PσP2 ⃒hPE ⃒ and Y = Pσ2S ⃦ ̃
⃦hSE ‖ in (21). It can be readily observed that X is an exponential RV with cdf given by
⎛ ⎞
x
(51)
−
FX ⎝x⎠ = 1 − e γ4
, x⩾0.

⃦ 2
⃦̃
Furthermore, according to [36], ⃦hSE ‖ ∼ χ 22(NT − 1) with pdf given by
⎛ ⎞
NT − 2
y y
(52)
−
fY ⎝y⎠ = ( ) e γ3 , y⩾0.
Γ NT − 1 γN3 T − 1

By utilizing the order statistics, the cdf of γ E is given by

( )
X
FγE (z) = Pr ⩽z
Y +1
∫∞ (( )) ( ) (53)
= FX z y + 1 fY y dy.
0

By substituting (51) and (52) into (53), and then applying the integral identity [29, Eq. (3.351.3)], the desired result in (26) is obtained. Moreover, by
taking derivative of (26) with respect to z the pdf of γE can be obtained as (27).

Appendix C. Proof of Proposition 1

We notice that the average secrecy rate in (31) can be re-expressed as

∫ ∞ ( ∫ x1 ) ( )
1 1 − χ γE (x2 )
Cs = dx2 fγPRL x1 dx1
2ln(2) 0 0 1 + x2

1
∫∞ ( ) ( )
1
∫ ∞ ∫ x1
χ γE (x2 )
( ) (54)
= ln 1 + x fγPRL x1 dx1 + fγPRL x1 dx2 dx1 ,
2ln(2) 0 2ln(2) 0 0 1 + x2
⏟̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅⏞⏞̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅⏟ ⏟̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅⏞⏞̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅⏟
Ξ1 Ξ2

In the high SNR regime with ρp →∞ and ρs →∞ we have ln(1 + x) ≈ ln(x). Therefore, Ξ1 can be approximated as
∫∞ ( ) ( )
1
Ξ∞
1 = ln x fγPRL x dx, (55)
2ln(2) 0
By substituting (25) into (55), we have

11
Z. Mobini AEUE - International Journal of Electronics and Communications 128 (2021) 153495

∑ η ∫ ∞ − a x ρ ( )ηk +n
1 1 NT− 2
b2k ∑∞
( − 1)n e 0 x
Ξ∞
1 = NR
( ) k
lnxdx
2ln(2) a1 Γ NR k=0 k!a2 n=0 n!(ηk + n) 0 a0 ρ b2 ρ
(56)
∑ R− 1
∫ ∞ − a x ρ ( )θk +n
1 1 N
bθ2k ∑∞
( − 1)n e 0 x
+ ( ) lnxdx,
2ln(2) a2NT − 1 Γ NT − 1 k=0 k!ak1 n=0 n!(θk + n) 0 a0 ρ b2 ρ
(
∑ (− 1)n xα+n
where we have used the series expression of γ α, x) = ∞ n=0 n!(α+n) [29, Eq. (8.354.1)]. By applying [29, Eq. (4.352.1)], and perform some algebraic
manipulations, Ξ1 is derived as
( ) ( )
∑ η +n
1 1 N T− 2
1 ∑ ∞
( − 1)n a0k Γ(ηk + n)
Ξ∞
1 = log2 a0 ρ + ( ) n ψ η k + n
2ln(2) aN1 R Γ NR k=0 k!ak2 n=0 n!b2
( ) (57)
∑ R− 1
1 1 N
1 ∑ ∞
( − 1)n a0θk +n Γ(θk + n)
+ ( ) ψ θk + n .
2ln(2) a2NT − 1 Γ NT − 1 k=0 k!ak1 n=0 n!bn2

We now turn our attention to derive Ξ2 In order to derive the asymptotic expression of Ξ2 , we change the order of integration in (54) and rewrite Ξ2
as [44]
∫∞ [ ( )]
[1 − FγE (x2 )]
Ξ2 = 1 − FγPRL x2 dx2 . (58)
0 1 + x2
∑ ∑
By applying the Taylor series expansion γ(α, x) = kj=0 (− 1)j xα+j /j!(α +j) +o(xk ) and ex = kj=0 xj /j! +o(xk ) in (23), we observe that FγPRL (x2 ) ≈ 0
when γ 1 →∞ and γ2 →∞ [44]. Therefore, the asymptotic expression for Ξ2 can be derived as
∫∞
1 − FγE (x2 )
Ξ∞
2 = dx2 . (59)
0 1 + x2
At high SNR regime the received SINR at E in (21) can be approximated as
⃒
ρ ⃒hPE |2
γE ≈ ⃦P . (60)
ρ ⃦hSE wZF ‖2
S t

Accordingly, FγE (x) can be obtained as

( ) ( )− (NT − 1)
a3
Fγ E x = 1 − 1 + x . (61)
a4

Therefore, by substituting (61) into (59) and then by applying [29, Eq. (3.197.1)], Ξ2 can be approximated as
( )
1 a3
Ξ∞
2 = F1 NT − 1, 1; NT , 1 − . (62)
NT − 12 a4
By invoking (54), (57), and (62), we derive the asymptotic average secrecy rate as (32)

Appendix D. Proof of Proposition 2

By adopting the proposed approach in [49], the secrecy outage probability of the system can be lower bounded as
( ) ( ) ∫∞ ( ) ( )
Pout r ⩾PLout r = FγPR rx fγE x dx. (63)
0

Now by substituting (23) and (27) into (63), the lower bound of secrecy outage probability can be expressed as
( ) 1 ∑
N T− 2
1
PLout r = ( ) (ℋ(ηk , μ1 , ζ1 ) − ℋ(ηk , μ2 , ζ2 ))
γN1 R Γ NR k=0 k!γk2
(64)
1 ∑
NR− 1
1
+ NT − 1
( ) (ℋ(θk , μ1 , ζ1 ) − ℋ(θk , μ2 , ζ2 )),
γ2 Γ NT − 1 k=0 k!γk1

where
⎛ ⎞ ⎛ ⎞⎛ ⎞
⎜ ⎟ ( )NT − 1 ∫ ∞ ⎜ ⎟⎜(N − 1)e− ζx ⎟
⎜ ⎟ 1 γ ⎜ ⎟⎜ T e− ζx ⎟
ℋ⎜λ, μ, ζ⎟ = λ 4 γ⎜λ, rμx⎟⎜ ( )N T + ( )NT − 1 ⎟dx. (65)
⎝ ⎠ μ γ3 0 ⎝ ⎠⎝ γ4 γ4
⎠
γ
+ x γ 4 γ
+ x
3 3

Applying the series expansion of γ(a, x) [29, Eq. (8.354.1)] and then utilizing the integral identity [29, Eq. (9.211.4)], to solve the resultant integrals,
we derive the exact lower bound on the secrecy outage probability as in (39).

12
Z. Mobini AEUE - International Journal of Electronics and Communications 128 (2021) 153495

References [25] Ngo HQ, Suraweera HA, Matthaiou M, Larsson EG. Multipair full-duplex relaying
with massive arrays and linear processing. IEEE J Sel Areas Commun 2014;32:
1721–37.
[1] Zhang L, Xiao M, Wu G, Alam M, Liang Y, Li S. A survey of advanced techniques for
[26] Jafarian F, Mobini Z, Mohammadi M. Secure cooperative network with multi-
spectrum sharing in 5G networks. IEEE Wireless Commun Mag 2017;24(5):44–51.
antenna full-duplex receiver. IEEE Syst J 2019;13:2786–94.
[2] Akyildiz IF, Lee W, Vuran MC, Mohanty S. A survey on spectrum management in
[27] Shang Z, Zhang T, Cai Y, Liu Y, Yang W. Secure spectrum-sharing wiretap networks
cognitive radio networks. IEEE Commun Mag 2008;46(4):40–8.
with full-duplex relaying. IEEE Access 2019; 7: 181610–181625.
[3] Manna R, Louie RHY, Li Y, Vucetic B. Cooperative spectrum sharing in cognitive
[28] Li M, Huang Y, Yin H, Wang Y, Cai C. Improving the security and spectrum
radio networks with multiple antennas. IEEE Trans Signal Process 2011;59(11):
efficiency in overlay cognitive full-duplex wireless networks. IEEE Access 2019; 7:
5509–22.
68359–68372.
[4] Li Q, Xu D. Minimizing secrecy outage probability for primary users in cognitive
[29] Gradshteyn IS, Ryzhik IM. Table of integrals, series and products. 7th ed. Academic
radio networks. AEU-Int J Electron Commun 2018;83:353–8.
Press; 2007.
[5] Zou XLY, Liang Y. Secrecy outage and diversity analysis of cognitive radio systems.
[30] Abramowitz M, Stegun IA. Handbook of Mathematical Functions With Formulas,
IEEE J Sel Areas Commun 2014;32:2222–36.
Graphs, and Mathematical Tables. 9th ed. New York: Dover; 1970.
[6] Elkashlan M, Wang L, Duong TQ, Karagiannidis GK, Nallanathan A. On the security
[31] Duarte M, Dick C, Sabharwal A. Experiment-driven characterization of full-duplex
of cognitive radio networks. IEEE Trans Veh Technol 2015;64(8):3790–5.
wireless systems. IEEE Trans Wireless Commun 2012;11:4296–307.
[7] Sharma RK, Rawat DB. Advances on security threats and countermeasures for
[32] Zhang R, Song L, Han Z, Jiao B. Physical layer security for two-way untrusted
cognitive radio networks: a survey. IEEE Commun Surveys Tut 2015;17(2):
relaying with friendly jammers. IEEE Trans Veh Technol 2012;61(8):3693–704.
1023–43.
[33] Zhang H, Xing H, Cheng J, Nallanathan A, Leung VCM. Secure resource allocation
[8] Zhang T, Huang Y, Cai Y, Yang W. Secure transmission in spectrum sharing
for OFDMA two-way relay wireless sensor networks without and with cooperative
relaying networks with multiple antennas. IEEE Commun Lett 2016;20(4):824–7.
jamming. IEEE Trans Ind Informat 2016;12(5):1714–25.
[9] Nandan N, Majhi S, Wu H. Secure beamforming for MIMO-NOMA-based cognitive
[34] Kuhestani A, Mohammadi A, Yeoh PL. Optimal power allocation and secrecy sum
radio network. IEEE Commun Lett 2018;22(8):1708–11.
rate in two-way untrusted relaying networks with an external jammer. IEEE Trans
[10] Chen B, Chen Y, Chen Y, Cao Y, Ding Z, Zhao N, Wang X. Secure primary
Commun 2018;66(6):2671–84.
transmission assisted by a secondary full-duplex NOMA relay. IEEE Trans Veh
[35] Horn RA, Johnson CA. Matrix analysis. 2nd ed. Cambridge, U.K.: Cambridge Univ.
Technol 2019;68(7):7214–9.
Press; 2013.
[11] Mobini Z, Mohammadi M, Tellambura C. Wireless-powered full-duplex relay and
[36] Mohammadi M, Chalise BK, Suraweera HA, Zhong C, Zheng G, Krikidis I.
friendly jamming for secure cooperative communications. IEEE Trans Inf Forensics
Throughput analysis and optimization of wireless-powered multiple antenna full-
Security 2019;14:621–34.
duplex relay systems. IEEE Trans Commun 2016;64:1769–85.
[12] Mobini Z. Secrecy performance of non-orthogonal multiple access cognitive
[37] Huang Y, Wang J, Zhong C, Duong TQ, Karagiannidis GK. Secure transmission in
untrusted relaying with friendly jamming. AEU – Int J Electron Commun 2020;118:
cooperative relaying networks with multiple antennas. IEEE Trans Wireless
153–6.
Commun 2016;15(10):6843–56.
[13] Lei H, Gao C, Ansari IS, Guo Y, Zou Y, Pan G, Qaraqe KA. Secrecy outage
[38] Wang L, Kim KJ, Duong TQ, Elkashlan M, Poor HV. Security enhancement of
performance of transmit antenna selection for MIMO underlay cognitive radio
cooperative single carrier systems. IEEE Trans Inf Forensics Security 2015;10(1):
systems over Nakagami- m channels. IEEE Trans Veh Technol 2017;66(3):2237–50.
90–103.
[14] Zhang N, Lu N, Cheng N, Mark JW, Shen XS. Cooperative spectrum access towards
[39] Huang J, Swindlehurst AL. Robust secure transmission in MISO channels based on
secure information transfer for CRNs. IEEE J Sel Areas Commun 2013;31(11):
worst-case optimization. IEEE Trans Signal Process 2012;60(4):1696–707.
2453–64.
[40] He X, Yener A. Cooperation with an untrusted relay: a secrecy perspective. IEEE
[15] Mobini Z, Khabbazian M. Asymptotic gain analysis of cooperative broadcast in
Trans Inf Theory 2010;56(8):3807–27.
linear wireless networks. IEEE Trans Wireless Commun 2016;15(1):485–97.
[41] Zhu G, Zhong C, Suraweera HA, Zhang Z, Yuen C. Outage probability of dual-hop
[16] Abedi MR, Mokari N, Javan MR, Yanikomeroglu H. Secure communication in
multiple antenna AF systems with linear processing in the presence of co-channel
OFDMA-based cognitive radio networks: an incentivized secondary network
interference. IEEE Trans Wireless Commun 2014;13(4):2308–21.
coexistence approach. IEEE Trans Veh Technol 2017;66(2):1171–85.
[42] Bloch MR, Barros J, Rodrigues MRD, McLaughlin SW. Wireless information-
[17] Mokari N, Parsaeefard S, Saeedi H, Azmi P. Cooperative secure resource allocation
theoretic security. IEEE Trans Inf Theory 2008;54(6):2515–34.
in cognitive radio networks with guaranteed secrecy rate for primary users. IEEE
[43] Wang L, Elkashlan M, Huang J, Schober R, Mallik RK. Secure transmission with
Trans Wireless Commun 2014;13(2):1058–73.
antenna selection in MIMO Nakagami-m fading channels. IEEE Trans Wireless
[18] Xu D, Li Q. Resource allocation for secure communications in cooperative cognitive
Commun 2014;13(11):6054–67.
wireless powered communication networks. IEEE Syst J 2019;13(3):2431–42.
[44] Wang L, Yang N, Elkashlan M, Yeoh PL, Yuan J. Physical layer security of maximal
[19] Xu D, Li Q. Resource allocation for cognitive radio with primary user secrecy
ratio combining in two-wave with diffuse power fading channels. IEEE Trans Inf
outage constraint. IEEE Syst J 2018;12(1):893–904.
Forensics Security 2014;9(2):247–58.
[20] Zhang Z, Chai X, Long K, Vasilakos AV, Hanzo L. Full-duplex techniques for 5G
[45] Peppas KP, Sagias NC, Maras A. Physical layer security for multiple-antenna
networks: Self-interference cancellation, protocol design, and relay selection. IEEE
systems: a unified approach. IEEE Trans Commun 2016;64(1):314–28.
Commun Mag May 2015;53(5):128–37.
[46] Lozano A, Tulino AM, Verdu S. High-SNR power offset in multiantenna
[21] Riihonen T, Werner S, Wichman R. Mitigation of loopback self-interference in full-
communication. IEEE Trans Inf Theory 2005;51(12):4134–51.
duplex MIMO relays. IEEE Trans Signal Process 2011;59:5983–93.
[47] Jin S, McKay MR, Zhong C, Wong K. Ergodic capacity analysis of amplify-and-
[22] Psomas C, Mohammadi M, Krikidis I, Suraweera HA. Impact of directionality on
forward MIMO dual-hop systems. IEEE Trans Inf Theory 2010;56(5):2204–24.
interference mitigation in full-duplex cellular networks. IEEE Trans Wireless
[48] Lei H, Wang D, Park K, Ansari IS, Jiang J, Pan G, Alouini M. Safeguarding UAV IoT
Commun 2017;16(1):487–502.
communication systems against randomly located eavesdroppers. IEEE Internet
[23] Mobini Z, Mohammadi M, Chalise BK, Suraweera HA, Ding Z. Beamforming design
Things J 2020;7(2):1230–44.
and performance analysis of full-duplex cooperative NOMA systems. IEEE Trans
[49] Lei H, Gao C, Ansari IS, Guo Y, Pan G, Qaraqe KA. On physical-layer security over
Wireless Commun 2019;18:3295–311.
SIMO generalized-K fading channels. IEEE Trans Veh Technol 2016;65(9):7780–5.
[24] Suraweera HA, Krikidis I, Zheng G, Yuen C, Smith PJ. Low-complexity end-to-end
performance optimization in MIMO full-duplex relay systems. IEEE Trans Wireless
Commun 2014;13:913–27.

13