3480 IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY, VOL.

16, 2021

Physical Layer Security Enhancement

With Reconfigurable Intelligent
Surface-Aided Networks
Jiayi Zhang , Senior Member, IEEE, Hongyang Du , Student Member, IEEE, Qiang Sun ,
Bo Ai , Senior Member, IEEE, and Derrick Wing Kwan Ng , Fellow, IEEE

Abstract— Reconfigurable intelligent surface (RIS)-aided wire- I. I NTRODUCTION

less communications have drawn significant attention recently.
We study the physical layer security of the downlink RIS-aided
transmission framework for randomly located users in the pres-
ence of a multi-antenna eavesdropper. To show the advantages
R ECENTLY, reconfigurable intelligent surface (RIS) has
been proposed as a promising technique as it can achieve
high spectral-/energy-efficiency through adaptively controlling
of RIS-aided networks, we consider two practical scenarios:
Communication with and without RIS. In both cases, we apply the wireless signal propagation environment [1]. Specifically,
the stochastic geometry theory to derive exact probability density RIS is a planar array which comprises a large number of
function (PDF) and cumulative distribution function (CDF) of nearly passive reflecting elements. By equipping the RIS
the received signal-to-interference-plus-noise ratio. Furthermore, with a controller, each element of the RIS can independently
the obtained PDF and CDF are exploited to evaluate important introduce a phase shift on the reflected signal. Besides, a RIS
security performance of wireless communication including the can be easily coated on existing infrastructures such as walls of
secrecy outage probability, the probability of nonzero secrecy buildings, facilitating low-cost and low-complexity implemen-
capacity, and the average secrecy rate. Monte-Carlo simulations tation. By smartly adjusting the phase shifts induced by all the
are subsequently conducted to validate the accuracy of our reflecting elements, the RIS can bring various potential advan-
analytical results. Compared with traditional MIMO systems,
tages to the system such as enriching the channel by deliber-
the RIS-aided system offers better performance in terms of
physical layer security. In particular, the security performance ately introducing more multi-paths, increasing the coverage
is improved significantly by increasing the number of reflecting area, or beamforming, while consuming very low amount of
elements equipped in a RIS. However, adopting RIS equipped energy due to the passive nature of its elements [2]. Recently,
with a small number of reflecting elements cannot improve the RIS has been introduced into many wireless communication
system performance when the path loss of NLoS is small. systems. For instance, the authors in [3] studied a RIS-aided
Index Terms— Fisher-Snedecor F -distribution, MIMO, recon-
single-user multiple-input single-output (MISO) system and
figurable intelligent surface, stochastic geometry. optimized the introduced phase shifts to maximize the total
received signal power at the users. In [4], various designs
Manuscript received December 16, 2020; revised April 7, 2021 and were proposed for both the transmit power allocation at a base
May 11, 2021; accepted May 15, 2021. Date of publication May 24, 2021; station (BS) and the phases introduced by the RIS elements
date of current version June 8, 2021. This work was supported in part to maximize the energy and the spectral efficiency of a RIS-
by the National Key Research and Development Program of China under aided multi-user MISO system. Besides, the authors of [5]
Grant 2020YFB1807201; in part by the National Natural Science Foundation
of China under Grant 61971027, Grant 61971467, Grant U1834210, and
considered a downlink (DL) multiuser communication system
Grant 61961130391; and in part by the Beijing Natural Science Foundation where the signal-to-interference-plus noise ratio (SINR) was
under Grant L202013. The work of Derrick Wing Kwan Ng was supported maximized for given phase shifts introduced by a RIS.
in part by the University of New South Wales (UNSW) Digital Grid Futures In practice, communication security is always a fundamental
Institute, UNSW, Sydney, through the Cross-Disciplinary Fund Scheme and problem in wireless networks due to its broadcast nature.
in part by the Australian Research Council’s Discovery Project under Grant
DP210102169. The associate editor coordinating the review of this manuscript Besides traditional encryption methods adopted in the appli-
and approving it for publication was Prof. Stefano Tomasin. (Corresponding cation layer, physical layer security (PLS) [6]–[8] serves as
author: Qiang Sun.) an alternative for providing secure communication in fast
Jiayi Zhang is with the School of Information Science and Technology, access fifth-generation (5G) networks. Furthermore, PLS can
Nantong University, Nantong 226019, China, and also with the School
of Electronic and Information Engineering, Beijing Jiaotong University,
achieve high-quality safety performance without requiring
Beijing 100044, China. actual key distribution, which is a perfect match with the
Hongyang Du is with the School of Electronic and Information Engineering, requirements of 5G. To fully exploit the advantages of 5G,
Beijing Jiaotong University, Beijing 100044, China. multi-antenna technology has become a powerful tool for
Qiang Sun is with the School of Information Science and Technology, enhancing the PLS in wireless fading networks, e.g. [9]–[15].
Nantong University, Nantong 226019, China (e-mail: sunqiang@ntu.edu.cn).
Bo Ai is with the State Key Laboratory of Rail Traffic Control and Safety, In particular, with the degrees of freedom provided by multiple
Beijing Jiaotong University, Beijing 100044, China, and also with the Henan antennas, a transmitter can steer its beamforming direction to
Joint International Research Laboratory of Intelligent Networking and Data exploit the maximum directivity gain to reduce the potential
Analysis, Zhengzhou University, Zhengzhou 450001, China. of signal leakage to eavesdroppers. However, there are only a
Derrick Wing Kwan Ng is with the School of Electrical Engineering and
Telecommunications, University of New South Wales, Sydney, NSW 2052,
few works considering the PLS in the emerging RIS-based
Australia. communication systems, despite its great importance for
Digital Object Identifier 10.1109/TIFS.2021.3083409 modern wireless systems. For example, a RIS-aided secure
1556-6021 © 2021 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: University of Electronic Science and Tech of China. Downloaded on June 08,2024 at 02:18:52 UTC from IEEE Xplore. Restrictions apply.
 ZHANG et al.: PHYSICAL LAYER SECURITY ENHANCEMENT WITH RIS-AIDED NETWORKS 3481

communication system was investigated in [16], [17] but their TABLE I

communication systems only consist of a transmitter, one M ATHEMATICAL S YMBOLS AND F UNCTIONS
legitimate receiver, single eavesdropper, and a RIS with limited
practical applications. Furthermore, authors in [18] studied a
DL MISO broadcast system where a BS transmits independent
data streams to multiple legitimate receivers securely in the
existence of multiple eavesdroppers. In order to conceive a
practical RIS framework, user positions have to be taken into
account using stochastic geometry for analyzing the system
performance. Stochastic geometry is an efficient mathematical
tool for capturing the topological randomness of networks
[19], [20]. However, there are only a few works studying the
impact of user-locations on the security performance.
Motivated by the aforementioned reasons, in this paper,
we study the security performance of a RIS-aided DL multiple-
input multiple-output (MIMO) system for randomly located
roaming multi-antenna users in the presence of a multi-antenna
eavesdropper. Moreover, we aim to answer a fundamental
question “How much improvement can RIS bring to physi-
cal layer security in wireless communications?”. The main
contributions of this paper are summarized as follows:
• To show the advantages of exploiting RIS to enhance
the PLS in wireless communications, we consider two
practical scenarios: RIS is adopted and its absence. For
each case, we derive the exact closed-form expressions
of SINR and its PDF and CDF, respectively. Further-
more, considering a general set-up when the direct links
between the BS and the RIS-aided users exist, we obtain
the CDF and outage probability (OP) expressions.
• Exploiting the tools from stochastic geometry, we propose
a novel PLS analysis framework of RIS-aided communi- II. S YSTEM M ODEL AND P RELIMINARIES
cation systems. Novel expressions are derived to charac- A. System Description
terize the PLS performance, namely the secrecy outage
probability (SOP), the probability of nonzero secrecy Let us consider the DL MIMO-RIS system and assume that
capacity (PNSC), and the average secrecy rate (ASR). a BS equipped with M transmit antennas (TAs) communicates
The derived results can provide insights, i.e., the security with N users each of which is equipped with K receive anten-
performance is improved significantly by increasing the nas (RAs). a RIS is installed between the BS and the users for
number of reflecting elements. assisting end-to-end communication. By jointly optimizing the
• We derive highly accurate and simplified closed-form transmit precoding at the BS and the reflect phase shifts at the
approximations of the CDF expressions of SINR for RIS according to the wireless channel conditions, the reflected
both scenarios. Furthermore, we present an asymptotic signals can be added constructively at the desired receiver
OP analysis in the high-SNR regime. The derived results to enhance the received signal power, which can potentially
show that adopting RIS equipped with a small number yield a secure transmission [22]. In addition, a malicious
of reflecting elements cannot improve the system perfor- eavesdropper who is equipped with K RAs aims to eavesdrop
mance if the path loss of NLoS is small. the information of the desired user. We assume that the number
of reflecting elements on the RIS is L. The RIS is equipped
The remainder of the paper is organized as follows. with a controller to coordinate between the BS and the RIS for
In Section II, we introduce the system model of the RIS-aided both channel acquisition and data transmission [3]. As such,
DL MIMO communication system and derive the exact closed- the signal received from the BS can be manipulated by the
form of PDF and CDF expressions of SINR for users and the RIS via adjusting the phase shifts and amplitude coefficients
eavesdropper. Section III formulates the security performance of the RIS elements.
analysis in two scenarios and present the corresponding exact On the other hand, various practical approaches have been
expressions of the performance. In Section IV, we derive the proposed in the literature for the channel estimation of
approximations to the statistics of SINR in two scenarios, RIS-aided links [23]–[25]. Thus, we can assume that the global
and analyze the OP in high-SINR regime. Section V presents channel state information (CSI) of users is perfectly known for
the CDF and OP of scenario 2 when the direct links exist joint design of beamforming [26]. The CSI of the eavesdropper
between BS and RIS-aided users. Section VI shows the is typically unavailable as the eavesdropper is passive to hide
simulation results and the accuracy of the obtained expressions its existence. Therefore, the passive eavesdropping scenario
is validated via Monte-Carlo simulations. Finally, Section VII is considered [27]. To quantify the performance of such a
concludes this paper. RIS-aided communication system, we adopt SOP, PNSC, and
Mathematical notations: A list of mathematical symbols ASR as performance metrics of interested.
and functions most frequently used in this paper is available Furthermore, we assume that the users are located on a
in Table I. disc with a radius r2 according to homogeneous Poisson point

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on June 08,2024 at 02:18:52 UTC from IEEE Xplore. Restrictions apply.
 3482 IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY, VOL. 16, 2021

have been derived as [32, eq. (30)]. As a result, the CDF of

the user locations is given by
r2 r02
FR (r ) =  2  −   , if r0 < r < R, (2)
r2 − r02 r2 2 − r02
where r0 is the minimum distance which is used to avoid
encountering a singularity [33]. Thus, the probability of the
distance between the user and the BS is less than r1 can be
expressed as
r2 r2
Fig. 1. Two practical scenarios: Communication without (left hand side) and B2  Pr (r ≤ r1 ) = FR (r1 ) =  2 1 2  −  2 0 2  . (3)
with RIS (right hand side). r2 − r0 r2 − r0

processes (HPPP) [28] with density λ and the eavesdropper D. Path Loss Model
chooses a position that is close to the legitimate user.
In practice, the direct transmission link between the BS and For the first communication scenario without a RIS, dif-
the users may be blocked by trees or buildings. Such assump- ferent path loss equations are applied to model the LoS and
tion is applicable for 5G and mmWave communication systems NLoS links as [29], [31]
that are known to suffer from high path and penetration losses  −α
d n ,L1 , if BS → user n link is LoS link.
resulting in signal blockages. In order to unveil the benefits in L (d) = u−α (4)
adopting the RIS, we consider the following two practice of
2
du n ,NL , if BS → user n link is NLoS link,
communication scenarios:
where α1 and α2 are the LoS and NLoS path loss expo-
• As shown in Fig. 1 (a), for the first scenario, RIS is not nents, respectively. Typical values of α1 and α2 are defined
adopted for enhancing the PLS and the quality of wireless in [34, Table 1], while α1 < α2 hold in general.
communication. Specifically, both LoS and NLoS links For the second communication scenario with the assistance
exist in our system. Without loss of generality, we focus by the RIS, recently, the free-space path loss models of
our attention on user n. We use dυ,σ to denote the distance RIS-aided wireless communications are developed for different
between the BS and user n (n = 1, . . . , N). According situations in [35, Proposition 1]. Authors in [35] proposed
to Table I, we have υ = u n , σ = L for LoS links and that the free-space path loss of RIS-aided communications is
υ = u n , σ = NL for NLoS links. proportional to (du R ,L du R −u n ,L )2 in the far field case. Thus,
• For the second scenario, a RIS is deployed to leverage
we obtain
the LoS components with respect to both the BS and  
the users to assist their end-to-end communication of L du R ,L , du R −u n ,L = C L 1 C L 2 du−2 d −2
R ,L u R −u n ,L
, (5)
Fig. 1 (b). Thus, the NLoS users in scenario 1 now can
communicate with BS through two LoS links with the where C L 1 and C L 2 denote the path loss intercepts of BS-RIS
help of RIS. Moreover, the locations of the BS and the and RIS-user links, respectively.
RIS are fixed, hence we assume that the distance between
the BS and the RIS is known and denoted by du R ,L . E. Small-Scale Fading
In addition, the distance between the RIS and user n is
To unify the system performance with different channel
denoted by du R −u n ,L .
environment, generalized fading distributions have been pro-
B. Blockage Model posed that include the most common fading distributions as
special cases [8], [36], [37]. It has been shown recently that
A blockage model was proposed in [29], which can be the Fisher-Snedecor F composite fading model can provide
regarded as an accurate approximation of the statistical block- a more comprehensive modeling and characterization of the
age model [30] and incorporates the LoS ball model proposed simultaneous occurrence of multipath fading and shadowing,
in [31] as a special case. In the considered system model, which is generally more accurate than most other general-
we adopt the blockage model to divide the users process in the ized fading models [38]. Furthermore, the Fisher-Snedecor F
spherical region around the BS into two independent HPPPs: model is more general and includes several fading distributions
LoS users process and NLoS users process. In particular, as special cases, while being more mathematically tractable.
we define q L (r ) as the probability that a link of length r For example, authors in [32] conceived a RIS-aided MIMO
is LoS. Each access link of separation r is assumed to be framework using Nakagami-m distribution. Note that Fisher-
LoS with probability B1 if r ≤ r1 and 0 otherwise: Snedecor F distribution includes the case of Nakagami-m
 distribution for m s → ∞ as one of its subsequent special
B1 , if r ≤ r1
q L (r ) = , (1) cases, such as Rayleigh (m = 1) and one-sided Gaussian
0, otherwise
(m = 1/2). Motivated by this, the Fisher-Snedecor F dis-
where 0 ≤ B1 ≤ 1. The parameter B1 can be interpreted as the tribution has been adopted to analyze the RIS-aided systems,
average LoS area in a circular ball with a radius of r1 around i.e., RIS-aided Internet-of-Things networks [39].
the BS. To conceive a practical RIS framework, we assume that the
small-scale fading of each link follows Fisher-Snedecor F
C. User Model fading distributions [39]. The channel correlation between
Let us assume that the users are located according to a HPPP RIS elements may exist because the electrical size of RIS’s
within the disc shown as Fig. 1. The PDF of the user locations reflecting elements is between λ/8 and λ/4 in principle,

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on June 08,2024 at 02:18:52 UTC from IEEE Xplore. Restrictions apply.
 ZHANG et al.: PHYSICAL LAYER SECURITY ENHANCEMENT WITH RIS-AIDED NETWORKS 3483

where λ is a wavelength of the signal [40]. However, it is hard for each TA, G υ,κ and gυ,κ are respectively the array gains
to model the correlation in RIS because a tractable model for of main and sidelobes. In practice, the BS can adjust their
capturing such unique characteristics has not been reported in antennas according to the CSI. In the following, we denote
the literature yet. Therefore, we consider a scenario where the the boresight direction of the antennas as 0◦ . For simplifying
correlation is weak enough to be ignored, i.e., the electrical the performance analysis, different antennas of the authorized
size of RIS’s reflecting elements is larger than λ/4 because the users and the malicious eavesdropper are assumed to have
frequency of signals is high [41], [42]. Thus, we can assume same array gains. Thus, without loss of generality, we assume
that the small scale fading components are independent of each that Gυ,κ = Gυ , G υ,κ = G υ and gυ,κ = gυ .
other.
For the first communication scenario, in order to charac- G. SINR Analysis
terize the LoS and NLoS links between the BS and users
(eavesdropper), the small-scale fading matrices are defined as Let us consider a composite channel model of large-scale
⎡ υ,σ υ,σ ⎤
and small-scale fading. It is assumed that the distance dυ,σ ,
q1,1 · · · q1,M υ = u 1 , . . . , u N , u e , and σ = L, NL are independent but
⎢ . .. .. ⎥ not identically distributed (i.n.i.d.) and the large-scale fading
Qυ,σ = ⎢⎣ .
. .
⎥
. ⎦, (6)
is represented by the path loss. In the DL transmission,
υ,σ υ,σ
q K ,1 · · · q K ,M the complex baseband transmitted signal at the BS can be
then expressed as
where Qυ,σ (σ = L, NL) is a K × M matrix. Letting υ = u n ,
we obtain the small-scale fading matrix between the BS and 
N
user n. Using [43, Eq. (5)], the PDF of the elements of (6) x= pu n su n , (11)
can be expressed as n=1

2m m ((m s − 1) )m s x 2m−1 where su n (n = 1, . . . , N) are i.i.d. random variables (RVs)

f X (x) =  m+m s , (7) with zero mean and unit variance, denoting the information-
B (m, m s ) mr 2 + (m s − 1)  bearing symbols of users, and pu n ∈ C M×1 is the correspond-
ing beamforming vector. The transmit power consumed at the
where  = E r 2 is the mean power, m and m s are physical
BS can be expressed as
parameters which represent the fading severity and shadowing
parameters, respectively. 
N
 2
P= pu  . (12)
For the second case, the small-scale fading matrices for the n
LoS links between the BS and users (eavesdropper) are same n=1
as the first case. In addition, in order to model the LoS links For the first communication scenario, e.g., Fig. 1 (a), there are
between the BS and the RIS, the small-scale fading matrix is both LoS and NLoS links. The signal received by user n or the
defined as eavesdropper from the BS can be expressed as
⎡ u R ,L u R ,L ⎤ 
q1,1 · · · q1,M  
⎢ . yυ,σ = G υ L dυ,σ vυ Qυ,σ pυ sυ
.. .. ⎥
Qu R ,L = ⎢
⎣ .
. . . ⎦,
⎥ (8)   
N
u R ,L
q L ,1 u R ,L
· · · q L ,M + G υ L dυ,σ vυ Qυ,σ pi si +vυ N0 , (13)
i =υ
where Qu R ,L is a L × M matrix.   
In order to model the LoS links between the RIS and users Multi−user interference
(eavesdropper), the small-scale fading matrices are defined as where υ = u n , u e , σ = L, NL, vu n ∈ C1×K is the detection
⎡ υ,L υ,L ⎤ vector of user n, and N0 ∈ C K ×1 denotes the additive white
q1,1 · · · q1,L
⎢ . .. .. ⎥
Gaussian noise, which is modeled as a realization of a zero-
Qυ,L = ⎢⎣ .
. . . ⎦,
⎥ (9) mean complex circularly symmetric Gaussian variable with
variance σ N2 .
q Kυ,L
,1 · · · q Kυ,L,L For the second communication scenario, e.g., Fig. 1 (b),
where Qυ,L is a K × L matrix. Letting υ = u R − u n , we get there are LoS and RIS-aided links. For LoS links, the signal
the small-scale fading matrix between the RIS and user n. received by the user or the eavesdropper can be obtained by
letting σ = L in (13). For RIS-aided links, the signal can be
expressed as
F. Directional Beamforming
υ
Highly directional beamforming antenna arrays are yRIS
  
deployed at the BS to compensate the significant path-loss = G υ L du R ,L , du R −υ,L vυ Qu R −υ,L Qu R ,L pυ sυ
in the considered system. For mathematical tractability and
similar to [29], [30], the antenna pattern of users can be   
N
approximated by a sectored antenna model in [44] which is + vυ N0 + G υ L du R ,L , du R−υ,L vυ Qu R−υ,L Qu R ,L pi si ,
given by i =υ
   
G υ,κ , if |θ | ≤ θc , Multi−user interference
Gυ,κ (θ ) = (κ = 1, . . . , K ), (10) (14)
gυ,κ , otherwise,
where θ , distributed in [0, 2π ], is the angle between the BS where υ = u n , u e , = diag [β1 φ1 , β2 φ2 , . . . , β L φ L ] is
and the user, θc denotes the beamwidth of the main lobe, a diagonal matrix accounting for the effective phase shift

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on June 08,2024 at 02:18:52 UTC from IEEE Xplore. Restrictions apply.
 3484 IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY, VOL. 16, 2021

introduced by all the elements of the RIS,1 β ∈ (0, 1] where υ = u R − u n , u R − u e , and

represents the amplitude reflection coefficient, φ = exp ( j θ), ⎡ υ,L   u ,L     ⎤
n = 1, . . . , L, θ ∈ [0, 2π), denotes the phase shift. q1,1 · q1,m R
··· υ,L
q1,L · q Lu R,m,L
⎢ ⎥
In the literature, numerous methods have been proposed  ⎢ .. .. .. ⎥
hυ,RIS =⎢ . . . ⎥ (21)
to obtain the effective phase shift introduced by all the ⎣       ⎦
elements of the RIS, e.g., [3], [32], [46]–[48]. To mitigate υ,L u R ,L υ,L u R ,L
q Q,1 · q1,m ··· q Q,L · q L ,m
the interference at the n th user and perform well both in
RIS and non-RIS scenarios, we adopt the design of passive is a Q × L matrix.
beamforming process and the downlink user-detection vectors With the help of (18) and (20), we derive the exact PDF
proposed in [32]. The passive beamforming matrix at RIS can and CDF expressions for each scenario’s SINR in terms
be obtained as [32, eq. (13)] of multivariate Fox’s H -function [49, eq. (A-1)], which are
H̃−1 X summarized in the following Theorems.
= , (15) Theorem 1: For the first communication scenario, let
βmax Z υ,σ = SINRυ,σ , the CDF of Z υ,σ is expressed as
where βmax is obtained by finding the maximum ampli-
tude coefficient, X is the signal sent by BS, and H̃ can 
Q M
1
be formulated from Qu R −u 1 ,L Qu R ,L , . . ., Qu R −u N ,L Qu R ,L FZ υ,σ (z) =   (HCDF1 ,2 − HCDF1 ,0 ), (22)
=1
 m s  (m  )
as [32, eq. (9)].
The detection vector of the n th user can be written where HCDF1 ,h̄ (h̄ = 0, 2) is derived as (23), shown at
as [32, eq. (19)] G p QM
the bottom of the next page, A1 = β 2 υ Qu2nσ 2 , {m  }i=1 =
vu n = Tu n xu n , (16)  M  Q  Q M  max
 υ,σ  M
N  Q
m υ,σ
q,n n=1 , m s i=1 = m s q,n n=1 , and
q=1 q=1
where Tu n is derived by the left singular vectors of H̃. With   Q
= υ,σ
QM M
the aid of the classic maximal ratio combining technique, { }i=1 q,n n=1 .
xu n can be expressed as q=1
Proof: Please refer to Appendix A.
TH
u hu n Theorem 2: For the first communication scenario, the PDF
xu n =  Hn , (17)
T hu  of the SINR for the user n and eavesdropper can be given by
un n

where hu n is the u n th column of the effective channel matrix 

Q M
1
Qu R −u n ,L Qu R ,L . Thus, the SINRs involved in both scenarios f Z υ,σ (z) =   (HPDF1 ,2 − HPDF1 ,0 ), (24)
=1
 m s  (m  )
can be obtained.
For the first case, the received SINR at user n and the where HPDF1 ,h̄ (h̄ = 0, 2) is derived as (25) shown at the
eavesdropper are denoted as SINRυ,σ , and it can be expressed bottom of next page.
as2 Proof: Following similar procedures as in Appendix B,
 2  −ασ
Gυ hυ,σ  F dυ,σ pu n we can derive the PDF of Z υ,σ by calculating
SINRυ,σ = , (18)  r ασ
βmax Q σ N
2 2 2 2
f Y (y) = x f X (x y) f D (x)dx, (26)
where pu n denotes the transmit power of the BS for user, r0ασ
Q  K − M − 1 is the effective antenna gain, and
⎡ υ,σ where f X (·) is given as [50, eq. (23)] and f D (·) has been
υ,σ ⎤
q1,1 · · · q1,M derived as (A-4).
⎢ .. ⎥
hυ,σ = ⎣ ...
 ..
. . ⎦ (19) Then, we focus on the derivation for the performance metric
υ,σ υ,σ in the second scenario.
q Q,1 · · · q Q,M Theorem 3: For the second communication scenario, let
is a Q × M matrix which denotes the channel gain of user Z υ,RIS = SINRυ,RIS , the CDF of Z υ,RIS can be expressed
n. In addition, we can see that the eavesdropper’s SINR is as
affected by the random directivity gain G u e (θ ). 
QL
HCDF2 ,2 − HCDF2 ,0
For the second case, the SINR of LoS path can be obtained FZ υ,RIS (z) = , (27)
by letting σ = L in (18). For the RIS-aided path, with the help 
2    
=1  m si,  m i,
of the design of passive beamforming process [32], the SINR i=1
with the optimal phase shift design of RIS’s reflector array
can be derived as [32, eq. (29)] where HCDF2 ,h̄ (h̄ = 0, 2) is derived as (28)
 2  −2
Gυ C L 1 C L 2 hυ,RIS  F du R ,L dυ,L pu n shown
⎛ at the bottom of the ⎞next page, =
SINRυ,RIS = , (20) T
βmax
2 Q2σ 2
N ⎜ A2 zr22 m 1,1 m 2,1 A2 zr22 m 1,Q L m 2,Q L ⎟
⎝2   ,..., 2 
  ⎠ , A2 =
1 We assume that the phase shifts can be continuously varied in [0, 2π ) to m si,1 −1 γ̄1 m si,Q L −1 γ̄ Q L
i=1 i=1
 M &Q
characterize the fundamental performance of RIS. Although using the RIS  Q L
Gυ C L 1 C L 2 (du R ,L )−2 pu n υ,L
with discrete phase shifts causes performance loss [45], our results serve as
βmax
2 Q 2σ 2 ,
m 1, i=1 = m q,n ,
performance upper bounds for the considered system which are still useful N n=1 q=1
 &
M Q  &
M Q
for characterizing the fundamental performance of the RIS-aided system.  Q L  Q L
2 To facilitate effective eavesdropping, the eavesdropper should try to close
m 1,s i=1 = m s υ,L
q,n , 1, i=1 = υ,L
q,n ,
to BS because the SINR of the eavesdropper increases as du e ,σ decreases. n=1 q=1 n=1 q=1

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on June 08,2024 at 02:18:52 UTC from IEEE Xplore. Restrictions apply.
 ZHANG et al.: PHYSICAL LAYER SECURITY ENHANCEMENT WITH RIS-AIDED NETWORKS 3485

 L  L  L  L
R ,L R ,L (m  ) → (m 1, ); (m 2,), (m s ) → (m s1, ); (m s2, ), Q M → Q L
m 2, i=1 = m un,m , m 2,s i=1 = m s un,m , and
 L   n=1
L
n=1 and ασ = NL, where A → B means replacing A with B, and
R ,L
2, i=1 = un,m . the results for the RIS-aided scenario will follow.
n=1
Proof: Please refer to Appendix B.
Theorem 4: For the second communication scenario, III. S ECURITY P ERFORMANCE A NALYSIS
the PDF of Z υ,RIS can be deduced as The secrecy rate over fading wiretap channels [51] is defined
as the difference between the main channel rate and the

QL
(HPDF2 ,2 − HPDF2 ,0 )
f Z υ,RIS (z) = , (29) wiretap channel rate as
2     
=1  m si,  m i,   Cu n ,σ − C2 , Z u n ,σ > Z u e ,σ .
i=1
Cs Z u n ,σ , Z u e ,σ =
0, otherwise,
where HPDF2 ,h̄ (h̄ = 0, 2) is derived as (30) shown at the ⎧ * +
⎨log 1 + Z u n ,σ , Z
bottom of the page. 2 u n ,σ > Z u e ,σ .
= 1 + Z u e ,σ
Proof: Following similar procedures as in Appendix B, ⎩
we can express the PDF of Z υ,RIS with the help of (26). 0, otherwise,
Remark 1: By comparing Theorem 1 and Theorem 3 (31)
or Theorem 2 and Theorem 4, we can see that, caused by which means that a positive secrecy rate can be assured if and
the characteristics of Fisher-Snedecor F distribution, the CDF only if the received SINR at user n has a superior quality than
and PDF of the SINR for the RIS-aided path are very similar that at the eavesdropper.
to that of the LoS or NLoS path. This insight is very useful In the considered RIS-aided system, we assume that the
because we only need to further investigate Theorem 1 and location of users are random and define
Theorem 2 for performance analysis of the first communication
scenario. In other words, we can get the results for the PA  [PLoS , PNLoS ] (32)
RIS-aided scenario through simple parameter transformation.
and
 can let m  → m
Specifically, we  2, , m s → m s1,
 1, m  m s2,,
 (m  ) →  m 1,  m 2, ,  m s →  m s1,  m s2, , PB PrLoS Pr G ,Pr LoS Pr g ,Pr NLoS Pr G ,Pr NLoSPr g , (33)

⎛  ⎞
A1 zrh̄ασ m 1

*
  +
⎜     ⎟
⎜  1 − 2 ; 1, · · · , 1 : (1, 1) , 1−m s , 1 ; . . . ; (1, 1) , 1−m s , 1 ⎟
γ̄1 m s1 −1
2r 2 ⎜  ..ασ 1 QM ⎟
0,1:1,2;...;1,2 ⎜  ⎟
HCDF1 ,h̄   2 h̄2  H1,2:2,1;...;2,1 ⎜  . * + ⎟.
r2 − r0 ασ ⎜  2   ⎟
⎜ ασ  (0; 1, . . . , 1) − ; 1, . . . , 1 : (m 1 , 1) ; . . . ; m Q M , 1 ⎟
⎝ A1 zrh̄ m Q M  ασ ⎠
 
γ̄ Q M m s −1  QM
(23)

⎛  ⎞
A1 zrh̄α m 1

 * +
⎜       ⎟
⎜  − 2 ; 1, . . . , 1 : (1, 1) , 1−m s , 1 ; . . . ; (1, 1) , 1− m s , 1 ⎟
γ̄1 m s1 − 1
2r 2+α
A1 ⎜ ..  α 1 QM ⎟
0,1:1,2;...;1,2 ⎜  ⎟
HPDF1 ,h̄   2 h̄ 2  H1,2:2,1;...;2,1 ⎜ .  * + ⎟.
r2 − r0 α ⎜  2   ⎟
⎜ A zr α m  (1; 1, . . . , 1) −1− ; 1, . . . , 1 : (m 1 , 1) ; . . . ; m Q M , 1 ⎟
⎝ 1 h̄ Q M  ⎠
  α
γ̄ Q M m s Q M − 1 
(25)

⎛  ⎞
 (0; 1, . . . , 1) : (1, 1) , 1 − m s , 12 ; . . . ; (1, 1) , 1 − m s , 12
rh̄2  i,1 i=1 i,Q L i=1
⎜  * + ⎟
  2  H 0,1:1,2;...;1,2 ⎝  ⎠.
 (0; 1, . . . , 1) − 2 ; 1, . . . , 1 : m , 12 ; . . . ; m 2
HCDF2 ,h̄
r2 − r02 1,2:2,1;...;2,1  i,1 i,Q L , 1
2 i=1 i=1

(28)

⎛   2  2 ⎞

rh̄4 A2 
0,1:1,2;...;1,2 ⎝ 
(−1; 1, . . . , 1) : (1, 1) , 1 − m s i,1 , 1 i=1
; . . . ; (1, 1) , 1 − m s i,Q L , 1 i=1 ⎠
HPDF2 ,h̄   2  H1,2:2,1;...;2,1      .
r2 − r0
2
 2
(1; 1, . . . , 1) (−2; 1, . . . , 1) : m i,1 , 1 i=1 ; . . . ; m i,Q L , 1 i=1
2

(30)

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on June 08,2024 at 02:18:52 UTC from IEEE Xplore. Restrictions apply.
 3486 IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY, VOL. 16, 2021

θc
where PrLoS  B1 B2 , PrNLoS  1 − B1 B2 , Pr G  180 , Pr g  Fox’s H -function, so Q, M and L have a greater impact on
θc
1 − 180 , PLoS represents the probability that the path is LoS in the SOP than the channel parameters. Thus, it is obvious that
two scenarios, PNLoS denotes the probability that the path is communication system designers can increase L of RIS for
NLoS in the first scenario or RIS-aided in the second scenario, lower SOP.
PG is the probability that the eavesdropper’s directivity gain is
the same as the user’s, and Pg represents the probability that C. PNSC Characterization
the eavesdropper’s directional gain is gu e . Besides, we assume Another fundamental performance metric of the PLS is the
that the eavesdropper has the same path loss as user n because probability of non-zero secrecy capacity (PNSC) which can
they are close to each other. be defined as
⎡   ⎤
Pr  Z u n ,L > Z u e ,L,G
A. Outage Probability Characterization ⎢ Pr Z u n ,L > Z u e ,L,g ⎥
P N SC = PB ⎢ 
⎣ Pr Z u n ,NL > Z u e ,NL,G ⎦
⎥
The OP is defined as the probability that the instantaneous  
SINR is less than Z th , where Z th is the determined SNR Pr Z u n ,NL > Z u e ,NL,g
threshold. In the first communication scenario, the OP can T
= PB P N Z 1,G , P N Z 1,g , P N Z 2,G , P N Z 2,G . (40)
be directly calculated as
T Thus, we exploit (40) and Theorems 1-4 to arrive the
O P = PA FZ u n ,L (Z th ), FZ u n ,NL (Z th ) , (34) following proportions.  
which can be evaluated directly with the help of (22). Proposition 2: Let P N SCσ,λ̄  Pr Z u n ,σ > Rs Z u e ,σ,λ̄
Remark 2: In the second communication scenario, the OP denotes the element of the matrix in (40) and we can express
can be obtained by replacing FZ u n ,NL (Z th ) in (34) with P N SCσ,λ̄ as
FZ u n ,RIS (Z th ). We can observe that OP decreases when the P N SCσ,λ̄ = P2,2 − P2,0 − P0,2 + P0,0 , (41)
channel condition of user n is improved. Moreover, a RIS
equipped with more elements will also make the OP lower. where P p,q ( p = 0, 2 and q = 0, 2) can be writ-
, as (42) at the top of-Tthe next page, H P N SC 
ten
B. SOP Characterization A1 zr αp m 1 A1 zrqα m 2Q M
  ,· · ·,   , H P1 and H P2 are expressed
The secrecy outage probability (SOP), is defined as the eγ̄1 m s1−1 eγ̄2Q M m s2Q M −1
probability that the instantaneous secrecy capacity falls below as (43) and (44) shown at the bottom of the next page,
a target secrecy rate threshold. In the first communication respectively.
scenario, the SOP can be written as Proof: Please refer to Appendix D.
⎡   ⎤ Remark 4: For the second communication scenario,
Pr  Z u n ,L ≤ Rs Z u e ,L,G + Rs − 1
⎢ Pr Z u ,L ≤ Rs Z u ,L,g + Rs − 1 ⎥ the PNSC of the RIS-aided path can be easily obtained
S O P = PB ⎢  n e
⎣ Pr Z u n ,NL ≤ Rs Z u e ,NL,G + Rs − 1 ⎦
⎥ according to Remark 1. From (41), we can see that PNSC
  will increase when user n has good communication
Pr Z u n ,NL ≤ Rs Z u e ,NL,g + Rs − 1 conditions or when the eavesdropper is in a bad
T communication environment. Besides, we can observe
= PB S O P1,G , S O P1,g , S O P2,G , S O P2,G , (35)
that PNSC is also more susceptible to Q, M, and L, because
where Rt is the target secrecy rate, Rs = 2 Rt [52], Z u e ,L,G these parameters directly determine the dimension of the
denotes that the eavesdropper’s directional gain is G u e and multivariate Fox’s H -function. In general, system designers
Z u e ,L,g denotes that the eavesdropper’s directional gain is gu e . can improve PNSC by increasing the size of the RIS.
With the help of (35) and Theorems 1-4, we derive the
following propositions.   D. ASR Characterization
Proposition 1: Let S O Pσ,λ̄  Pr Z u n ,σ ≤ Rs Z u e ,σ,λ̄ + Rs −1 The ASR describes the difference between the rate of
denotes the element of the matrix in (35), where λ̄ = G, g. the main channel and wiretap channel over instantaneous
We can express S O Pσ,λ̄ as SINR can be expressed as
⎡   ⎤
S O Pσ,λ̄ = S2,2 − S2,0 − S0,2 + S0,0 , (36) E Cs  Z u n ,L , Z u e ,L,G
⎢ E Cs Z u ,L , Z u ,L,g ⎥
where S p,q ( p = 0, 2 and q = 0, 2) is derived AS R = PB ⎢  n e
⎣ E Cs Z u n ,NL , Z u e ,NL,G ⎦
 ⎥
as
, (37) at the top of the next
-T page, H S O P   
E Cs Z u n ,NL , Z u e ,NL,g
A1 zr αp m 1 A1 zrqα m 2Q M+1
 ,. . .,  , e , H S1 and H S1 are T
γ̄1 m s1 −1 γ̄2Q M+1 m s2Q M+1 −1 = PB ASC1,G , ASC1,g , ASC2,G , ASC2,G . (45)
respectively given as (38) and (39) shown at the bottom We derive the following proposition using (45) and
of the next page, and e is a positive number close to zero Theorems 1-4.
(e.g., e = 10−6 ). Proposition
 3: We  assume that AS Rσ,λ̄ =
Proof: Please refer to Appendix C. E Cs Z u n ,σ,Z u e ,σ,λ̄ is the element of the matrix in (45) and
Remark 3: In the second communication scenario, the SOP we can expressed AS Rσ,λ̄ as
of the RIS-aided path can be obtained with the help of
Remark 1. From (36), we can see that the SOP will decrease 
2
 
when the fading parameter m u n and m su n increase because AS Rσ,λ̄ = A2,2,i − A2,0,i − A0,2,i + A0,0,i
of better communication conditions. In addition, we can see i=1
 
that Q, M, and L will affect the dimension of the multivariate − A2,2,3 − A2,0,3 − A0,2,3 + A0,0,3 , (46)

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on June 08,2024 at 02:18:52 UTC from IEEE Xplore. Restrictions apply.
 ZHANG et al.: PHYSICAL LAYER SECURITY ENHANCEMENT WITH RIS-AIDED NETWORKS 3487

⎛ ⎞ ⎛ ⎞ ⎛ ⎞
where A p,q,i ( p = 0, 2, q = 0, 2, and i = 1, 2)
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
can be derived
, as (47) shown at the bottom of -the page, H A1,2  ⎝−α2 ; 0,. . ., 0, 0⎠ ; ⎝1, 1, . . . , 1, 0⎠ ; ⎝0, 1,. . ., 1, 0⎠,
T         
A1 zr αp m 1 A1 zrqα m 2Q M+1 QM QM QM
H AS R  
γ̄1 m s1 −1
, . . . ,  , e , H A1 , and H A2,2 can be expressed as (51), shown at the bottom of
γ̄2Q M+1 m s2Q M+1 −1
H A2 , and A p,q,3 ( p = 0, 2 and q = 0, 2) are respec- the page.
tively given as (48), (49) Proof: Please refer to Appendix E.
, and (50) shown at the bottom
-T Remark 5: For second communication scenario, the ASR
A1 zr αp m 1 A1 zrqα m Q M+1 of the RIS-aided path can also be obtained according to
of the page, H ASC2  
γ̄1 m s1−1
 ,· · ·,  , e ,
γ̄ Q M+1 m s Q M+1−1 Remark 1. From (46), as expected, better channel conditions

*      +
4 A1r p 2rq2+ασ  H S1 : (1, 1) , 1 − m s1 , 1 ; . . . ; (1, 1) , 1 − m s2Q M , 1 ; −
S p,q = 
0,3:1,2;...;1,2;1,0 
2 × H4,4:2,1;...;2,1;0,1 H S O P    . (37)
αu n αu e r22 − r02 H S2 : (m 1 , 1) ; . . . ; m 2Q M , 1 ; (0, 1)
⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞
⎜ 2 ⎟ ⎜ 2 ⎟ ⎜ ⎟ ⎜ ⎟
H S1  ⎝1 − ; 1, . . . , 1, 0, . . . , 0, 0⎠ ; ⎝− ; 0, . . . , 0, 1, . . . , 1, 0⎠ ; ⎝2, −1, . . . , −1⎠ ; ⎝0, 1,. . ., 1⎠ . (38)
α       α            
QM QM QM QM 2Q M+1 2Q M+1
⎛ ⎞⎛ ⎞⎛ ⎞⎛ ⎞
⎜ ⎟⎜ 2 ⎟⎜ ⎟⎜ 2 ⎟
H S2  ⎝0; 1,. . ., 1, 0,· · ·, 0, 0⎠ ⎝− ; 1,. . ., 1, 0,· · ·, 0, 0⎠;⎝1; 0, . . . , 0; 1,· · ·, 1, 0⎠;⎝−1 − ; 0,· · ·, 0, 1,. . ., 1, 0⎠ . (39)
      α             α      
QM QM QM QM QM QM QM QM

,     -
4 A1r p 2rq2+ασ  H S : (1, 1) , 1 − m s , 1 ; . . . ; (1, 1) , 1 − m s , 1
0,3:1,2;...;1,2  1 1 2Q M
P p,q =  2 H3,4:2,1;...;2,1 H P N SC    . (42)
eαu n αu e r22 − r02 H S2 : (m 1 , 1) ; . . . ; m 2Q M , 1
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
⎜ 2 ⎟ ⎜ 2 ⎟ ⎜ ⎟
H P1 = ⎝1 − ; 1, . . . , 1, 0, . . . , 0⎠ ; ⎝− ; 0, . . . , 0, 1, . . . , 1⎠ ; ⎝0, 1, . . . , 1⎠ . (43)
α       α         
QM QM QM QM 2Q M

⎛ ⎞⎛ ⎞ ⎛ ⎞⎛ ⎞
⎜ ⎟⎜ 2 ⎟ ⎜ ⎟⎜ 2 ⎟
H P2  ⎝0; 1, . . . , 1, 0, . . . , 0⎠ ⎝− ; 1, . . . , 1, 0, . . . , 0⎠ ; ⎝1; 0, . . . , 0; 1, . . . , 1⎠;⎝−1− ; 0, . . . , 0, 1, . . . , 1⎠ . (44)
      α             α      
QM QM QM QM QM QM QM QM

*     +
4 A1r p 2rq2+ασ  H A1 : (1, 1) , 1 − m s1 , 1 ;· · ·; (1, 1) , 1 − m s2Q M , 1 ;
2 H5,4:2,1;...;2,1;0,1 H AS R   
0,4:1,2;...;1,2;1,0
A p,q,i =  . (47)
e ln 2αu n αu e r22 − r02 −H A2 : (m 1 , 1) ;· · ·; m 2Q M , 1 ; (1, 1)
⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞
⎜ 2 ⎟ ⎜ 2 ⎟ ⎜ ⎟ ⎜ ⎟
H A1  ⎝1 − ; 1, . . . , 1, 0, . . . , 0, 0⎠ ; ⎝− ; 0,. . ., 0, 1,· · ·, 1, 0⎠ ; ⎝1, −1,· · ·, −1, 1,· · ·, 1, 0⎠ ; ⎝0, 1,· · ·, 1, 0⎠ .
α       α               
QM QM QM QM QM QM 2Q M
(48)
⎛ ⎞⎛ ⎞ ⎛ ⎞ ⎛ ⎞
⎜ ⎟⎜ 2 ⎟ ⎜ ⎟ ⎜ 2 ⎟
H A2  ⎝0; 1,· · ·, 1, 0,· · ·, 0⎠ ⎝− ; 1,· · ·, 1, 0,· · ·, 0⎠ ; ⎝1; 0, . . . , 0; 1,. . ., 1, 0⎠ ; ⎝−1 − ; 0, . . . , 0, 1,. . ., 1, 0⎠ .
      α             α      
QM Q M+1 QM Q M+1 QM QM QM QM
(49)
*      +
2 A1rq2+ασ  H A1,2 : (1, 1) , 1 − m s1 , 1 ; . . . ; (1, 1) , 1 − m s Q M , 1 ;
A p,q,i   H 0,3:1,2;...;1,2;1,0
H ASC2    . (50)
e ln 2αu e r22 − r02 4,2:2,1;...;2,1;0,1  −H A2,2 : (m 1 , 1) ; . . . ; m Q M , 1 ; (1, 1)
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
⎜ ⎟ ⎜ ⎟ ⎜ 2 ⎟
H A2,2  ⎝0; 1, . . . , 1, 0⎠ ; ⎝1; 1, . . . , 1, 0⎠ ; ⎝−1 − ; 1, . . . , 1, 0⎠ . (51)
      α   
QM QM QM

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on June 08,2024 at 02:18:52 UTC from IEEE Xplore. Restrictions apply.
 3488 IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY, VOL. 16, 2021

will result in a higher ASR. In addition, we can also observe

that the ASR can be improved by assuring larger RIS because
the multivariate Fox’s H -function in ASR is also more easily
affected by Q, M, and L as SOP and PNSC.

IV. A SYMPTOTIC A NALYSIS

 In the
2 SINR expression (18) of scenario 1, we observe that
 
hυ,σ F is the sum of Fisher-Snedecor F RVs. An accu-
rate closed-form approximation to the distribution of sum
of Fisher-Snedecor F RVs using single Fisher-Snedecor F
distribution has been given in [53, Theorem 3]. Thus, eq. (18) Fig. 2. RIS communication system with the direct links.
can be expressed as
 −ασ In the following, we focus on OP and study the asymptotic
SINRυ,σ  ≈ A1 dυ,σ F1 , (52) behavior in the high-SINR regime, i.e., when γF1 → ∞,
  to gain more insights.
where A1 is defined after (22), F1 ∼ F m F1 , m s F1 , γ̄F1 is a For the first communication scenario, the OP can be
single
 Fisher-Snedecor F RV which is adopted to approximate expressed as
hυ,σ  F , where m F1 , m s F1 , and γ̄F1 can be obtained from the
2
 2 T
parameters of  hυ,σ  F with the help of [53, eq. (14)]. O P = PA FZ u n ,L (Z th ), FZ u n ,NL (Z th ) , (56)
 2
In the SINR expression (20) of scenario 2,  hυ,RIS  F is which can be evaluated with the aid of the CDF of Z υ,σ  .
the sum of product of two Fisher-Snedecor F RVs. With To obtain more engineering insights from (56), the asymptotic
the help of [8, eq. (18)] and [8, eq. (10)], we can exploit behavior in the high-SINR regime is analyzed.
a single Fisher-Snedecor F RV to approximate the product Proposition 4: In the high-SINR regime, the OP can be
of two Fisher-Snedecor
2 F RVs. Therefore, the distribution approximated as
of hυ,RIS  F can also be approximated by a single Fisher- . /T
Snedecor F distribution. Following the similar parameters O P = PA FZ u n ,L (Z th ), FZ u n ,NL (Z th ) , (57)
transformation method as in (52), eq. (20) can be approximated
as where

−m F 2+ασm F1 2+ασm F1
 mF mF
SINRυ,RIS ≈ A2 dυ,L −2 F2 , (53) 2γ̄F1 1 r2 −r0 m F1m F1−1A1 1 Z th 1
  FZ u n,σ  =      m ,
where F2 ∼ F m F2 , m s F2 , γ̄F2 is a single Fisher-Snedecor 2+ασ m F1 r22 −r02 B m s F1 , m F1 m s F1 −1 F1
F RV, and A2 is defined after (27). (58)
Theorem 5: For the first communication scenario, let
Z υ,σ  = SINRυ,σ  , the CDF of Z υ,σ  can be expressed as and σ  = L, NL.
Proof: Please refer to Appendix G.
m F 2 −m F1  
2γ̄F1 1 G 1,0 − G 1,2 Furthermore, at high-SINRs, the OP can be written as O P ∝
FZ υ,σ  (z) =      , (54) γ̄F−G d
, where G d denotes the diversity order. From (57),
ασ r22 − r02  m F1  m s F1 1
we can observe that G d = m F1 .
where G 1,h̄ (h̄ = 0, 2) is derived as As for the second communication scenario, the OP and
⎛   ⎞ corresponding asymptotic expression can be obtained through

A m s F1 − 1  1, 1 + 2 parameter transformation in Remark 6.
3,0 ⎝
G 1,h̄ = rh̄2 G 2,0
1  ασ ⎠ . (55)
ασ −m F1  0, m , 2
rh̄ m F1 z γ̄F1  s F1 ασ

Proof: Please refer to Appendix F. V. RIS-A IDED S YSTEM W ITH D IRECT L INKS
Remark 6: For the second communication scenario, let For the second communication scenario, we assume that
Z υ,RIS = SINRυ,RIS , the CDF of Z υ,RIS can be obtained there is no direct links between the BS and the users when
through simple parameter transformation in (54). Specifically, the BS beams toward the RIS. This assumption is reasonable
we can let A1 → A2 , ασ → 2, m F1 → m F2 , m s F1 → m s F2 , when the signals have high frequency which are easily blocked
γ̄F1 → γ̄F2 , G 1,0 → G 2,0 , and G 1,2 → G 2,2 . by obstructions [54].
Thus, with the help of the approximated CDF of SINRυ,σ  However, as shown in Fig. 2, a more general SINR expres-
and SINRυ,RIS , we can re-derive the performance metrics, sion with direct channels considered can be given as
i.e., OP, SOP, PNSC, and ASR, to simplify the involved  2  2 
−2 
calculations. Note that the multivariate Fox’s H -functions SINRυ,RD = A2 dυ,L hυ,σ  F + 
hυ,RIS  F . (59)
in the CDF expressions of SINRυ,σ and SINRυ,RIS can be
expressed in the form of a multi-fold Mellin-Barnes type Note that deriving the exact statistical characteristics of
contour integration where the order of integration increases SINRυ,RD are challenging, if not impossible, for calculat-
with the number of elements in the RIS. Besides, the Meijer G- ing the performance metrics. Thus, with the help of (52)
functions in the CDF expressions of SINRυ,σ and SINRυ,RIS and (53), we can obtain the high-quality approximated CDF
can be written as a one-fold Mellin-Barnes integration. There- of SINRυ,RD .
fore, the re-derivations of OP, SOP, PNSC, and ASR follow Theorem 6: For the second communication scenario with
similar methods as Section III A-D, but the results are simpler. direct links, let Z υ,R D = SINRυ,R D , the CDF of Z υ,R D can

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on June 08,2024 at 02:18:52 UTC from IEEE Xplore. Restrictions apply.
 ZHANG et al.: PHYSICAL LAYER SECURITY ENHANCEMENT WITH RIS-AIDED NETWORKS 3489

Fig. 4. Outage probability versus the number of reflecting elements equipped

Fig. 3. Outage probability versus the transmit power for user n. in RIS.

be expressed as
, -m F  −1

2
A−1 zr 2 m F 
rh̄2 r22 − r02 G
FZ υ,R D (z) =  2 h̄      ,
=1
m s F − 1 γ̄F  m s F  m F
(60)
where G is derived as (61), shown at the bottom of the next
page, and G mp11,n 1 ;m 2 ,n 2 ;m 3 ,n 3
,q1 ; p2 ,q2 ; p3 ,q3 (·) denotes the Bivariate Meijer’s
G-function [55].
Proof: Please refer to Appendix H.
Therefore, the OP of the second scenario with direct links can
T
be expressed as O P = PA FZ υ,L (Z th ), FZ υ,R D (Z th ) .
Fig. 5. Secrecy outage probability versus the transmit power for user n.
VI. N UMERICAL R ESULTS
In this section, analytical results are presented to illustrate Figure 4 illustrates the OP performance versus the number
the advantages of applying the RIS to enhance the security of reflecting elements equipped in RIS, with K = 4, M = 2,
of the DL MIMO communication system. We assume that the Z th = 0 dB, and pum = 25 dB. In Fig. 4, we can observe
noise variances σ N at user n and the eavesdropper are identical that the OP decreases as L increases. This is because the
and σ N = 0 dB. The transmit power pu n is defined in dB increases of RIS’s elements offer more degrees of freedom
with respect to the noise variance. The array gains of main for efficient beamforming and interference management which
and sidelobes are set to G υ = 30 dB and gυ = −10 dB. For can significantly improve the system performance. However,
the small scale fading, we set m u n = 5, m u e = 3, m su n = 5, when the path loss of NLoS is small, applying a RIS equipped
m su e = 3, γ̄u n = γ̄u e = −10 dB. The path loss exponent with a small number of elements, i.e., less than 25, cannot
is set to α1 = 2. Moreover, in the considered simulation offer a lower OP than the traditional MIMO system. This
scenario, we assume that r2 = 400, r1 = 300, r0 = 1, is because the signal propagation distance in the direct link
βmax = 1, du R ,L = 30 and B1 = 0.3. The numerical results are between the BS and user is shorter then the end-to-end link
verified via Monte-Carlo simulations by averaging the obtain when RIS is used. In contrast, we can observe that when the
performance over 106 realizations. path loss of NLoS is large, i.e., when α2 = 3, the performance
Figure 3 depicts the OP performance versus transmit power gain brought by RIS is significant. Furthermore, when L is
pu n with K = 4, M = 2, Z th = 0 dB. As it can be observed, sufficiently large such that the reflected signal power by the
the OP decreases as transmit power and L increase, which RIS dominates the total received power at user n, we can
means that the introduction of a RIS can provide a higher observe that there is a diminishing return in the slope of OP
quality of communication for the considered MIMO wireless due to channel hardening [56]. Thus, to achieve a low OP
communication system. Moreover, when α2 = 2.5, we can at the user, there exists an trade-off between the number of
observe that RIS cannot improve the system performance reflecting elements in RIS and the transmit power of the BS.
when the number of reflecting elements is small. This is Furthermore, we can find that an outage floor exists in OP
because the OP is dominated by the path loss. When the when L → ∞. This is because the SINR first increases with L,
signal is propagates to the user via the direct links, the signal which results in a decreasing OP. As L continues increasing,
propagation distance is shorter. Furthermore, while consider- the power of noise becomes negligible compared to the power
ing the direct links exists between BS and RIS-aided users, of received signal. Hence, the SINR becomes saturated which
we can observe the OP is slightly lower than while assuming is independent of L, leading to a saturated OP.
no direct links exist. In addition, analytical results perfectly Figure 5 depicts the SOP performance versus transmit power
match the approximate ones and Monte-Carlo simulations pu n under different θc with Rt = 1, K = 4, M = 2, and
well. In addition, the asymptotic expressions match well the α2 = 3. For a certain setting of the parameters, the SOP
exact ones at high-SINR values thus proving their validity and decreases as the transmit power of the BS increases.
versatility. Besides, it can be easily observed that the SOP increases

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on June 08,2024 at 02:18:52 UTC from IEEE Xplore. Restrictions apply.
 3490 IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY, VOL. 16, 2021

Figure 7 illustrates the ASR as a function of the transmit

power pu n for different settings of L with K = 4, M = 2,
and α2 = 3. As expected, there is a perfect match between
our analytical and simulated results. Like SOP and PNSC,
when the transmit power is not large, there is only a small
difference for the ASR between communication with RIS and
without RIS. This is also because adopting RIS increases
the signal propagation distance. In contrast, when the power
is moderate to high, RIS is very useful for enhancing the
security of the physical layer, and as analyzed in Remark 5,
the larger the L is, the more obvious the performance gain is.
Furthermore, we can observe that there is a diminishing return
in the ASR gain due to increasing L. This is because when
L is large, the security performance of the system is mainly
Fig. 6. Probability of non-zero secrecy capacity versus transmit power for
user n. limited by other factors, such as the channel quality and the
path loss.

VII. C ONCLUSION
We proposed a new RIS-aided secure communication sys-
tem and presented the exact of expressions for CDF and PDF
of the SINR for both cases of the RIS is used or not used.
Our analytical framework showed the analytical performance
expressions of RIS-aided communication subsume the coun-
terpart case without exploiting RIS which can be obtained
by some simple parameter transformation of the former case.
Secrecy metrics, including the SOP, PNSC, and ASR, were
all derived with closed-form expressions in terms of the
multivariate Fox’s H -function. Numerical results confirmed
that RIS can bring significant performance gains and enhance
Fig. 7. Average security rate versus transmit power for user n. the PLS of the considered system, especially when the number
of reflecting elements is sufficiently large and the path loss of
NLoS links is severe.
as θc decreases. This is because a large value of θc offers the
eavesdroppers a higher possibility in exploiting the larger array
gains. Again, it is evident that the analytical results match the A PPENDIX A
Monte-Carlo simulations well. P ROOF OF T HEOREM 1
Figure 6 shows the analytical and simulated PNSC versus
transmit power pu n under different θc with Rt = 1, K = 4, Based on the result derived in (18) and exploiting the fact
M = 2, and α2 = 3. As it can be observed, the PNSC that the elements of Qυ,σ are i.n.i.d., the effective channel gain
decreases as θc increases because the eavesdropper is more vector of user n (eavesdropper) can be transformed into
likely to receive signals. Moreover, similar to the results
 M 
Q  
in Fig. 5, the use of RIS can enhance the security of the  2  υ,σ 2
system. Furthermore, if the transmit signal power is not hυ,σ  F = qq,n  . (A-1)
high or the number of reflecting elements on the RIS is not q=1 n=1
large, the performance gain brought by RIS is rather then
limited, as the signal propagation distance in the direct link Note that the elements of Qυ,σ obey the Fisher-Snedecor
between the BS and user is shorter then the end-to-end link F distribution with fading parameters m υ,σ , m s
υ,σ
and
υ,σ
QM
when the RIS is used. However, with the increase of transmit . For the sake of presentation, we define {q  } i=1 =
 Q
power, the communication scenario with RIS enjoys a better υ,σ  M
security performance. qq,n n=1 .
q=1

0,1;1,3;1,3
G = G 0,2;3,2;3,2
⎛    ⎞
 2  2  2  A−1r 2 zm F
 1−m , 1−m −m , 1−   2 h̄
m   1−m F2 , 1−m s F2 −m F2 , 1− m F   1
⎜ m F +1  F1 s F1 F1 ⎟
⎜    m s F1 −1 γ̄F1 ⎟
⎜ =1  =1  =1  ⎟
×⎜ 2    ⎟.
⎜ 2
 2   2  −1 2 ⎟
⎝   A r zm ⎠
m , 2+ m
F F  0, − m F  0, − m F   2 h̄  F2
=1 =1
  m −1 γ̄
=1 =1 s F2 F2
(61)

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on June 08,2024 at 02:18:52 UTC from IEEE Xplore. Restrictions apply.
 ZHANG et al.: PHYSICAL LAYER SECURITY ENHANCEMENT WITH RIS-AIDED NETWORKS 3491

 L 
Q    
Letting γ = γ̄ |q |2 / , where γ̄ = E[ ], we have  υ,L 2  u R ,L 2
  m = qq,n  qn,m  . (B-1)
m  m  · m s − 1 γ̄ s γ m  −1 q=1n=1
f γ (γ ) =     m +m . (A-2)  
B m  , m s m  γ + m s − 1 γ̄  s υ,L u R ,L
Note that qq,n ∼ F m υ,L υ,L υ,L
q,n ,m sq,n ,q,n and qn,m ∼
0
Q M  
u R ,L u R ,L u R ,L
Thus, let X = γ , the CDF of X is obtained with the F m n,m ,m sn,m ,n,m are i.n.i.d. Fisher-Snedecor F RVs.
=1    L & Q
help of [50, eq. (23)] and [49, eq (A.1)] as QL υ,L u R ,L
Again, we define that {q }i=1 = qq,n qn,m

Q M    * + n=1 q=1
1 1 QM in order to make our proof process more concise. Thus, using
FX (x) =   ···
 m s  (m  ) L1 L2 |q |2
=1 L Q M 2π j [58, eq. (18)] and letting N = 2 and γ = γ̄ 1, 2, , the MGF

Q M of γ can be expressed as
1 ,  -
× , -  (m  − ζ )  (ζ ) 
0
Q M 2,3  m 1, m 2,  1, 1−m s1, , 1−m s2,
 1+ ζ =1 G 3,2    m 1, , m 2,
m s1,−1 m s2,−1 γ̄ γ 
=1 Mγ (γ ) =         .
, -ζ  m 1,  m s1,  m 2,  m s2,
  xm 
×  m s +ζ   dζ1 · · · dζ Q M . (B-2)
γ̄ m s − 1
0L
Q
(A-3) Letting X = |γ |, we can
 α =1
Let D = du m ,σ σ , hence, using [32, eq. (30)], we derive derive the PDF of X as
the PDF of D as 
1
2 f X (x) = L−1 {M X (s); x} = M X (s)e xs ds, (B-3)
 d ασ −1 , if r0ασ < d < r2ασ . (A-4)
2
f D (d) =  2 2π j L
r2 − r0 ασ
2
where L−1 {·} denotes the inverse Laplace transform. The MGF
Let us define Y = X/D, because X and D are statistically of the X can be expressed with the aid of (B-2) and [21,
independent, the CDF of Y can be formulated as eq. (8.4.3.1)]
 r ασ   QL
(2π j)−Q L
2
FY (y) = P {X ≤ y D} = FX (x y) f D (x)dx. (A-5) 1
f X (x) =        
r0ασ 2π j L  m 1,  m s1,  m 2,  m s2,
=1
Then, by replacing (A-3) and (A-4) into (A-5) and exchanging   , -−ζ
the order of integration according to Fubini’s theorem, we get m 1, m 2,
× · · · ϒ (ζ )   
FY (y) LQL L1 m s1, −1 m s2, −1 γ̄ s

Q M * + × e xs dζ1 · · · dζ Q L ds, (B-4)
2 1 1 QM
=  2 
r2 − r02 ασ =1  (m st )  (m  ) 2π j where
      
1 ϒ (ζ ) =  (−ζ )  m s1, − ζ  m s2, − ζ
× ··· , -    
L1 L2 LQM Q0M ×  m 1, + ζ  m 2, + ζ (B-5)
 1+ ζ
=1 Note that the order of integration can be interchangeable
, -ζ according to Fubini’s theorem, we can re-write (B-4) as

Q M
  ym 
× (m  − ζ ) (ζ )  m s +ζ   
QL * +
γ̄ m s − 1 1 1 QL
=1 f X (x) =        
 m 1,  m s1,  m 2,  m s2, 2π j
 r2ασ
0
Q M =1
ζ + α2σ −1
  , -−ζ
× x =1 dx dζ1 dζ2 · · · dζ Q M . (A-6) m 1, m 2,
r0ασ × ··· ϒ (ζ )   
   L Q L L1 m s1, −1 m s2, − 1 γ̄
IA
 Q0L ζ
I A can easily be deduced. Letting Z = A1 Y , 1 

we obtain (22). The proof is now complete. × s =1 e xs ds dζ1 · · · dζ Q L . (B-6)

2π j L
  
A PPENDIX B IB1

P ROOF OF T HEOREM 3 Letting xs = −t and using [21, eq. (8.315.1)], we can derive
Based on the result derived in (20) and exploiting the fact IB1 as
that the elements of Q3 and Qυ,L are i.n.i.d., the effective ⎛ ⎞
0L
Q
channel gain vector of user n (eavesdropper) can be trans- −1− ζ QL
IB1 = x =1  −1 ⎝− ζ ⎠. (B-7)
formed into
L  =1
 2 Q   2
   υ,L u R ,L 
hυ,RIS F =  qq,n qn,m  Substituting (B-7) into (B-6), we obtain the PDF of X,
q=1n=1 which can be written as Fox’s H function. The CDF of X can

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on June 08,2024 at 02:18:52 UTC from IEEE Xplore. Restrictions apply.
 3492 IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY, VOL. 16, 2021

BS and to the RIS is very small. For example, the maximum is

less than 5 meters. Thus, let Y = X/D, hence, we can derive
the CDF of Y using (A-5)
FY (y)
2 
QL
(2π j)−Q L
= 2         
r2 − r0 α =1  m 1,  m s1,  m 2,  m s2,
2

  , -−ζ
ϒ (ζ) xm 1,m 2,
× ··· * +   
LQL L1 0L m s1, −1 m s2, −1 γ̄
 1− ζ
=1
 r2α
0
L
α −1− ζ
2

Fig. 8. The difference between the expected distance from the user to the × y =1 dy dζ1 · · · dζ Q L . (B-11)
r0 α
BS and the expected distance from the user to the RIS.   
IB3
1x IB3 can be solved easily. Let α = 2 and Z = A2 Y ,
be expressed as FX (x) = 0 fr (r ) dr . Thus, we can rewrite
the CDF of X as we obtain (27) and complete the proof.

FX (x) A PPENDIX C

QL
(2π j)−Q L P ROOF OF P ROPOSITION 1
=        
=1
 m 1,  m s1,  m 2,  m s2, S O Pσ,λ̄ can be expressed as
  , -−ζ  ∞
ϒ (ζ) m 1, m 2, S O Pσ,λ̄ = FZ u n ,σ (Rs Z + Rs − 1) f Z u e ,σ,λ̄ (Z )d Z . (C-1)
× ··· * L +   
LQL L1 0 m s1, −1 m s2, − 1 γ̄ 0
 − ζ Substituting (22) and (24) into (C-1), we obtain
=1
 x −1−
0L
Q
ζ S O Pσ,λ̄
   
× r =1 dr dζ1 · · · dζ Q L , (B-8)

Q 
M Q M  −1 m u e ,s2  −1 m u e ,2
0   =    
IB2
1 =1 2 =1  m u n ,s1  m u n ,1
 ∞
where IB2 can be solved as  
× Hu n ,C D F1,2 − Hu n ,C D F1,0
 x −1−
0L
Q
ζ −
0L
Q
ζ 0 
1 × Hu e ,P D F1 ,2 − Hu e ,P D F1 ,0 d Z
IB2 = r =1 dr = x =1 . (B-9)
0L
Q
0 − ζ  S2,2 − S2,0 − S0,2 + S0,0 . (C-2)
=1

Substituting (B-9) into (B-8) and using [21, eq. (8.331.1)], Substituting (23) and (25) into S p,q and changing the order
equation (B-8) can be expressed as of integration, the Integral part in S p,q can be expressed as

 0
Q M 0
Q M

QL
(2π j)−Q L ∞ ζu n ,1 ζu e ,2
FX (x) =         I1 = (Rs Z + Rs − 1)1 =1 Z 2 =1 dZ. (C-3)
=1
 m 1,  m s1,  m 2,  m s2, 0
  , -−ζ
Rs
xm 1,m 2, Let t = Rs −1 Z , I1 can be written as
× ···   
L Q L L1 m s1, − 1 m s2, − 1 γ̄ 0 0
ζu n ,1 * +1+
Q M Q M
ζu e ,2
ϒ (ζ ) 1 =1
Rs − 1 2 =1
× * + dζ1 · · · dζ Q L . (B-10) I1 = (Rs − 1)
0L Rs
 1− ζ 0 0

Q M Q M
=1 ∞ ζu n ,1 ζu e ,2
1 =1 2 =1
× (t + 1) (t) dt , (C-4)
In order to avoid complex trigonometric operations,  0
 
we assume that dυ,σ and dυ,L are i.n.i.d. RVs. To verify that I2
this assumption will only cause small errors for our considered
system, we perform Monte-Carlo simulation for the location let L{ p(t)} = P(x). Using the property of Laplace trans-
1t
of user n with r0 = 1. In the simulation, we generate the form, we have L 0 p(z)dz = P(x) x . According to the final
location of user n for 106 times according to [32, eq. (30)]. value theorem, it follows that
Fig. 8 shows that when the distance from the RIS to the BS, * t +
defined as rRIS , and the range of the system model, r2 , change, P(e)
lim p(z)dz = e = P(e). (C-5)
the difference between the mean distance from user n to the t →∞ 0 e

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on June 08,2024 at 02:18:52 UTC from IEEE Xplore. Restrictions apply.
 ZHANG et al.: PHYSICAL LAYER SECURITY ENHANCEMENT WITH RIS-AIDED NETWORKS 3493

 ∞  
Using (C-5) and [49, eq. (2.9.6)], we can solve I2 as × log2 (1 + Z ) Hu n ,C D F1,2 − Hu n ,C D F1,0
⎧ ⎫ ⎛ ⎞
⎪ 0 0 0 
ζu e ,2⎪
Q M Q M
⎨ ζu n ,1 ⎬ 
QM × Hu e ,P D F1 ,2 − Hu e ,P D F1 ,0 d Z
I2 = L (t + 1)1 =1 t 2 =1 =  ⎝1+ ζu e ,2⎠
⎪
⎩ ⎪
⎭  A2,2,i − A2,0,i − A0,2,i + A0,0,i . (E-3)
2 =1
⎛ ⎞ Substituting (22) and (24) into (E-2), we can express the

QM 
QM 
QM
integral part of Iσ,λ̄,i (i = 1, 2) as
×  ⎝1+ ζu e ,2 , 2 + ζu n ,1 + ζu e ,2 , s⎠ ,
2 =1 1 =1 2 =1  0
Q M 0
Q M
∞ ζu n ,1 + ζu e ,2
(C-6) I4 = log2 (Z + 1) Z 1 =1 2 =1
dZ, (E-4)
0
where (·) is the tricomi confluent hypergeo-
metric function [21, eq. (9.210.2)]. Using [59, which has been solved using the final value theorem as
eq. (07.33.07.0003.01)], (C-6) and (C-4) and letting [53, eq. (49)]. Thus, I3 can be written as
 Q M  Q M  , -
{t }2Q
=1
M+1
= ζ u ,
n 1 1 =1 , ζ u ,
e 2 2 =1 , s , we can derive 0
Q M 0
Q M
  (1 − s)  − ζu n ,1 + ζu e ,2
2
S2,2 as (36) to complete the proof. 1 1 =1 2 =1
I4 = , -
A PPENDIX D
e ln 2 S 0
Q M 0
Q M
 1− ζu n ,1 − ζu e ,2
P ROOF OF P ROPOSITION 2 1 =1 2 =1
⎛ ⎞
P N SCσ,λ̄ is defined as 
QM 
QM
 ∞ × ⎝1 + ζu n ,1 + ζu e ,2 ⎠ es ds. (E-5)
P N SCσ,λ̄ = FZ u e ,σ (Z ) f Z u n ,σ (Z ) d Z . (D-1) 1 =1 2 =1
0
Substituting (22) and (24) into (D-1), we obtain that With the help of (E-2), we obtain Iσ,λ̄,i (i = 1, 2). Similarly,
following the same methodology, one can easily derive Iσ,λ̄,3 .

Q M
1 
Q M
1 The proof is now completed.
P N SCσ,λ̄ =        
1 =1 m u n ,s1  m u n ,1 2 =1 m u e ,s2  m u e ,2
 ∞ A PPENDIX F
  P ROOF OF T HEOREM 5
× Hu n ,C D F1,2 − Hu n ,C D F1 ,0
0  The CDF of F1 is given as [43, eq. 12]
× Hu e ,P D F1 ,2 − Hu e ,P D F1 ,0 d Z
FF1 (F1 )
 P2,2 − P2,0 − P0,2 + P0,0 . (D-2)
m F1 m F1 −1 F1 m F1
By changing the order of integration, we can express the =   m mF
integral part of P p,q as B m F1 , m s F1 m s F1 − 1 F1 γ̄F1 1
, -
0 0 −m F1 F1

Q M Q M
∞ ζu n ,1 + ζu e ,2 +1 × 2 F1 m F1 , m F1 +m s F1 , m F1 +1;   mF .
I3 = Z 1 =1 2 =1
d Z, (D-3) m s F1 −1 γ̄F1 1
0 (F-1)
which can be solved using the final value theorem as
[53, eq. (34)]. Thus, (41) is obtained, which completes the Let us define Y = F1 /D, the CDF of Y can be formulated as
proof.  r ασ
2
FY (y) = P {F1 ≤ y D} = FF1 (F1 y) f D (F1 )dF1 . (F-2)
r0ασ
A PPENDIUX E
P ROOF OF P ROPOSITION 3 Substituting (F-1) and (A-4) into (F-2) and using
ASCσ,λ can be expressed as [21, eq. (9.113)], we obtain


2 2
  FY (y)
ASCσ,λ = Iσ,λ,i −Iσ,λ,3= A2,2,i−A2,0,i−A0,2,i+A0,0,i −m F
m F1 m F1 −1 y m F1 γ̄F1 1
i=1
 i=1 =     m
− A2,2,3 − A2,0,3 − A0,2,3 + A0,0,3 , (E-1) ασ r22 − r02 B m F1 , m s F1 m s F1 − 1 F1
 
 m F1 + 1 1
where ×    
⎧ 1∞  m F1  m F1 + m s F1 π j
⎨Iσ,λ̄,1 = 10 log2 (1 + Z) f Z u n ,σ (Z ) FZ u e ,σ (Z ) d Z ,    ⎛ ⎞t
∞  −m F
m F1 y γ̄F1 1
I = log2 (1 + Z ) f Z u e ,σ (Z ) FZ u n ,σ (Z ) d Z , (E-2)  m F1 + t  m F1 + m s F1 + t
⎩ σ,λ̄,2 10∞ ×   ⎝ ⎠
Iσ,λ̄,3 = 0 log2 (1 + Z) f Z u e ,σ (Z ) d Z L  m F1 + 1 + t m s F1 − 1
and  r ασ
2
F1 t +m F1 + ασ −1 dF1 dt,
2

Q M 
Q M × (F-3)
1 1 r0ασ
AS Rσ,λ̄,i =           
1 =1 m u n ,s1  m u n ,1 2 =1  m u e ,s2  m u e ,2 I F1

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on June 08,2024 at 02:18:52 UTC from IEEE Xplore. Restrictions apply.
 3494 IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY, VOL. 16, 2021

⎛    ⎞ ⎛   ⎞s   2 
  (−s)  −  −
A m s F1 − 1  1, 1 + 2
1 A m − 1 m s F s α s
 ⎝ 1 s F1
3,0 ⎝ 1 σ
G 2,0
1 ασ ⎠ = ⎠   ds. (G-1)
ασ −m F1  0, m , 2 ασ −m F1
rh̄ m F1 z γ̄F1  s F1 ασ 2π j L rh̄ m F1 z γ̄F1  (1 − s)  1 + ασ − s
2

, -m F * +2  

2
xm F 
1 1
FX (x) =   * +
m s F − 1 γ̄F 2π j L2 L1 2 
0 
=1  m F + s
=1
* +
    0
2
 m F + s  m s F + m F + s  (−s )  s + m F , -s

2
=1 xm F
× * +   ds1 ds2 . (H-1)
    0
2 m s F − 1 γ̄F
=1  m s F  m F  s + m F + 1
=1

where the integration path of L goes from t − inf j to [2] X. Yu, V. Jamali, D. Xu, D. W. K. Ng, and R. Schober, “Smart
t + inf j and t ∈ R. I F1 can be solved easily. With the and reconfigurable wireless communications: From IRS modeling
help of [21, eq. (9.301)], [21, eq. (9.31.5)], [21, eq. (9.113)], to algorithm design,” 2021, arXiv:2103.07046. [Online]. Available:
http://arxiv.org/abs/2103.07046
[21, eq. (8.384.1)] and [21, eq. (8.331.1)], FY (y) can be [3] Q. Wu and R. Zhang, “Intelligent reflecting surface enhanced wireless
expressed by the Meijer’s G-function. Let Z υ,σ  = A1 Y , network via joint active and passive beamforming,” IEEE Trans. Wireless
we obtain (54) that complete the proof. Commun., vol. 18, no. 11, pp. 5394–5409, Nov. 2019.
[4] C. Huang, A. Zappone, G. C. Alexandropoulos, M. Debbah, and
C. Yuen, “Reconfigurable intelligent surfaces for energy efficiency in
A PPENDIX G wireless communication,” IEEE Trans. Wireless Commun., vol. 18, no. 8,
P ROOF OF P ROPOSITION 4 pp. 4157–4170, Aug. 2019.
[5] Q.-U.-A. Nadeem, A. Kammoun, A. Chaaban, M. Debbah, and
With the help of [21, eq. (9.301)], eq. (55) can be expressed M.-S. Alouini, “Large intelligent surface assisted MIMO communica-
as (G-1), shown at the top of the page.
  tions,” 2019, arXiv:1903.08127. [Online]. Available: https://arxiv.org/
m −1 abs/1903.08127
When γF1 → ∞, we have α sF1 −m F → ∞. The [6] X. Chen, D. W. K. Ng, W. H. Gerstacker, and H.-H. Chen, “A survey on
r2 σ m F1 z γ̄F 1
1 multiple-antenna techniques for physical layer security,” IEEE Commun.
Mellin-Barnes integral over L can be calculated approx- Surveys Tuts., vol. 19, no. 2, pp. 1027–1053, 2nd Quart., 2017.
imately by evaluating the residue at −m F according to [7] A. D. Wyner, “The wire-tap channel,” Bell Syst. Tech. J., vol. 54, no. 8,
[49, Theorem 1.7]. Thus, we obtain pp. 1355–1387, Oct. 1975.
⎛ 2 ⎞
[8] H. Du, J. Zhang, K. P. Peppas, H. Zhao, B. Ai, and X. Zhang, “On
   1, 1 + the distribution of the ratio of products of Fisher-Snedecor F random
3,0 ⎜ −1  ασ ⎟
A m variables and its applications,” IEEE Trans. Veh. Technol., vol. 69, no. 2,
G 2,0 ⎜ 1 s F1  ⎟ γ̄F → ∞ pp. 1855–1866, Feb. 2020.
⎝ ασ −m F1  ⎠ −−−1
rh̄ m F1 z γ̄F1  0, m , 2 −−−→ [9] H.-M. Wang, T.-X. Zheng, J. Yuan, D. Towsley, and M. H. Lee,
s F1 “Physical layer security in heterogeneous cellular networks,” IEEE
ασ
 , -m F Trans. Commun., vol. 64, no. 3, pp. 1204–1219, Mar. 2016.
 m s F1 + m F1 r2ασ m F1 y 1 [10] Y.-W.-P. Hong, P.-C. Lan, and C.-C.-J. Kuo, “Enhancing physical-
×   m F1   . (G-2) layer secrecy in multiantenna wireless systems: An overview of signal
m F1 α2 + m F1 γ̄F1 m s F1 − 1 processing approaches,” IEEE Signal Process. Mag., vol. 30, no. 5,
σ
pp. 29–40, Sep. 2013.
Substituting (G-2) into (54), after some algebraic manipu- [11] S. Goel and R. Negi, “Guaranteeing secrecy using artificial noise,” IEEE
Trans. Wireless Commun., vol. 7, no. 6, pp. 2180–2189, Jun. 2008.
lations, we obtain (57) and complete the proof. [12] T.-X. Zheng, H.-M. Wang, and Q. Yin, “On transmission secrecy outage
of a multi-antenna system with randomly located eavesdroppers,” IEEE
A PPENDIX H Commun. Lett., vol. 18, no. 8, pp. 1299–1302, Aug. 2014.
[13] G. Geraci, H. S. Dhillon, J. G. Andrews, J. Yuan, and I. B. Collings,
P ROOF OF T HEOREM 6 “Physical layer security in downlink multi-antenna cellular networks,”
With the help of (52) and (53), the CDF of Z υ,R D can be IEEE Trans. Commun., vol. 62, no. 6, pp. 2006–2021, Jun. 2014.
regarded as the CDF of sum of two Fisher F RVs. Letting [14] H.-M. Wang, C. Wang, D. W. K. Ng, M. H. Lee, and J. Xiao, “Artificial
noise assisted secure transmission for distributed antenna systems,” IEEE
X = F1 + F2 , using [53, eq. (12)] and [49], we can derive Trans. Signal Process., vol. 64, no. 15, pp. 4050–4064, Aug. 2016.
the CDF of X as (H-1), shown at the top of the page. [15] X. Zhou, R. K. Ganti, and J. G. Andrews, “Secure wireless network
Let Y = X/D, hence, we can derive the CDF of Y with the connectivity with multi-antenna transmission,” IEEE Trans. Wireless
help of (A-5). Using Z = A2 Y , we can obtain the CDF of Z Commun., vol. 10, no. 2, pp. 425–430, Feb. 2011.
in terms of Bivariate Meijer’s G-function as (60) to complete [16] X. Yu, D. Xu, and R. Schober, “Enabling secure wireless communica-
tions via intelligent reflecting surfaces,” in Proc. IEEE Global Commun.
the proof. Conf. (GLOBECOM), Dec. 2019, pp. 1–6.
[17] M. Cui, G. Zhang, and R. Zhang, “Secure wireless communication via
R EFERENCES intelligent reflecting surface,” IEEE Wireless Commun. Lett., vol. 8,
no. 5, pp. 1410–1414, Oct. 2019.
[1] J. Zhang, E. Björnson, M. Matthaiou, D. W. K. Ng, H. Yang, and [18] J. Chen, Y.-C. Liang, Y. Pei, and H. Guo, “Intelligent reflecting surface:
D. J. Love, “Prospective multiple antenna technologies for beyond 5G,” A programmable wireless environment for physical layer security,” IEEE
IEEE J. Sel. Areas Commun., vol. 38, no. 8, pp. 1637–1660, Aug. 2020. Access, vol. 7, pp. 82599–82612, Jun. 2019.

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on June 08,2024 at 02:18:52 UTC from IEEE Xplore. Restrictions apply.
 ZHANG et al.: PHYSICAL LAYER SECURITY ENHANCEMENT WITH RIS-AIDED NETWORKS 3495

[19] S. N. Chiu, D. Stoyan, W. S. Kendall, and J. Mecke, Stochastic Geometry [39] A. U. Makarfi, K. M. Rabie, O. Kaiwartya, O. S. Badarneh, X. Li, and
and Its Applications. Hoboken, NJ, USA: Wiley, 2013. R. Kharel, “Reconfigurable intelligent surface enabled IoT networks in
[20] M. Haenggi, Stochastic Geometry for Wireless Networks. Cambridge, generalized fading channels,” in Proc. IEEE Int. Conf. Commun. (ICC),
U.K.: Cambridge Univ. Press, 2012. Jun. 2020, pp. 1–6.
[21] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and [40] T. J. Cui, S. Liu, and L. Zhang, “Information metamaterials and
Products, 7th ed. New York, NY, USA: Academic, 2007. metasurfaces,” J. Mater. Chem. C, vol. 5, no. 15, pp. 3644–3668, 2017.
[22] X. Yu, D. Xu, Y. Sun, D. W. K. Ng, and R. Schober, “Robust and secure [41] D. Dardari, “Communicating with large intelligent surfaces: Fundamen-
wireless communications via intelligent reflecting surfaces,” IEEE J. Sel. tal limits and models,” IEEE J. Sel. Areas Commun., vol. 38, no. 11,
Areas Commun., vol. 38, no. 11, pp. 2637–2652, Nov. 2020. pp. 2526–2537, Nov. 2020.
[23] Z. Wang, L. Liu, and S. Cui, “Channel estimation for intelligent [42] E. Bjornson and L. Sanguinetti, “Power scaling laws and near-field
reflecting surface assisted multiuser communications: Framework, algo- behaviors of massive MIMO and intelligent reflecting surfaces,” IEEE
rithms, and analysis,” IEEE Trans. Wireless Commun., vol. 19, no. 10, Open J. Commun. Soc., vol. 1, pp. 1306–1324, 2020.
pp. 6607–6620, Oct. 2020. [43] S. K. Yoo et al., “A comprehensive analysis of the achievable chan-
[24] G. T. de Araujo, A. L. F. de Almeida, and R. Boyer, “Channel estimation nel capacity in F composite fading channels,” IEEE Access, vol. 7,
for intelligent reflecting surface assisted MIMO systems: A tensor pp. 34078–34094, Mar. 2019.
modeling approach,” IEEE J. Sel. Topics Signal Process., vol. 15, no. 3, [44] J. Wang, L. Kong, and M.-Y. Wu, “Capacity of wireless ad hoc networks
pp. 789–802, Apr. 2021. using practical directional antennas,” in Proc. IEEE Wireless Commun.
[25] C. Liu, X. Liu, D. W. K. Ng, and J. Yuan, “Deep residual learning Netw. Conf., Apr. 2010, pp. 1–6.
for channel estimation in intelligent reflecting surface-assisted multi- [45] Q. Wu and R. Zhang, “Beamforming optimization for wireless network
user communications,” 2020, arXiv:2009.01423. [Online]. Available: aided by intelligent reflecting surface with discrete phase shifts,” IEEE
http://arxiv.org/abs/2009.01423 Trans. Commun., vol. 68, no. 3, pp. 1838–1851, Mar. 2020.
[26] A. Mukherjee, S. A. A. Fakoorian, J. Huang, and A. L. Swindlehurst, [46] X. Yu, D. Xu, and R. Schober, “Optimal beamforming for MISO com-
“Principles of physical layer security in multiuser wireless networks: munications via intelligent reflecting surfaces,” in Proc. IEEE 21st Int.
A survey,” IEEE Commun. Surveys Tuts., vol. 16, no. 3, pp. 1550–1573, Workshop Signal Process. Adv. Wireless Commun. (SPAWC), May 2020,
3rd Quart., 2014. pp. 1–5.
[27] L. Yang, J. Yang, W. Xie, M. O. Hasna, T. Tsiftsis, and M. D. Renzo, [47] A. M. Elbir, A. Papazafeiropoulos, P. Kourtessis, and S. Chatzinotas,
“Secrecy performance analysis of RIS-aided wireless communication “Deep channel learning for large intelligent surfaces aided mmWave
systems,” IEEE Trans. Veh. Technol., vol. 69, no. 10, pp. 12296–12300, massive MIMO systems,” IEEE Wireless Commun. Lett., vol. 9, no. 9,
Oct. 2020. pp. 1447–1451, Sep. 2020.
[28] M. Haenggi, J. G. Andrews, F. Baccelli, O. Dousse, and [48] H. Du, J. Zhang, J. Cheng, and B. Ai, “Millimeter wave communica-
M. Franceschetti, “Stochastic geometry and random graphs for the tions with reconfigurable intelligent surfaces: Performance analysis and
analysis and design of wireless networks,” IEEE J. Sel. Areas Commun., optimization,” IEEE Trans. Commun., vol. 69, no. 4, pp. 2752–2768,
vol. 27, no. 7, pp. 1029–1046, Sep. 2009. Apr. 2021.
[29] S. Singh, M. N. Kulkarni, A. Ghosh, and J. G. Andrews, “Tractable [49] A. Mathai, R. K. Saxena, and H. Haubold, The H Function: Theory and
model for rate in self-backhauled millimeter wave cellular networks,” Applications. New York, NY, USA: Springer-Verlag, 2010.
IEEE J. Sel. Areas Commun., vol. 33, no. 10, pp. 2196–2211, Oct. 2015. [50] Y. A. Rahama, M. H. Ismail, and M. S. Hassan, “On the sum of
[30] M. Di Renzo, “Stochastic geometry modeling and analysis of multi- independent Fox’s H -function variates with applications,” IEEE Trans.
tier millimeter wave cellular networks,” IEEE Trans. Wireless Commun., Veh. Technol., vol. 67, no. 8, pp. 6752–6760, Aug. 2018.
vol. 14, no. 9, pp. 5038–5057, Sep. 2015. [51] M. Bloch, J. Barros, M. R. D. Rodrigues, and S. W. McLaughlin, “Wire-
[31] T. Bai and R. W. Heath, Jr., “Coverage and rate analysis for millimeter- less information-theoretic security,” IEEE Trans. Inf. Theory, vol. 54,
wave cellular networks,” IEEE Trans. Wireless Commun., vol. 14, no. 2, no. 6, pp. 2515–2534, Jun. 2008.
pp. 1100–1114, Feb. 2015. [52] L. Kong and G. Kaddoum, “On physical layer security over the
[32] T. Hou, Y. Liu, Z. Song, X. Sun, Y. Chen, and L. Hanzo, “MIMO Fisher-Snedecor F wiretap fading channels,” IEEE Access, vol. 6,
assisted networks relying on intelligent reflective surfaces,” 2019, pp. 39466–39472, Jun. 2018.
arXiv:1910.00959. [Online]. Available: http://arxiv.org/abs/1910.00959 [53] H. Du, J. Zhang, K. P. Peppas, H. Zhao, B. Ai, and X. Zhang, “Sum of
[33] Z. Ding, F. Adachi, and H. V. Poor, “The application of MIMO to non- Fisher-Snedecor F random variables and its applications,” IEEE Open
orthogonal multiple access,” IEEE Trans. Wireless Commun., vol. 15, J. Commun. Soc., vol. 1, pp. 342–356, Mar. 2020.
no. 1, pp. 537–552, Jan. 2016. [54] Z. Zhang, C. Zhang, C. Jiang, F. Jia, J. Ge, and F. Gong, “Improv-
[34] M. R. Akdeniz et al., “Millimeter wave channel modeling and cellular ing physical layer security for reconfigurable intelligent surface aided
capacity evaluation,” IEEE J. Sel. Areas Commun., vol. 32, no. 6, NOMA 6G networks,” IEEE Trans. Veh. Technol., early access,
pp. 1164–1179, Jun. 2014. Mar. 24, 2021, doi: 10.1109/TVT.2021.3068774.
[35] W. Tang et al., “Wireless communications with reconfigurable intelligent [55] M. Shah, “On generalizations of some results and their applications,”
surface: Path loss modeling and experimental measurement,” IEEE Collectanea Math., vol. 24, no. 3, pp. 249–266, 1973.
Trans. Wireless Commun., vol. 20, no. 1, pp. 421–439, Jan. 2021. [56] M. Jung, W. Saad, Y. Jang, G. Kong, and S. Choi, “Performance analysis
[36] G. Nauryzbayev, K. M. Rabie, M. Abdallah, and B. Adebisi, “On the of large intelligent surfaces (LISs): Asymptotic data rate and channel
performance analysis of WPT-based dual-hop AF relaying networks in hardening effects,” IEEE Trans. Wireless Commun., vol. 19, no. 3,
α - μ fading,” IEEE Access, vol. 6, pp. 37138–37149, Jun. 2018. pp. 2052–2065, Mar. 2020.
[37] K. Rabie, B. Adebisi, G. Nauryzbayev, O. S. Badarneh, X. Li, and [57] Q. Tao, J. Wang, and C. Zhong, “Performance analysis of intelligent
M.-S. Alouini, “Full-duplex energy-harvesting enabled relay networks reflecting surface aided communication systems,” IEEE Commun. Lett.,
in generalized fading channels,” IEEE Wireless Commun. Lett., vol. 8, vol. 24, no. 11, pp. 2464–2468, Nov. 2020.
no. 2, pp. 384–387, Apr. 2019. [58] O. S. Badarneh, S. Muhaidat, P. C. Sofotasios, S. L. Cotton, K. Rabie,
[38] S. K. Yoo, S. L. Cotton, P. C. Sofotasios, M. Matthaiou, M. Valkama, and B. D da Costa, “The N∗ Fisher-Snedecor F cascaded fading model,”
and G. K. Karagiannidis, “The Fishe-Snedecor F distribution: A simple in Proc. IEEE WiMbo, Oct. 2018, pp. 1–7.
and accurate composite fading model,” IEEE Commun. Lett., vol. 21, [59] Wolfram. The Wolfram Functions Site. Accessed: May 2021. [Online].
no. 7, pp. 1661–1664, Jul. 2017. Available: http://functions.wolfram.com

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on June 08,2024 at 02:18:52 UTC from IEEE Xplore. Restrictions apply.