1192 IEEE WIRELESS COMMUNICATIONS LETTERS, VOL. 10, NO.

6, JUNE 2021

Random Shifting Intelligent Reflecting Surface

for OTP Encrypted Data Transmission
Zijie Ji , Phee Lep Yeoh , Member, IEEE, Gaojie Chen , Senior Member, IEEE,
Cunhua Pan , Member, IEEE, Yan Zhang , Member, IEEE, Zunwen He , Member, IEEE,
Hao Yin, and Yonghui Li , Fellow, IEEE

Abstract—In this letter, we propose a novel encrypted data artificial noise vectors, whilst IRS based secret key generation
transmission scheme using an intelligent reflecting surface (IRS) remains unexplored.
to generate secret keys in wireless communication networks. We Physical layer secret key generation is an alternative security
show that perfectly secure one-time pad (OTP) communications approach to computational complexity based encryption [7],
can be established by using a simple random phase shifting of where secret keys are generated by legitimate transceivers
the IRS elements. To maximize the secure transmission rate,
we design an optimal time slot allocation algorithm for the
using their reciprocal time-variant channels, while preventing
IRS secret key generation and the encrypted data transmission eavesdroppers from inferring the generated keys through its
phases. Moreover, a theoretical expression of the key genera- own wireless channel observations due to spatial decorrela-
tion rate is derived based on Poisson point process (PPP) for the tion. Notably, the key generation rate (KGR) is limited by the
practical scenario when eavesdroppers’ channel state information movement speed of the legitimate transceivers, which makes
(CSI) is unavailable. Simulation results show that employing our it impractical for stationary or slow-moving wireless devices.
IRS-based scheme can significantly improve the encrypted data To achieve secure one-time pad (OTP) [8], artificial random-
transmission performance for a wide-range of wireless channel ness is introduced to boost the KGR. In [9], a method to
gains and system parameters. induce randomness was proposed by selecting different local
Index Terms—Intelligent reflecting surface, one-time pad, pilot constellations. In [10], private random precoding vec-
physical layer security, reconfigurable intelligent surface, secret tors were designed to induce randomness in multiple-input
key generation. multiple-output (MIMO) systems. The authors in [11] utilized
artificial noise to scramble eavesdropping channels. We note
that all the above contributions focused on increasing ran-
I. I NTRODUCTION domness at transceiver ends, rather than increasing the rate
NTELLIGENT reflecting surface (IRS) is a promising can- of channel randomness in the wireless environment.
I didate technology for future mobile communication systems
due to its low-cost deployment and high spectral and energy
In this letter, we consider the use of IRS to induce virtual
fast fading channels, where artificial randomness is introduced
efficiency [1], [2]. By adaptively adjusting phase shifts of in the propagation environment. Compared to the beamform-
large-scale passive reflecting elements, IRS can reconfigure the ing based scheme [12] which requires high IRS hardware
electromagnetic propagation environment of wireless devices, accuracy, we propose a simpler random phase shifting scheme
thereby improving their communication performance [3]. for secret key generation, which only relies on the fast phase
Recently, physical layer security in IRS assisted networks has switching to support OTP encrypted data transmission. To pro-
also attracted extensive attention [4]–[6]. However, most of vide detailed insights, we derive the secure transmission rate
these contributions focused on designing beamforming and and KGR of our proposed scheme. Based on the maximum
transmission rate, we develop an optimal time slot alloca-
Manuscript received December 29, 2020; revised February 17, 2021; tion algorithm for the key generation and data transmission
accepted February 20, 2021. Date of publication February 23, 2021; date phases. For randomly distributed eavesdroppers whose chan-
of current version June 9, 2021. This work was supported in part by nel state information (CSI) is unknown, we further consider
the National Key Research and Development Program of China under
Grant 2020YFB1804901; in part by the National Natural Science Foundation the use of a Poisson point process (PPP) model to derive
of China under Grant 61871035; in part by the China Scholarship Council the KGR that is only related to the distribution intensity of
Scholarship; and in part by Ericsson Company. The associate editor coordinat- eavesdroppers. Finally, the advantages of our scheme and the
ing the review of this article and approving it for publication was J. Harshan. validity of the analysis is verified through simulations. We
(Corresponding author: Yan Zhang.)
Zijie Ji, Yan Zhang, and Zunwen He are with the School of Information and
show that the secure transmission rate is enhanced signifi-
Electronics, Beijing Institute of Technology, Beijing 100081, China (e-mail: cantly by random phase shifted IRS compared with no IRS and
[email protected]; [email protected]; [email protected]). IRS with fixed phase shifts. The impacts of important system
Phee Lep Yeoh and Yonghui Li are with the School of Electrical and design parameters including the length of coherence interval,
Information Engineering, University of Sydney, Sydney, NSW 2006, Australia the number of IRS elements, signal-to-noise ratio (SNR), the
(e-mail: [email protected]; [email protected]).
Gaojie Chen is with the Department of Engineering, University of Leicester, distribution intensity and distribution radius of eavesdroppers,
Leicester LE1 7RH, U.K. (e-mail: [email protected]). are highlighted in the simulation results.
Cunhua Pan is with the School of Electronic Engineering and Computer
Science, Queen Mary University of London, London E1 4NS, U.K. (e-mail:
[email protected]). II. S YSTEM M ODEL
Hao Yin is with the Institute of China Electronic System Engineering
Corporation, Beijing 100141, China (e-mail: [email protected]). Our proposed system model consists of a single-antenna
Digital Object Identifier 10.1109/LWC.2021.3061549 base station (Alice) and a single-antenna user (Bob) wanting
2162-2345 
c 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: Jaypee Insituite of Information Technology-Noida Sec 128 (L3). Downloaded on December 10,2021 at 17:50:53 UTC from IEEE Xplore. Restrictions apply.
 JI et al.: RANDOM SHIFTING IRS FOR OTP ENCRYPTED DATA TRANSMISSION 1193

the equivalent channels change randomly in each round, which

helps to significantly increase the KGR. The equivalent chan-
nels estimated at each node can be calculated from their
observations as
(q)H  
(q) yb,1 s1 (q) H (q)
hB = √ = gab +ẑb = hab +hrb Θ har +ẑb ,
P s1 22
(q)H  
(q) ya,2 s2 (q) H (q)
Fig. 1. Time slot allocation for the IRS-assisted secret key generation phase hA = √ = gba +ẑa = hba +hra Θ hbr +ẑa ,
and the OTP encrypted data transmission phase. P s2 22
(q)H  
(q) yek ,1 s1 (q) H (q)
to exchange data securely over the public wireless channels, hAEk = √ = gae +ẑe k = hae k +hre Θ har +ẑek ,
P s1 22
k k
and K single-antenna eavesdroppers (Eves) attempting to cap-
(q)H  
ture the transmitted information. Alice and Bob exploit their (q) ye ,2 s2 (q)
reciprocal channels in time-division duplex (TDD) mode to hBEk = √ k 2
= gbek +ẑek = hbek +hH
rek Θ
(q)
hbr +ẑek , (3)
P s2 2
generate keys, then encrypt and decrypt the transmitted data
using the OTP approach. We assume that Alice is fixed and where ẑi ∼ CN (0, σ̂i2 ) denotes the estimation error at
Bob is a slow-moving user which incurs a relative long coher- node i, i ∈ {a, b, ek }, ·2 is the Euclidean norm, and
ence interval and thus a limited KGR. To significantly increase σ̂i2 = σi2 /P . Assume that σa2 = σb2 = σe2k = σ 2 ,
the KGR, we propose to use an N-element IRS (Rose) that (q) (q)
and gab = gba are the reciprocal combined channels
can artificially adjust phase shifts and induce randomness in
for Alice and Bob to generate shared keys. To eliminate
the wireless propagation environment. Considering Bob is a
the impact of amplitude difference caused by hardware, the
mobile user, we assume all Eves are passive and located close
estimations are linearly normalized before quantization as
to Alice so that their observed channels can be correlated with (q) (q)
the legitimate ones. h̃I = hI /hI 2 , I ∈ {A, B, AEk , BEk }, where hI =
(1) (2) (Q)
[hI , hI , . . . , hI ]. Given these channels, the maximum
III. R ANDOM S HIFTING IRS FOR OTP T RANSMISSION achievable KGR can be expressed as [7], [10], [13]
⎛ ⎞
Fig. 1 shows our scheme which is composed of two phases: Q  
1 (q) (q) (q) (q)
secret key generation and encrypted data transmission. Firstly, RSKG = min⎝ I h̃A ; h̃B |h̃AE , h̃BE ⎠, (4)
k ∈K 2QΔT k k
Alice and Bob exchange pilots during the first 2Q time slots, q=1
assisted by Rose which simply reflects the transmissions with
a random set of phase shifts in each uplink and downlink where K = {1, 2, . . . , K }, and ΔT is the duration of one time
round. During the odd time slots, Alice sends a pilot sequence slot. Here, we consider the average KGR per unit time.
distributed on different sub-carriers s1 ∼ CN (0, I), and the Theorem 1: When the number of IRS elements N is suffi-
received signal at Bob or the kth Eve can be expressed as ciently large,1 the maximum achievable KGR in our proposed
(q) √  (q)
 scheme can be given by
yi,1 = P hai + hH ri Θ har s1 + zi , i ∈ {b, ek }, (1)
(k )
RSKG = min (RSKG )
k ∈K
where hai ∈ C1×1 , har ∈ CN ×1 ,
and hri ∈ CN ×1
denote the ⎡ ⎛  2 ⎞⎤
direct channel from Alice to node i, the incident channel from
1 1 − ρ2Ek
⎢ ⎜ ⎟⎥
Alice to Rose, and the reflected channel from Rose to Bob or = min ⎣ log2 ⎝ ⎠⎦, (5)
Eve, respectively. Let Θ(q) = diag(e j θ1,q , e j θ2,q , . . . , e j θN ,q ) k ∈K 2ΔT 1 + 2ρL ρ2E − 2ρ2E − ρ2L
k k
denote the diagonal IRS phase shifting matrix in the qth round
of channel training, where q = 1, 2, . . . , Q. θn,q denotes the (q) (q)∗ (q) (q)∗
where ρEk = E{h̃A h̃BE } = E{h̃B h̃BE }, ρL =
k k
discrete phase shift introduced by the nth reflecting element, (q) (q)∗ (q)
which is generated based on an independent random vari- E{h̃A h̃B } denote the cross-correlations between h̃A (or
(q) (q) (q) (q)
able (RV) uniformly distributed in a set of discrete values h̃B ) and h̃BE , and that between h̃A and h̃B , respectively.
k
(2B −1)2π
{0, 22π
B ,..., 2B
}, where B is the number of quantiza- E{·} represents the expectation with respect to (w.r.t.) the Q
tion bits for phase shifts. In (1), P is the transmit power, samples.
and zi ∼ CN (0, σi2 I) denotes the independent and identi- Proof: By invoking the central limit theorem, the composite
cally distributed (i.i.d.) complex additive white Gaussian noise channels can be approximated by the Gaussian distribution for
vector. a sufficiently large number of N [1]. Thus, the conditional
Similarly, during the even time slots, Bob transmits s2 ∼ mutual information can be calculated as [11]
CN (0, I) with the same power, and Alice or Eve receives  
(q) (q) (q) (q)
√   I h̃A ; h̃B |h̃AE , h̃BE
k k
(q)
yi,2 = P hbi + hH (q)    
ri Θ hbr s2 + zi , i ∈ {a, ek }, (2) (q) (q) (q) (q) (q) (q) (q)
= H h̃A |h̃AE , h̃BE − H h̃A |h̃B , h̃AE , h̃BE
where hbi ∈ C1×1 and hbr ∈ CN ×1 are the channel  k k
   k k

det WAAEk BEk det WBAEk BEk

from Bob to node i and Rose, respectively. We assume = log2    , (6)
det WAEk BEk det WABAEk BEk
that all involved channels are quasi-static. They remain con-
stant in each coherence interval and change independently 1 IRS is commonly equipped with tens or hundreds of elements to achieve
between different coherence intervals. The IRS ensures that high performance [2], so this assumption is reasonable.

Authorized licensed use limited to: Jaypee Insituite of Information Technology-Noida Sec 128 (L3). Downloaded on December 10,2021 at 17:50:53 UTC from IEEE Xplore. Restrictions apply.
 1194 IEEE WIRELESS COMMUNICATIONS LETTERS, VOL. 10, NO. 6, JUNE 2021

where det(·) is the matrix determinant, and Algorithm 1 Proposed Optimal Time Slot Allocation
⎧⎛ ⎞ ⎫ 1: Let Rose be in the receiving mode. Alice and Bob alternatively send pilot
⎪ (q) ⎪ signals so that Rose can estimate har and hH
⎪ h̃
⎨⎜ A ⎟ ⎪
⎬ rb . Alice estimates hab and
(q) ⎟ (q)∗ (q)∗ (q)∗ sends its phase to Rose, then Θ and RMRT can be calculated based
⎜
WAAEk BEk = E ⎝ h̃AE ⎠ h̃A h̃AE h̃BE on (10).
⎪
⎪ k k k ⎪
⎪ 2: Initialize Q (t) = Qth .
⎩ h̃ (q) ⎭ (t)
BEk
⎡ ⎤ 3: repeat (Observation accumulation for RSKG )
KAA KAAEk KABEk 4: For given Q (t) , let Rose be in the reflecting mode with randomly
= ⎣ KAEk A KAEk AEk KAEk BEk ⎦. (7) (q) (q)
selected Θ(q) . Alice and Bob collect channel observations hA , hB ,
KBEk A KBEk AEk KBEk BEk (q) (t)
and hBE and normalize them. The corresponding RSKG can be
k
computed according to (5).
(q) (q)∗ Initialize Qmin = Q (t) and Qmax = L/2 − 1.
KIJ = E{h̃I h̃J }, I , J ∈ {A, B, AEk , BEk } is the 5:
correlation function. When two nodes are more than half- 6: repeat (Bisection search for Q (t) )
7: Set Q = (Qmin + Qmax )/2, and calculate ΔR =
wavelength apart, their observed channels can be deemed (t)
QRSKG −(L−2Q)RMRT . If ΔR ≤ 0, update Qmin = Q; otherwise
uncorrelated [9]. Hence, here we only consider ρEk and ρL for update Qmax = Q.
simplicity. Since the channel observations of two nodes sepa- 8: until (Qmax − Qmin ) ≤ 1. Record this Q as Q (t) .
rated by more than half of the wavelength can be considered 9: Update Q (t) = Q (t) + 1.
uncorrelated [9], [13] and all channels have been normalized, 10: until Q (t) ≥ Q (t−1) . Finally, we obtain Q  = Q (t−1) .
we have
 
  1 0 ρEk
det WAAEk BEk = det 0 1 0 = 1 − ρ2Ek . (8) remaining (L − 2Q) time slots is used for encrypted data trans-
ρEk 0 1 mission. We observe that there exists a tradeoff between the
two phases: if Q is too small, the generated keys will be insuf-
Similarly, the determinants of other matrices in (6) can be
ficient to encrypt all the data to be transmitted; whereas if Q is
given as det(WBAEk BEk ) = 1 − ρ2Ek , det(WAEk BEk ) = 1, too large, the remaining time slots for data transmission will
and det(WABAEk BEk ) = 1 + 2ρL ρ2Ek − 2ρ2Ek − ρ2L . In addi- be insufficient and thus some generated keys will be wasted.
tion, since the phase shift coefficients are randomly selected, Therefore, the secure transmission throughput in our proposed
the autocorrelation between two channels sampled in different scheme depends on the minimum of the generated key bits and
(q ) (q ) the transmitted data bits, where the secure transmission rate is
rounds denoted as h̃I 1 and h̃I 2 is expressed as
derived as
N 
(q1 ,q2 )
σh2ij + σh2rj σh2ir n=1 E{e j θn,q1 }E{e −j θn,q2 } RSKG Q/L, if α ≤ (L − 2Q)/Q,
ρI = 2 2 2
, (9) REDT =
RMRT (L − 2Q)/L, otherwise,
(11)
σh + N σh σh
ij rj ir
where α = RSKG /RMRT is the rate ratio. To achieve the
where σh2ij , σh2rj , and σh2ir are the average power of hij , hrj ,n , highest secure transmission rate, Q should be optimized so that
and hir,n (the nth element of hrj and hir ), i , j ∈ {a, b,ek }, the length of generated keys and the length of transmitted data
respectively. Since θn,q ∼ U(0, 2π), E{e j θn,q } = 0. If N  are closest to each other, i.e., Q  = arg minQ∈Q |QRSKG −
(q ,q ) (L − 2Q)RMRT |, where Q = {Qth , . . . , L/2 − 1} is the set
σh2ij /(σh2rj σh2ir ), it shows that ρI 1 2 → 0. This condition
always holds under our assumption on N, especially when the of Q, and Qth is the threshold to ensure a sufficient number of
IRS reflected channels are strong. Thus, the channel samples samples,2 and thereby the capacity of the proposed scheme can
in different rounds can be deemed as independent RVs, and be expressed as CEDT = REDT (Q  ). In practice, because Q
the maximum achievable KGR can be computed by (5) based is an integer, given the expression of RSKG and RMRT , we can
on the weak law of large numbers. find Q  by using the optimization approach in Algorithm 1.
Based on (5), Alice and Bob can determine relevant param- All channel knowledge related to calculation is acquired and
eters and generate keys. We note that standard key generation processed at the IRS end as described in Step 1 of Algorithm 1,
processing such as equally likely quantization, reconciliation while Alice and Bob only need to generate keys based on their
and privacy amplification should be considered as in [7]. channel observations and the value of Q  sent from the IRS.
After key generation, the data to be transmitted is encrypted
by XORing with the generated keys using the OTP approach, IV. P ERFORMANCE A NALYSIS FOR R ANDOMLY
and then we consider that the ciphertext is transmitted at D ISTRIBUTED E AVESDROPPERS
the maximum rate for the IRS-assisted system using maxi- (k )
We find that RSKG in (5) depends on the minimum RSKG ,
mum rate transmission (MRT) as considered in [1]. We note (q) (q)
that since the keys have been shared between Alice and which is computed based on knowledge of h̃A , h̃B , and
(q) (q)
Bob, this encrypted data transmission stage can be either h̃BE , where h̃BE is typically assumed to be known at Alice
k k
uplink or downlink. Taking the downlink for an example, the and Bob when the untrusted nodes are other users in the same
transmission rate is given by network. However, we note that the CSI of passive and mali-
1   cious Eves may be challenging to estimate and may not be
RMRT = log2 1 + γb |hab + hH
rb Θ har |2 , (10) perfectly known at Alice and Bob. To this end, in this section
ΔT
2 Although infinite samples are required to achieve R
where γb = P /σb2 is the reference SNR and Θ is the optimal SKG theoreti-
cally [14], in practice we only need to ensure the number of samples is
IRS phase shift matrix defined in [1, (19)]. Let L denote the sufficiently large to obtain reliable statistics according to the weak law of
total number of symbols in each coherence interval, thus the large numbers.

Authorized licensed use limited to: Jaypee Insituite of Information Technology-Noida Sec 128 (L3). Downloaded on December 10,2021 at 17:50:53 UTC from IEEE Xplore. Restrictions apply.
 JI et al.: RANDOM SHIFTING IRS FOR OTP ENCRYPTED DATA TRANSMISSION 1195

we focus on the performance analysis of our proposed scheme

when there exists randomly distributed Eves.
Since the transmitted data has been XORed by OTP keys,
Eves will not impact on the encrypted transmission stage. Only
key generation can be compromised in the presence of Eves.
From (5), we can derive that
  
(k ) ln 2 ρ 2 −ρ 1 − ρ 2
∂RSKG Ek L Ek
2
=   < 0,
∂ρE ΔT 1 + ρL − 2ρ2E
k k
 
(k ) ln 2 ρL − ρEk2
∂RSKG
=   > 0, (12) Fig. 2. Encrypted transmission capacity CEDT versus transmit power P for
∂ρL ΔT (1 − ρL ) 1 + ρL − 2ρ2E different schemes, number of quantization bits B and number of elements N.
k

when ρL > ρ2Ek holds. This condition is always satisfied

because noise is the only factor that results in the decrease With this closed-form expression, we can investigate the influ-
(q) (q)
of ρL = E{|gab |2 }/(E{|gab |2 } + σ̂ 2 ), which shows the high ence of randomly distributed Eves on the security performance
(q) (q) (q)
similarity between observations h̃A and h̃B of the reciprocal of our proposed scheme. Correspondingly, hBE in Step 4 of
k
(q) Algorithm 1 is no longer required, and Alice and Bob only
channel gab , but ρEk is weakened by the difference between
(q) (q) need to determine a suitable λE according to their estimation
the combined channels gab and gbe in addition to noise. of the surrounding network conditions.
k
(k )
Hence, RSKG decreases as ρEk increases or ρL decreases.
For a given SNR, ρL is fixed, while ρEk is affected by
the positions of Eves. Since Eves are passive, we assume that V. S IMULATION R ESULTS AND D ISCUSSION
their exact locations are not known. Therefore, we consider In this section, simulation results are presented to show
a PPP to model the locations of randomly distributed Eves the effectiveness of the proposed scheme by comparing with
and analyze the impact on RSKG . Generally, a closer distance two benchmarks: (1) without IRS, (2) IRS with fixed optimal
between Alice and Eve k, dAEk , is associated with a stronger phases. We assume that the distance between Alice and Bob
correlation ρEk , so we need to find the expectation of the is dAB = 100 m. The central point of the IRS is located at a
closest distance dmin , which corresponds to the largest ρEk vertical distance of d1 = 5 m to the line that connects Alice
and RSKG . Specifically, we model the locations of Eves as a and Bob, and the horizontal distance between the IRS and Bob
homogenous PPP in the plane denoted by ΦE with intensity is set as d2 = 5 m. All Eves are randomly deployed within
λE . The probability density function (PDF) of dAEk of the a circle of radius 1 m centered at Alice and their distances
kth nearest Eve to Alice is expressed as [15] to Bob and Rose can be calculated correspondingly. The path
2k −1 loss is given by PL = (PL0 + 10ζlog10 (d/d0 )) dB, where
2
−λE πdAE 2λkE π k dAE PL0 = 30 dB is the path loss at reference distance d0 = 1 m,
k
fdAE (dAEk ) = e k , (13)
k Γ(k ) ζ is the path loss exponent, and d is the distance between the
transmitter and the receiver. The path loss exponents for the
where Γ(·) is the gamma function. As such, the expectation
Alice-Rose link (Eve-Rose link), Rose-Bob link, and Alice-
of dmin (i.e., dAE1 ) can be calculated as
 ∞ Bob link (Eve-Bob link) are ζAR = ζER = 2.2, ζRB = 2.5,
2 2 and ζAB = ζEB = 3.5, respectively. The small-scale fading of
EΦE (dmin ) = 2λE πdmin e −λE πdmin ddmin all involved channels follows a Rayleigh fading model, and the
0
 π  ∞ 1  correlation between channel coefficients of legitimate channels
2
2
−λE πr 2 1 and eavesdropping channels is characterized by (15). Other
= re dθdr = , (14) parameters are set as: the carrier frequency f c = 1 GHz, the
0 0 4λE
transmit power P = 20 dBm, the noise power σ 2 = −96 dBm,
which indicates that when λE is high, dmin is expected to be ΔT = 1 ms, L = 1000, Qth = 100, N = 50, B = 3, and K = 4
(q) (q) if not specified otherwise.
small. Furthermore, the correlation between gab and gbe can
k
be characterized by [9] Fig. 2 shows the advantages of the proposed scheme over
other benchmarks. IRS with fixed optimal phases yields higher
ρ̃(dAEk ) = [J0 (2πdAEk /λ)]2 , (15) RSKG and RMRT by improving the SNR compared with the
where J0 (·) is the Bessel function of the first kind, and λ is case without IRS, but only one set of keys can be generated
the wavelength. Thus, the KGR in (5) can be simplified as by using the existing channels. However, in our scheme, since
  2  the random IRS shifting is employed, the combined channels
1 1 − ρ2E max leads to Q  times of key generation and therefore CEDT is
RSKG = log , (16)
2ΔT 2 1 + 2ρL ρ2E max − 2ρ2E max − ρ2L increased. In addition, we see that the performance is sig-
nificantly improved when the transmit power P is increased
where the maximum correlation among all ρEk is since the corresponding high SNR benefits to both key gen-
  eration and encrypted data transmission. The larger number
(q) (q)
E{|gab ||gbe1 |}ρ̃ 1/4λE of elements N also results in better performance, but the gain
ρE max = ! ! . (17) brought by the same number of IRS elements will gradually
(q) (q)
E{|gab |2 } + σ̂ 2 E{|gbe1 |2 } + σ̂ 2 decrease as N increases. It can be observed that the number

Authorized licensed use limited to: Jaypee Insituite of Information Technology-Noida Sec 128 (L3). Downloaded on December 10,2021 at 17:50:53 UTC from IEEE Xplore. Restrictions apply.
 1196 IEEE WIRELESS COMMUNICATIONS LETTERS, VOL. 10, NO. 6, JUNE 2021

We note that, similar to previous works such as [1] and [5],

the performance improvement is sensitive to the placement of
the IRS. In our proposed scheme, the IRS should be deployed
near the transceivers since the large-scale fading effect due to
increasing distance will increase the autocorrelation in (9) thus
impair the randomness of keys when the IRS is far away.

VI. C ONCLUSION
We investigated the confidential transmission in IRS assisted
wireless networks with multiple passive eavesdroppers. A new
Fig. 3. Impact of time slot allocation Q on encrypted transmission rate encrypted data transmission scheme was designed, where the
REDT for different coherence interval L and transmit power P. OTP secret keys were provided by random IRS phase shifting.
The KGR was derived based on the assumption that all Eves
were located near Alice, and an optimal time slot allocation
algorithm was proposed to maximize the secure transmission
rate. For practical implementations, we further analyzed the
impact of randomly distributed Eves whose CSI is unavailable.
The effectiveness of the proposed scheme and correctness of
our analysis were validated by the simulation and theoretical
results.

R EFERENCES
[1] Q. Wu and R. Zhang, “Intelligent reflecting surface enhanced wireless
network via joint active and passive beamforming,” IEEE Trans. Wireless
Commun., vol. 18, no. 11, pp. 5394–5409, Nov. 2019.
Fig. 4. Impact of randomly distributed Eves on secret key generation rate [2] S. Gong et al., “Toward smart wireless communications via intelligent
RSKG for different distribution radius R and intensity λE . reflecting surfaces: A contemporary survey,” IEEE Commun. Surveys
Tuts., vol. 22, no. 4, pp. 2283–2314, 4th Quart., 2020.
[3] C. Pan et al., “Multicell MIMO communications relying on intelligent
reflecting surfaces,” IEEE Trans. Wireless Commun., vol. 19, no. 8,
of quantization bits for phase shifts B basically does not affect pp. 5218–5233, Aug. 2020.
CEDT , which indicates that our scheme can still provide good [4] S. Fang, G. Chen, and Y. Li, “Joint optimization for secure intelli-
performance when equipping low-resolution hardware at the gent reflecting surface assisted UAV networks,” IEEE Wireless Commun.
Lett., vol. 10, no. 2, pp. 276–280, Feb. 2021.
IRS.
[5] X. Guan, Q. Wu, and R. Zhang, “Intelligent reflecting surface assisted
Fig. 3 depicts the impact of time slot allocation on REDT as secrecy communication: Is artificial noise helpful or not?” IEEE Wireless
discussed in Section III. At first, as more keys are generated, Commun. Lett., vol. 9, no. 6, pp. 778–782, Jun. 2020.
REDT increases with Q, then as Q further increases, REDT [6] S. Hong, C. Pan, H. Ren, K. Wang, and A. Nallanathan, “Artificial-noise-
decreases due to insufficient time slots assigned for encrypted aided secure MIMO wireless communications via intelligent reflecting
data transmission. The Q corresponding to each peak is Q  , surface,” IEEE Trans. Commun., vol. 68, no. 12, pp. 7851–7866,
Dec. 2020.
which can be found by applying Algorithm 1, and the value of [7] J. Zhang, G. Li, A. Marshall, A. Hu, and L. Hanzo, “A new frontier for
each peak pointed by the arrow is CEDT . Under the same P, IoT security emerging from three decades of key generation relying on
Q  /L basically remains unchanged and the slight performance wireless channels,” IEEE Access, vol. 8, pp. 138406–138446, 2020.
improvement comes from the increased accuracy of allocation [8] G. Li, Z. Zhang, J. Zhang, and A. Hu, “Encrypting wireless communica-
tions on the fly using one-time pad and key generation,” IEEE Internet
proportion. This demonstrates that our scheme can be applied Things J., vol. 8, no. 1, pp. 357–369, Jan. 2021.
to various scenarios with a wide range of coherence intervals [9] N. Aldaghri and H. Mahdavifar, “Physical layer secret key generation
L. When P decreases, the achievable REDT reduces as well, in static environments,” IEEE Trans. Inf. Forensics Security, vol. 15,
which is consistent with the analysis in Fig. 2. Meanwhile, we pp. 2692–2705, Feb. 2020.
note that Q  becomes larger because the SNR impacts more on [10] H. Taha and E. Alsusa, “Secret key exchange using private random
secret key generation, which compromises reciprocity between precoding in MIMO FDD and TDD systems,” IEEE Trans. Veh. Technol.,
vol. 66, no. 6, pp. 4823–4833, Jun. 2017.
legitimate channels and thereby decreases ρL and RSKG . [11] Y. Chen, K. Huang, Y. Zhou, K. Ma, H. Jin, and X. Xu, “Physical layer
Fig. 4 investigates the impact of randomly distributed Eves key generation scheme through scrambling the correlated eavesdropping
and their distribution parameters. We can see that no matter channel,” IEEE Access, vol. 8, pp. 48982–48990, 2020.
how Eves are distributed, RSKG almost increases linearly with [12] Z. Ji et al., “Secret key generation for intelligent reflecting surface
the number of IRS elements N. Moreover, when the radius of assisted wireless communication networks,” IEEE Trans. Veh. Technol.,
vol. 70, no. 1, pp. 1030–1034, Jan. 2021.
ΦE is R = 0.1 m, a lower λE (less Eves) also improves RSKG , [13] C. D. T. Thai, J. Lee, and T. Q. S. Quek, “Physical-layer secret key gen-
which is consistent with our analysis in Section IV. However, eration with colluding untrusted relays,” IEEE Trans. Wireless Commun.,
this improvement becomes negligible when R is increased to vol. 15, no. 2, pp. 1517–1530, Feb. 2016.
1 m. This confirms the conclusion that Eves beyond several [14] C. T. Li and V. Anantharam, “One-shot variable-length secret key
wavelengths can be ignored in secret key generation. Finally, agreement approaching mutual information,” in Proc. Allerton Conf.,
Monticello, IL, USA, 2018, pp. 259–266.
simulation results given by (5) coincide well with theoretical [15] G. Chen and J. P. Coon, “Secrecy outage analysis in random wire-
results obtained from (16), which shows the correctness of our less networks with antenna selection and user ordering,” IEEE Wireless
analysis. Commun. Lett., vol. 6, no. 3, pp. 334–337, Jun. 2017.

Authorized licensed use limited to: Jaypee Insituite of Information Technology-Noida Sec 128 (L3). Downloaded on December 10,2021 at 17:50:53 UTC from IEEE Xplore. Restrictions apply.