1036 IEEE SIGNAL PROCESSING LETTERS, VOL.

28, 2021

Intelligent Reflecting Surface Assisted

Secret Key Generation
Xinjin Lu , Jing Lei , Yuxin Shi , and Wei Li

Abstract—In secret key generation of physical layer security incur some change to the incident signal, which could help the
technology, it is challenging to achieve high key capacity and low signal transmissions.
bit inconsistency rate. This paper investigates intelligent reflect- In recent years, more and more studies have applied IRSs to
ing surface (IRS)-assisted secret key generation, which aims to
maximize the secret key capacity by adjusting the placement of
physical layer security of wireless communications [10], [11].
the IRS units. Specifically, we first analyze and deduce the key The intuition of these researchers is that the IRS can be used
capacity expression of the IRS-assisted system from the perspective to improve the secrecy data rate under wiretap channel, which
of information theory. Then we investigate how to use the channel is called keyless information theory security [12]. In the key
state information (CSI) to place the IRS units effectively so as to generation of the physical layer security technology, how to
maximize the secret key capacity. Simulation results show that extract the key effectively and make full use of the channel
our scheme could improve the quality of secret key generation state information (CSI) to get more keys has always been an
significantly.
open issue. Since the IRS is able to configure the wireless
Index Terms—Intelligent reflecting surface, secret key capacity, channel in real-time via passive reflection, it has great potential
secret key generation. in improving the secret key capacity [13].
Aiming at the problem of wireless channel key generation,
I. INTRODUCTION this paper proposes a scheme based on IRS assisted secret
INCE Maurer [1] proposed that both legitimate commu- key generation. Different from the work in [13], we consider
S nication parties can extract the same key through a re-
lated random source in 1993, the key-based physical layer
a three-node model and optimize the model by the placement of
IRS units. By deriving the key capacity expression of the IRS
security mechanism has drawn significant research attention assisted system, we further optimize the placement of smart IRS
gradually [2], [3]. Due to the time-varying, short-term reci- units, or the switch state of IRS units. This scheme can maximize
procity and space-time uniqueness of wireless channels, it can the key capacity of the system when IRS units resources are lim-
be used as a natural source to generate secret keys. The sender ited. The simulation results show that the system with location
and receiver can obtain a secure shared key through channel optimization of the IRS units can not only effectively increase
probing, measurement quantization, information reconciliation key capacity, but also greatly reduce the bit inconsistency rate.
and privacy amplification. Secret key capacity is an important The main structure of this paper is as follows. Section II
parameter in the study of secret key generation. Since Ahlswede introduces the IRS assisted system model. Section III derives the
and Csiszar [4] et al. derived the theoretical upper limit of key capacity analytical formula for the system and proposes an
the secret key capacity in source-type model with wiretapper optimization scheme for the placement of IRS units. Section IV
and channel-type model with wiretapper, the research on the gives the simulation results and V concluded the full paper.
secret key capacity under more complex system models has been Notations: Throughout our discussions, the distribution of
successively developed [5], [6]. complex Gaussian random variable with mean 0 and variance
On the other hand, intelligent reflecting surface (IRS) [7], [8] σ 2 is denoted by ∼ CN (0, σ 2 ). C M ×N denotes norm, the trace
has emerged as a promising technology to improve communica- of a matrix and the space of M × N complex-valued matrices.
tion qualities through some adjustments. Generally, signals can
be controlled smartly by adjusting the reflection coefficients of
IRS such as the phase, amplitude, frequency, or even polariza- II. IRS ASSISTED SYSTEM MODEL
tion [8], [9]. In essence, IRS is composed of a large number of The idea of secret key generation is that the legitimate com-
reconfigurable and passive reflecting units whose location can munication parties extract the secure shared secret keys through
be also adaptively placed. These IRS units can independently channel probing, quantizing, information reconciliation and pri-
vacy amplification. The specific process of key generation based
Manuscript received November 25, 2020; revised January 10, 2021; accepted on CSI is shown in Fig. 1 and the steps are as follows.
February 16, 2021. Date of publication February 23, 2021; date of current version r Channel probing [14], [15]: The legitimate communication
June 4, 2021. The associate editor coordinating the review of this manuscript
and approving it for publication was Prof. Xun Cao. (Corresponding author: parties Alice and Bob successively send channel sounding
Jing Lei.) signals to each other within the coherence time, and both
The authors are with the College of Electronic Science and Technology, parties obtain channel characteristic observation values
National University of Defense Technology, Changsha 410000, China (e-mail:
based on the received signals.
[email protected]; [email protected]; [email protected];
[email protected]).
r Measurement quantization [16], [17]: Both parties of legit-
Digital Object Identifier 10.1109/LSP.2021.3061301 imate communication adopt the same quantization scheme

1070-9908 © 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 January 19,2022 at 19:12:13 UTC from IEEE Xplore. Restrictions apply.
 LU et al.: INTELLIGENT REFLECTING SURFACE ASSISTED SECRET KEY GENERATION 1037

Bob-Alice link hBA , Alice-Eve link hAE and Bob-Eve link hBE
are satisfied hΔ ∼ CN (0, σh2 Δ ), Δ ∈ (AB, BA, AE, BE). In
addition, the Alice-IRS-Bob link, the Bob-IRS-Alice link are
denoted by hAIB ∈ C N ×1 and hBIA ∈ C N ×1 , respectively. N
is the number of IRS reflecting units. It is assumed that Eve can
monitor the communication content between Alice and Bob,
but cannot actively interfere with the key establishment process.
Assuming that the distance between Eve and Alice or Bob is
greater than a half wavelength λ/2, the channel features of the
main channel and the eavesdropping channel are independent of
each other [22].
The IRS units receive all multi-path received signals and
reflect the combined signal from via IRS planar array. We denote
Ψ = [β1 Ψ1 , β2 Ψ2 , β3 Ψ3 , . . ., βN ΨN ]T as the vector associated
Fig. 1. The process of secret key generation based on wireless channel. with the effective phases shifts Ψi in all IRS units, where
βi ∈ {0, 1} indicates the switch state of the reflection units at
the current position. β = 1 means that IRS units is power on,
while β = 0 means that IRS is power off. It is assumed that the
units of IRS are separate of each other. Thus, the received signal
at Alice and Bob can be respectively written as
yA = (hBA + hBIA )x + nA
 

N
= hBA + hBIA βi Ψi x + nA
i

i=1

yB = (hAB + hAIB )x + nB
 

N
= hAB + hAIB βi Ψi x + nB
i
(1)
i=1
Fig. 2. IRS assisted system model of the secret key generation.
2 2
where nA ∼ CN (0, σB ), nB ∼ CN (0, σA ) denote the noise at
the legitimate users. We assume that hiBIA and hiAIB satisfy
to quantify the channel feature values obtained by channel hiΔ ∼ CN (0, σh2 i ), Δ ∈ (BIA, AIB).
Δ
probing to furture get the initial key.
r Information reconciliation [18], [19]: Due to factors such as
noise, interference, estimation error, half-duplex, etc., there III. SECRET KEY CAPACITY ANALYSIS AND SCHEME
may be inconsistent bits in the initial key. The two parties OPTIMIZATION OF IRS ASSISTED SYSTEM
in legitimate communication complete the verification of Secret key capacity is the upper bound of the key generation
inconsistent key bits through information exchange on rate. hA and hB represent the main channel state informa-
the common channel, and obtain consistent key bits. The tion (CSI) obtained by Alice and Bob, respectively. hE is the
interactive information can be key sequence number, parity eavesdropping channel information obtained by Eve. Therefore,
check matrix, etc. the secret key capacity can be expressed as a form of mutual
r Privacy amplification [20], [21]: During channel probing information.
and information reconciliation process, the eavesdropper
Eve may overhear some information about the key, which C = I(hA ; hB |hE ) (2)
poses a potential threat to the security of the key. Alice Based on the system model in Fig. 2, Alice and Bob send
and Bob can use universal hash functions from a publicly known probing signals to each other in turns within the channel
known set of such functions to obtain fixed size length coherence time. Alice and Bob’s estimation of CSI can be
output from longer input streams, which ensures that Eve expressed as:
can not get any information about the secret key.
The system model of IRS assisted secret key generation is hA = hBA + zA
depicted in Fig. 2, Alice and Bob are legitimate communication
= (hBA + hBIA )x + zA
nodes who aim to extract the secret key from wireless channel.
IRS is the IRS and Eve is passive eavesdropper. Alice, Bob  

N
and Eve are all equipped with a single antenna, and all adopt = hBA + hBIA βi Ψi x + zA
i

time-division duplex (TDD) working mode and half-duplex i=1

communication style, which can ensure the reciprocity between
hB = hAB + zB
the uplink and downlink channels within a coherent time. It
is assumed that channel coefficients of Alice-Bob link hAB , = (hAB + hAIB )x + zB

Authorized licensed use limited to: Jaypee Insituite of Information Technology-Noida Sec 128 (L3). Downloaded on January 19,2022 at 19:12:13 UTC from IEEE Xplore. Restrictions apply.
 1038 IEEE SIGNAL PROCESSING LETTERS, VOL. 28, 2021

 

N
= hAB + hiAIB βi Ψi x + zB (3)
i=1

where zA ∼ CN (0, σz2A ) and zB ∼ CN (0, σz2B ) are the obser-

vation noises at the nodes of Alice and Bob. It should be noted
that the noise term includes the noise of direct channel hBA or
hAB and all sub-channels hBIA or hAIB .
The channel estimation values hBA and hAB obey the fol-
lowing distribution.
hBA ∼ CN (0, σh2 )
BA

hAB ∼ CN (0, σh2 ) (4)

AB

where σh2 and σh2 can be described as:

BA AB


N
σh2 = σh2 BA + βi2 σh2 i
BA BIA
i=1


N Fig. 3. The comparison of the IRS-assisted system and the system without
σh2 = σh2 AB + βi2 σh2 i (5) IRS on secret key capacity.
AB AIB
i=1

Since the channel meets reciprocity in the coherent time, we can

We can take formula (5) into formula (10) to obtain the final
further get σh2 = σh2 .
AB BA secret key capacity:
Besides, since the distance between Eve and Alice or Bob is ⎛ 2
⎞
more than λ/2, the legitimate CSI of hAB and hBA could not σ 2
+ N
β 2 2
σ /σ 4
be included in hE . The formula (2) can be further simplified as ⎜ hAB i=1 i hAIBi z ⎟
I(hA ; hB ) = log2 ⎝1 + ⎠
2 N
1 + 2 σhAB + i=1 βi σhi2 2 /σz2
C = I(hA ; hB |hE ) AIB
(11)
= I(hA ; hB ) It can be seen from formula (11) that the secret key capacity
= Hd (hA ) + Hd (hB ) − Hd (hA , hB ) (6) of the system can be adjusted by the reflection factor βi of the
IRS units. Alice and Bob take turns to send probing signals in
where Hd (·) represents differential entropy. the channel probing process. The CSI of all sub-channels can
Ye et al. pointed out that legitimate communication parties be obtained through channel estimation, which could be used to
would establish the following observations in order to get shared further improve the probing of subsequent channel. Alice and
keys from legitimate channel [23]. Bob can adjust the smart reflector to obtain more keys by use of
x = h + na the CSI in the subsequent channel probing process.
If the number of IRS units is limited or the number of IRS
y = h + nb (7) units that can be turned on is limited, i.e. N i=1 βi  M , the
where h ∼ CN (0, σh2 ), na ∼ CN (0, σn2 a ) and nb ∼ parameters IRS βi needs to be adjusted effectively to maximize
CN (0, σn2 b ). And the secret key capacity can be expressed as the secret key capacity. This problem can be further transformed
⎛ ⎞ as:
σh2 max I(hA ; hB )
⎜ ⎟
I(x; y) = log2 ⎝1 + 2 σ2 ⎠ (8) N
σn2 a + σn2 b +
σn a nb
2
s.t. βi  M (12)
σh i=1
βi ∈ {0, 1}, i = 1, 2, . . . , N
We can take formula (4) into formula (8) to obtain the secret
key capacity of the system: In order to find the optimal placement of IRS units, we sort
⎛ ⎞ σh2 i corresponding to different hiAIB and take the first M
AIB
2
⎜ σh ⎟ positions with the largest σh2 i to place the IRS units.
I(hA ; hB ) = log2 ⎝1 + AB
σz2A σz2B ⎠
(9) AIB
2 2
σ zA + σ zB + σ 2
h
AB IV. SIMULATION RESULTS AND ANALYSIS
It is assumed that zA and zB are independent and identically We conduct Monte Carlo simulation verification on the pro-
distributed, and σz2A = σz2B = σz2 . Formula (9) can be further posed scheme. Fig. 3 shows the secret key capacity I(hA ; hB )
simplified as under different signal-to-noise ratio (SNR) when the number of
  IRS units is limited. It can be seen that the I(hA ; hB ) of the IRS
σh4 /σz4
I(hA ; hB ) = log2 1 + AB
(10) assisted system has been greatly improved compared with the
1 + 2σh2 /σz2 system without IRS. The I(hA ; hB ) of the IRS assisted system
AB

Authorized licensed use limited to: Jaypee Insituite of Information Technology-Noida Sec 128 (L3). Downloaded on January 19,2022 at 19:12:13 UTC from IEEE Xplore. Restrictions apply.
 LU et al.: INTELLIGENT REFLECTING SURFACE ASSISTED SECRET KEY GENERATION 1039

Fig. 4. Comparison of IRS-Random placement and IRS-Optimal placement Fig. 6. Comparison of IRS-Random placement and IRS-Optimal placement
on secret key capacity when R = 1/8, 1/4, 1/2. on bit inconsistency rate.

secret key capacity on the scheme of IRS-Random placement

and when N = 32, 64 128 and R = 1/4. We can see that the
I(hA ; hB ) increases with the increase of N and the I(hA ; hB )
on IRS-Optimal placement is greater. Since the selection space
of sub-channels available increases as the number of IRS units
increases, which can bring higher secret key capacity. In ad-
dition, the I(hA ; hB ) of the scheme based on the placement
optimization of the IRS units is much higher that of random
placement.
The IRS-assisted system can also obtain a lower bit incon-
sistency rate. Bit inconsistency rate Kincon /Wtotal refers to
the ratio of inconsistent bits Kincon to the total bits Wtotal
obtained by the legitimate nodes in each channel probing. Fig. 6.
compares the bit inconsistency rate performance of the differ-
ent schemes which adopt adaptive single-bit quantization with
guard interval [24]. It can be seen that the bit inconsistency
rate performance of IRS-Optimal placement is better than the
IRS-Random placement, which means that Alice and Bob can
Fig. 5. Comparison of IRS-Random placement and IRS-Optimal placement
on secret key capacity when N = 32, 64, 128. obtain the initial secret keys with a smaller bit inconsistency
rate. This makes information negotiation easier to implement in
the key generation process.
increases as the SNR increases. In addition, the I(hA ; hB ) on the
scheme of IRS-Optimal placement is higher than that of IRS-
V. CONCLUSION
Random placement. The main reason is that the sub-channels
path with better CSI are selected in the scheme of IRS-Optimal Aiming at the difficulty of secret key generation in the wireless
placement. channel under single path, this paper proposes an IRS-assisted
Fig. 4 compares the I(hA ; hB ) on the scheme of IRS-Random secret key generation scheme, which solves the key genera-
placement and when R = 1/8, 1/4, 1/2 and N = 128. R = tion problem by using IRS units to increase the number of
M/N means that M IRS units are selected from the N available sub-channels. We give the derivation of the secret key capacity
IRS units for practical use. It can be seen that the I(hA ; hB ) formula of the IRS assisted system, and further enhance the
increases with the increase of R. In addition, the I(hA ; hB ) secret key capacity of the system by optimizing the placement
on the scheme of IRS-Optimal placement is better than the of the IRS units. Simulation results shows that this scheme can
scheme of IRS-Random placement and this advantage increases not only save IRS resources, but also effectively improve the
with the decrease of R. Fig. 5 shows the comparison of the performance of secret key generation.

Authorized licensed use limited to: Jaypee Insituite of Information Technology-Noida Sec 128 (L3). Downloaded on January 19,2022 at 19:12:13 UTC from IEEE Xplore. Restrictions apply.
 1040 IEEE SIGNAL PROCESSING LETTERS, VOL. 28, 2021

REFERENCES [13] Z. Ji et al., “Secret key generation for intelligent reflecting surface assisted
wireless communication networks,” 2020, arXiv:2008.06304.
[1] U. M. Maurer, “Secret key agreement by public discussion from common [14] Y. Wei, K. Zeng, and P. Mohapatra, “Adaptive wireless channel probing
information,” IEEE Trans. Inf. Theory, vol. 39, no. 3, 733–742, May 1993. for shared key generation based on pid controller,” IEEE Trans. Mobile
[2] G. Bassi, P. Piantanida, and S. Shamai Shitz, “The wiretap channel with Comput., vol. 12, no. 9, pp. 1842–1852, Sep. 2013.
generalized feedback: Secure communication and key generation,” IEEE [15] Y. Peng, P. Wang, W. Xiang, and Y. Li, “Secret key generation based on
Trans. Inf. Theory, vol. 65, no. 4, pp. 2213–2233, Apr. 2019. estimated channel state information for TDD-OFDM systems over fading
[3] H. Hentil, V. Koivunen, and H. V. Poor, “Key generation for secure dis- channels,” IEEE Trans. Wireless Commun., vol. 16, no. 8, pp. 5176–5186,
tributed detection in iot using polar quantization,” in Proc. 53rd Asilomar Aug. 2017.
Conf. Signals, Syst., Comput., 2020, pp. 1916–1920. [16] N. Patwari, J. Croft, S. Jana, and S. K. Kasera, “High-rate uncorrelated bit
[4] R. Ahlswede and I. Csiszar, “Common randomness in information theory extraction for shared secret key generation from channel measurements,”
and cryptography. I. secret sharing,” IEEE Trans. Inf. Theory, vol. 39, no. 4, IEEE Trans. Mobile Comput., vol. 9, no. 1, pp. 17–30, Jan. 2010.
pp. 1121–1132, Jul. 1993. [17] C. Chen and M. A. Jensen, “Secret key establishment using temporally
[5] G. Bassi, P. Piantanida, and S. Shamai Shitz, “The secret key capacity of and spatially correlated wireless channel coefficients,” IEEE Trans. Mobile
a class of noisy channels with correlated sources,” Entropy, vol. 21, no. 8, Comput., vol. 10, no. 2, pp. 205–215, Feb. 2011.
p. 732, 2019. [18] D. Chen, Z. Qin, X. Mao, and P. Yang, “Smokegrenade: An efficient key
[6] G. Wang, C. Ottaviani, H. Guo, and S. Pirandola, “Improving the lower generation protocol with artificial interference,” IEEE Trans. Inf. Forensics
bound to the secret-key capacity of the thermal amplifier channel,” Eur. Secur., vol. 8, no. 11, pp. 1731–1745, Nov. 2013.
Phys. J. D, vol. 73, no. 1, 2019, Art. no. 17. [19] Y. Liu, S. C. Draper, and A. M. Sayeed, “Exploiting channel diversity in
[7] Q. Wu and R. Zhang, “Intelligent reflecting surface enhanced wireless secret key generation from multipath fading randomness,” IEEE Trans.
network via joint active and passive beamforming,” IEEE Trans. Wireless Inf. Forensics Secur., vol. 7, no. 5, pp. 1484–1497, Oct. 2012.
Commun., vol. 18, no. 11, pp. 5394–5409, Nov. 2019. [20] S. Wang and C. Li, “Discrete double-bit hashing,” IEEE Trans. Big Data,
[8] Q. Wu and R. Zhang, “Beamforming optimization for wireless network early access, Oct. 10, 2019, doi: 10.1109/TBDATA.2019.2946616.
aided by intelligent reflecting surface with discrete phase shifts,” IEEE [21] R. Tu et al., “Deep cross-modal hashing with hashing functions and unified
Trans. Commun., vol. 68, no. 3, pp. 1838–1851, Mar. 2020. hash codes jointly learning,” IEEE Trans. Knowl. Data Eng., early access,
[9] E. Basar, M. Di Renzo, J. De Rosny, M. Debbah, M. Alouini, and R. Zhang, Apr. 16, 2020, doi: 10.1109/TKDE.2020.2987312.
“Wireless communications through reconfigurable intelligent surfaces,” [22] W. Jakes, Microwave Mobile Communications. Piscataway, NJ, USA:
IEEE Access, vol. 7, pp. 116753–116773, 2019. IEEE Press, 1993.
[10] X. Yu, D. Xu, and R. Schober, “Enabling secure wireless communications [23] C. Ye, A. Reznik, and Y. Shah, “Extracting secrecy from jointly gaus-
via intelligent reflecting surfaces,” in Proc. IEEE Global. Commun. Conf., sian random variables,” in Proc. IEEE Int. Symp. Inf. Theory, 2006,
2020, pp. 1–6. pp. 2593–2597.
[11] H. Shen, W. Xu, S. Gong, Z. He, and C. Zhao, “Secrecy rate maximization [24] S. Jana, S. Premnath, M. Clark, S. Kasera, and N. Patwari, “On the
for intelligent reflecting surface assisted multi-antenna communications,” effectiveness of secret key extraction from wireless signal strength in real
IEEE Commun. Lett., vol. 23, no. 9, pp. 1488–1492, Sep. 2019. environments,” Proc. 15th Annu. Int. Conf. Mobile Comput. Netw., vol. 9
[12] A. D. Wyner, “The wire-tap channel,” Bell Syst. Tech. J., vol. 54, no. 8, 2009, pp. 321–332.
pp. 1355–1387, Oct. 1975.

Authorized licensed use limited to: Jaypee Insituite of Information Technology-Noida Sec 128 (L3). Downloaded on January 19,2022 at 19:12:13 UTC from IEEE Xplore. Restrictions apply.