Relay Selection for Improving Physical Layer Security in D2D Underlay Communications

This paper investigates the physical layer security of inband underlay Device-to-Device (D2D) communication, where the direct link between D2D users is not available. In this respect, optimal relay and suboptimal relay selections are utilized to secure the D2D transmission. The eavesdropper uses either maximal-ratio combining or selection combining to increase the wiretapped signals. The D2D secrecy performance analysis is performed regarding the secrecy outage probability (SOP) and the probability of non-zero secrecy capacity, and closed-form expressions are provided for both relay selection approaches and verified using Monte-Carlo simulation. From the numerical results, it is shown that increasing the number of D2D relays enhances the secrecy performance of D2D communications. Moreover, the exact asymptotic analysis of the SOP is provided. It turns out that the diversity order of both relay selection approaches is the same.

ciency [4]. The benefits of D2D communications are high 23 spectral efficiency, short delay, and low power consump-24 tion [5]. New mobile services for various proximity-based 25 applications are offered via D2D communications such as 26 content sharing, social networking, and multiplayer gaming. 27 Based on the spectrum band exploitation, two approaches 28 The associate editor coordinating the review of this manuscript and approving it for publication was Zhenzhou Tang . of D2D communications are presented; inband and out-band 29 D2D communications. The D2D users are allowed to share 30 the spectrum band with the cellular users in the former, 31 whereas the unlicensed spectrum band is utilized by the D2D 32 users in the latter [6]. The inband D2D communication can be 33 further subdivided into two categories: overlay and underlay. 34 Specifically, the spectrum band of the cellular network is 35 divided into non-overlapping frequency sets in overlay D2D 36 communications, where the cellular users utilize one set while 37 the D2D users use the other one. Hence, in overlay D2D com-38 munications, interference management between cellular and 39 D2D users is not required. In underlay D2D communications, 40 however, the same frequency is utilized by the cellular and 41 D2D users; therefore, interference management is crucial [7]. 42 A. RELATED WORK 43 Nowadays, wireless networks have been widely utilized in 44 daily life applications to transmit essential and secure infor-45 mation. Consequently, security is considered a critical issue 46 jammer was investigated in [21], [22], and [23]. 97 The importance of PLS techniques in relay networks lies 98 in the fact that an intermediate relay node is more vulnerable 99 to wiretapping than any other terminal. Signal relaying has 100 also been considerably used to increase the quality of ser-101 vice in cellular networks. Relaying techniques provide many 102 advantages, including extended coverage and a higher 103 data rate. The benefits of cooperative relaying systems in 104 the context of PLS have been widely investigated [24], 105 [25], [26], [27], [28], [29]. In [24], the PLS was inves-106 tigated for the cooperative non-orthogonal multiple access 107 (NOMA) system, where power allocation approaches were 108 proposed for the legitimate and jamming signals. The authors 109 in [30] proposed two relay selection techniques, namely 110 energy-aware relay selection and non-energy-aware relay 111 selection, to reduce the overhearing by multiple eavesdrop-112 pers in cognitive radio transmission. The trade-off of secu-113 rity vs reliability was studied for wireless communications 114 in [25], where an opportunistic relay selection approach was 115 introduced to increase the secrecy capacity of the cellular 116 network. In this regard, the relay selection of a cooperative 117 scenario was analyzed in [26] to guarantee a secure trans-118 mission for the cellular network. Towards this end, the relay 119 selection approach was proposed for guarding wireless trans-120 missions against eavesdropping attacks. The authors in [27] 121 studied the PLS of energy harvesting for cognitive radio 122 networks using the cooperative relaying technique. In [28], 123 the PLS of cooperative dual-hop NOMA for internet-of-thing 124 networks was investigated. The PLS of underlay multihop 125 D2D relaying was investigated in [29]. 126 The benefits of using D2D relays over direct D2D com-127 munications have been investigated in [31], [32], and [33] 128 especially when there is a long distance between D2D users 129 or poor link quality. As a result of sharing the same spec-130 trum band, interference is a serious issue for both D2D and 131 cellular users. Thus, interference management is necessary to 132 mitigate the effects of interference [7]. However, the interfer-133 ence can be utilized to increase the secrecy level for cellular 134 and D2D communications by confounding the eavesdropped 135 signal [34], [35]. To maximize the wiretapped signal quality, 136 the eavesdropper can utilize either maximum-ratio combining 137 (MRC) or selection combining (SC) [36]. For the PLS in 138 underlay D2D communications, many approaches can be 139 used to increase the security level of D2D communications. 140 More specifically, artificial noise and guard zone were used 141 to guarantee a secure transmission for the D2D communica-142 tions [37]. To secure D2D transmission, the BS was used as 143 a friendly jammer to confuse the wiretapper by producing 144 artificial noise [38]. Nevertheless, the quality of service of 145 cellular users should be taken into consideration. In order to 146 improve the PLS, spectrum partition and mode selection for 147 D2D inband communications were investigated in [39]. The 148 D2D relay guarantees a high-security cellular transmission 149 by sending jamming signals toward an eavesdropper to com-150 pensate for sharing the spectrum. As a result, the benefits for 151 cellular and D2D users are achieved, i.e., security enhance-152 ment for the former and high reliability and robustness for 153 the latter [40].

155
In this work, the PLS of the D2D communication is inves-156 tigated. Unlike the existing work on the PLS of the cellular 157 firm the provided analytical expressions.

207
The remaining part of the paper is organized as follows.

208
In Section II, the system model is described. The outage 209 probability of the cellular network is analyzed in Section III. communications. The main results are discussed in 214 Section VI. Finally, Section VII concludes the paper by 215 outlining the significant contributions and results of this 216 work. 218 We consider a downlink transmission scenario in D2D relay 219 communication as illustrated in Fig. 1, where the cellular 220 network shares its spectral band with the D2D network in 221 a particular environment. The D2D network consists of a 222 D2D transmitter, T , N R decode-and-forward (DF) D2D relays 223 (R k | k = 1, . . . ., N R ), a D2D receiver, D, each equipped 224 with a single antenna, and a multi-antenna eavesdropper, E, 225 equipped with N E antennas. We consider a cellular network 226 consisting of a BS, equipped with multiple antennas N B , com-227 municating with a single antenna cellular user, C. It is worth 228 mentioning that, as a result of severe shadowing, the direct 229 link between the D2D pair is unavailable in the proposed 230 scenario. Thus, communication from T to D can only be 231 set through the relays. Thus, the D2D transmissions require 232 two phases. The received signals from T are fully decoded 233 in the first phase by the D2D relays. In the second phase, 234 based on the highest secrecy capacity, the best D2D relay is 235 selected from the set of the D2D relays to forward the decoded 236 signal to D. During this phase, E may wiretap the transmis-237 sion of R k . The BS transmits in both phases. Furthermore, 238 we assume that all communication channels experience flat 239 fading with a Rayleigh distribution.

240
The channel coefficients for the T → R k , T → C, BS → 241 C, BS → E, R k → E, BS → R k , BS → D, and R k → D 242 links are denoted as h tr k , h tc , h bc , h be , h r k e , h br k , h bd , 243 and h r k d , respectively. In addition, the channel power gains 244 are indicated by |h ab | 2 , which are independent and expo-245 nentially distributed random variables with a mean of λ ab , 246 where ab ∈ {tr k , tc, bc, be, r k e, br k , bd, r k d}. Additionally, 247 the Euclidean distance is denoted as d ab . Also, the variances 248 of the additive white Gaussian noise (AWGN) at R k , D, 249 C, and E are denoted by σ 2 r k , σ 2 d , σ 2 c , and σ 2 e , respectively. 250 VOLUME 10, 2022 We assume that the D2D nodes, T and R k , transmit with equal 251 power P. In the first phase, the received signal at R k is given

264
(2) 265 where n d is the AWGN at R k . The received signal at C, in the 266 second phase, can be expressed as where n c is the AWGN at C. interference-and-noise ratio (SINR) at R k , in the first phase, 283 is given by . The channel capacity of the 288 T → R k link is given by 290 The SINR at D, in the second phase, is given by The SINR at C, γ C , in the second phase, is given by where h b i c represents the channel coefficients between the 299 selected antenna at the BS and C.
Moreover, the SINR at E is given by

311
The cellular outage probability, P out , is defined by where γ BC is the instantaneous SNR at C and 314 ϑ c = 2 R c − 1, R c being the data transmission rate. Antenna 315 selection approach is employed at the BS to avoid the high 316 hardware complexity while maintaining the diversity and reli-317 ability advantages from multiple antennas. More specifically, 318 the importance of using the antenna selection approach lies 319 in the fact that the power consumption and the complexity of 320 signal processing overhead are low as compared with other 321 techniques such as beamforming techniques. In this regards, 322 the transmitting antenna at BS is selected for the best data 323 transmission performance. Moreover, the maximum channel 324 gain can be determined by using The probability density function (PDF) of |h bc | 2 is given 327 by [41] 328 To derive the PDF of γ BC , we use [42] 331 The PDF of γ bc can be expressed in terms of the binomial 336 and f γ rc (.) is given by By plugging (14) and (15) into (13), the PDF of γ BC , after 339 some algebraic manipulations, can be obtained as By substituting (17) in (10), P out can be obtained.

347
To examine the D2D secrecy capacity, C S , the SOP is inves-348 tigated in this subsection. The SOP can be defined as the 349 probability that the achievable secrecy rate is less than a 350 predefined target secrecy rate, R s , for the D2D transmission.

351
Mathematically speaking, it can be expressed as [44] 352 where the index θ signifies the selected relay, R θ , for the ORS, 363 C R k D and C R k E are the channel capacities of the R k → D 364 and R k → E links, respectively, and [x] + = max(x, 0). The 365 channel capacities of the R k → D and R k → E links are 366 given by Mathematically speaking, the secrecy capacity, C ORS S , can be 371 expressed as Using (22), the SOP can be expressed as where υ ∈ {MRC, SC } and ϑ s = 2 2R s . The probability 379 density function (PDF) of the R θ → D link, F γ R θ D (·), can 380 be obtained as In this approach, the received signals are coherently com-384 bined. Hence, the channel gain between the selected relay and 385 E can be obtained as Using (24), the PDF of the R δ → D link, F γ SRS (.), can be 425 obtained as In this subsection, the SOP SRS MRC will be derived for the SRS 430 scheme. The SOP SRS MRC can be obtained by

446
Proof: See Appendix D.
Now, the SOP of the D2D communications at high SNR, i,e., 449 when γ d → ∞, is investigated to gain more insight into 450 the impact of the essential parameters of the system model. 451 We consider a scenario where D is much closer to the D2D 452 relay than the eavesdropper, which can be mathematically 453 described as ω 1 >> ω 3 . As ω 1 → ∞, the asymptotic 454 expression of the SOP can be expressed by [48] 455  (ω 2 + 1) 474

2) EAVESDROPPER CHANNEL WITH SC IN THE ORS SCHEME 475
Similarly, the asymptotic SOP for SC, SOP ∞ SC, ORS , can be 476 derived following a similar procedure to that given above as 477 β ω 2 ω 4 (k + 1) where G SRS d SC = N R and G SRS a SC can be derived as in (45) Following the same steps in the derivation of (26) and (28),

511
D v k can be derived as in (47) and (48), as shown at the bottom 512 of the next page, for MRC and SC, respectively, where A 4 = Following the same steps in the derivation of (34) and (36), can be derived as in (50) and (51), as shown at 522 the bottom of the next page, for MRC and SC, respectively, and Z 6 = ω 1 ω 2 (k+1) .

526
In this section, the analysis of the SOP and the PNSC for both 527 combining techniques, MRC and SC, are presented and val-528 idated through Monte-Carlo simulations. From the obtained 529 results, the impact of D2D relays and the number of eaves-530 dropper's antennas for both MRC and SC are investigated. 531 For simplicity, we assume ω 3 = 10 dB, and the noise variances 532 of all nodes are normalized to unity.

533
The SOP for MRC, SOP MRC , of the D2D communications, 534 is illustrated in Fig. 2 versus ω 1 for different values of N R for 535 both relay selection schemes, namely, ORS and SRS. The R s 536 is 1 b/s/Hz, and N E = 3. It can be observed that the SOP MRC 537 decreases as N R increases, showing the influence of cooper-538 ative communications on enhancing the D2D secrecy perfor-539 mance. It is worth mentioning that the SOP MRC increases as 540 ω 1 decreases. At SOP MRC = 10 −4 , a security enhancement of 541 6 dB and 9 dB of the ORS scheme over the SRS one at N R = 3 542 and N R = 5, respectively, can be noticed. In Fig. 2, SOP MRC 543 decreases from 0.025 to 0.0002 at 30 dB and N R =3 when the 544 OSR scheme is used as compared to the SRS scheme.

545
The results in Fig. 2 show that both schemes are close 546 to each other at low SNR. However, the gap increases 547 as the SNR increases. It can be inferred that the ORS guar-548 antees the optimal secrecy performance for the D2D link. 549 However, the CSI of the wiretapped link should be available 550 to utilize the ORS scheme, which is not always the case in 551 a practical scenario. Additionally, the asymptotic SOP are 552 In Fig. 3, the SOP SC of the D2D communications, is illus-561 trated for the ORS and SRS schemes. It can be seen that the 562 SOP SC increases with decreasing N R , implying an improve-563 ment in the D2D secrecy performance. From Figs. 2 and 3, 564 one can observe that the high-security level for D2D commu-565 nications is obtained when E employs SC in comparison to 566 the MRC approach. This is because the MRC gives a higher 567 SNR gain at E as compared to the SC approach.
VOLUME 10, 2022      6 plots the analytical and simulation results for the cel-585 lular outage probability, P out , versus µ 1 for different values 586 of N B at the BS. It can be observed that P out of the cellular link 587 decreases monotonically as µ 1 increases. Interestingly, the 588 reliability enhances significantly with increasing N B . Thus, 589 the data transmission of cellular communication improves 590 due to employing multiple antennas as the BS as compared to 591 a single antenna. Furthermore, it can be seen that analytical 592 results are also found to match the simulated ones, validating 593 the correctness of our analysis. Fig. 7    The expression P υ k can be obtained by

634
where β = 2 R s and α = β − 1. The PDF of γ R θ E for MRC 635 can be derived using [42] as is given by and f γ be (·) is given by By plugging (54) and (55)

669
In what follows, the integral in (60), I, is evaluated. First,   An earlier version of this paper (optimal relay selection 697 scheme analysis) was presented in part at the IEEE Vehicular 698 Technology Conference (VTC2020-Fall) [1]. This work is a 699 part of a Ph.D. thesis [2].