0% found this document useful (0 votes)

9 views

Secrecy Performance of Multi-User MISO VLC Broadcast Channels With Confidential Messages

Uploaded by

Tunha

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

9 views

Secrecy Performance of Multi-User MISO VLC Broadcast Channels With Confidential Messages

Uploaded by

Tunha

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 25

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/327897505

Secrecy Performance of Multi-user MISO VLC Broadcast Channels with

Conﬁdential Messages

Article in IEEE Transactions on Wireless Communications · September 2018

DOI: 10.1109/TWC.2018.2871055

CITATIONS READS
36 187

3 authors:

Mohamed Amine Arfaoui Ali Ghrayeb

Concordia University Montreal Texas A&M University at Qatar
54 PUBLICATIONS 769 CITATIONS 281 PUBLICATIONS 7,961 CITATIONS

SEE PROFILE SEE PROFILE

Chadi Assi
Concordia University Montreal
458 PUBLICATIONS 10,309 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

New Paradigms for Robust Cloud Networking Services View project

Scheduling service function chains View project

All content following this page was uploaded by Mohamed Amine Arfaoui on 03 October 2018.

The user has requested enhancement of the downloaded file.

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS 1

Secrecy Performance of Multi-User MISO VLC

Broadcast Channels With Confidential Messages
Mohamed Amine Arfaoui , Ali Ghrayeb , and Chadi M. Assi

1 Abstract— We study, in this paper, the secrecy performance systems was studied in terms of transmit data rates and signal- 38
2 of a multi-user (MU) multiple-input single-output visible light to-interference-plus-noise (SINR) ratio. In [5]–[7], a thorough 39
3 communication broadcast channel with confidential messages. review of the advantages of VLC was given, whereas, 40
4 The underlying system model comprises K +1 nodes: a transmit-
5 ter (Alice) equipped with N fixtures of LEDs and K spatially in [8], the potential of VLC for indoor communications was 41

of
6 dispersed users, each equipped with a single photo-diode. The discussed. In [9] and [10], the authors studied the fundamental 42

7 MU channel is modeled as deterministic and real-valued and limits of optical wireless channels. In [11], the viability of 43
8 assumed to be perfectly known to Alice, since all users are VLC for 5G wireless networks was investigated. 44
9 assumed to be active. We consider typical secrecy performance Security issues arise naturally in VLC broadcast channels 45
10 measures, namely, the max–min fairness, the harmonic mean,
11 the proportional fairness, and the weighted fairness. For each due to its open nature. Each receiver is able to receive signals 46

12 performance measure, we derive an achievable secrecy rate for that contain all information flows from the transmitter. Hence, 47

13 the system as a function of the precoding matrix. As such, some receivers may decode data that are not intended for them. 48

ro
14 we propose algorithms that yield the best precoding matrix for However, information flows should be kept confidential from 49
15 the derived secrecy rates, where we analyze their convergence non-intended receivers [12]. This requires that the transmitter 50
16 and computational complexity. In contrast, what has been con-
17 sidered in the literature so far is zero-forcing (ZF) precoding, employs security techniques to guarantee such confidentiality 51

18 which is suboptimal. We present several numerical examples requirements. On the other hand, physical layer security (PLS) 52

19 through which we demonstrate the substantial improvements in has achieved great success in enhancing the security of wire- 53
20 the secrecy performance achieved by the proposed techniques less communications or complementing existing cryptographic 54
21 compared with those achieved by the conventional ZF. However, schemes for radio-frequency (RF) broadcast channels [13]. 55
22 this comes at a slight increase in the complexity of the proposed
Due to this, there have been recently many attempts to extend
EP
56
23 techniques compared with ZF.
the previous studies to VLC. The potential of PLS stems 57
24 Index Terms— Broadcast, MISO, secrecy performance, VLC. from its ability of leveraging features of the surrounding 58

environments via sophisticated encoding techniques at the 59

25 I. I NTRODUCTION
physical layer [14]. Indeed, PLS schemes can be applied in 60
26 A. Motivation the same spirit to VLC systems. Furthermore, VLC systems 61

29
V ISIBLE light communication (VLC) is a new
communication technology that uses visible light
as a transmission medium, i.e., the light emitted by light
are characterized by many specificities that imply major differ-
ences compared to RF systems. Precisely, VLC channels are
quasi-static and real valued channels which seemingly simplify
62

30 sources is used for illumination and data communication the application of PLS techniques. However, due to the limited 65

31 purposes simultaneously. VLC has gained significant interest dynamic range of the emitting LEDs, VLC systems impose a 66
IEE

32 during the last decade, owing to its high speed and low peak-power constraint, i.e., an amplitude constraint, on the 67

33 deployment cost [1], robustness against interference and channel input which makes unbounded inputs, like Gaussian 68

34 abundance in the available spectrum [2]. Various aspects inputs, not admissible. As a result, the performance and the 69

35 of VLC systems have been studied in the literature. In [3], optimization of PLS schemes must be revisited in the VLC 70

36 the authors proposed a VLC end to end architecture with context due to its different operating constraints. 71

37 suitable modulation schemes. In [4], the performance of VLC

Manuscript received March 18, 2018; revised August 27, 2018; accepted B. Related Work 72
September 5, 2018. This work was supported in part by the Qatar National
Research Fund through NPRP under Grant NPRP8-052-2-029, in part by Information theoretic approaches in PLS were developed 73
FQRNT, and in part by Concordia University. The associate editor coordi- to achieve secure communication [15], [16]. Based on that, 74
nating the review of this paper and approving it for publication was N. Yang.
(Corresponding author: Ali Ghrayeb.) many techniques were then introduced in PLS to enhance the 75

M. A. Arfaoui and C. M. Assi are with the Concordia Institute for secrecy performance of VLC broadcast channels. The secrecy 76
Information Systems Engineering, Concordia University, Montreal, QC H3G performance of single-user (SU) MISO VLC wiretap systems 77
1M8, Canada (e-mail: m_arfaou@encs; [email protected]).
A. Ghrayeb is with the Electrical and Computer Engineering was investigated in [17]–[29]. Under perfect eavesdropper’s 78

Department, Texas A&M University at Qatar, Doha, Qatar (e-mail: channel state information (CSI), both beamforming and 79
[email protected]). zero-forcing beamforming were adopted [17], [18], whereas 80
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. artificial noise was considered in [19]–[21]. Under imperfect 81

Digital Object Identifier 10.1109/TWC.2018.2871055 eavesdropper’s CSI, robust beamfomring and artificial noise 82

1536-1276 © 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
2 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS

83 were employed in [22] and [23], respectively. The randomly sages to K spatially dispersed users, each equipped with 139

84 located terminals scenario with a single fixed user and a single PD. The transmitted messages are assumed to be 140

85 multiple randomly located eavesdroppers was investigated confidential such that each user is supposed to receive and 141

86 in [24]–[29], where the average secrecy performance was decode only his own message, i.e., users are ignorant about 142

87 analyzed using stochastic geometry. messages not intended for them. The channel input is subject 143

88 The secrecy capacity of a Gaussian wiretap channel under to an amplitude constraint. We develop within this framework 144

89 an amplitude constraint has not been determined in closed linear precoding schemes that enhance typical secrecy per- 145

90 form yet. In fact, from an information theoretic point-of- formance measures, namely: the max-min fairness (MMF), 146

91 view, finding the signaling schemes that achieve the secrecy the harmonic mean (HM), the proportional fairness (PF) and 147

92 capacity of a Gaussian wiretap channel under an amplitude the weighted fairness (WF). Specifically, we derive first an 148

93 constraint is quite challenging and it is still an open problem. achievable secrecy rate of a single user existing within the 149

94 This is attributed to the fact that, when input distributions of MU VLC network. Second, we formulate the problems of 150

95 unbounded support are not permissible, the optimal input dis- linear precoding schemes that maximize the aforementioned 151

of
96 tribution is either unknown, or only known to be discrete [30] secrecy performance measures of the considered system. Then, 152

97 for the special case of a degraded Gaussian single-input single- we propose iterative algorithms to find the best linear pre- 153

98 output (SISO) wiretap channel. VLC falls in this category coding schemes for all four ecrecy performance measures. 154

99 since amplitude constraints must be satisfied. Upper and lower We also analyze the computational complexity of the proposed 155

100 bounds on the capacity of the free-space optical intensity scheme and compare it to that of ZF where we show that 156

101 channel under peak and average optical power constraint were the increase in complexity due to the proposed schemes 157

102 derived in [31]. In addition prior works focused only on the can be, in the worst case scenario, the square root of the 158

103

104

105

106

107

108

109
ro
uniform distribution [17], the truncated Gaussian [19] and the
truncated generalized normal (TGN) [20], [32], where it was
shown that TGN is the best choice of input signaling till now,
since it encompasses several bounded input distributions and
one can enhance the secrecy performance of the system by
optimizing over its parameters.
The secrecy performance of MU-MISO broadcast chan-
nels has been studied in the literature [33]–[43]. However,
number of active users in the network. Finally, we compare the
performance of the proposed schemes to that of the con-
ventional ZF precoding [44]–[47] and we demonstrate that
substantial improvements can be achieved by the proposed
schemes.

D. Outline and Notations

159

160

161

162

163

164
EP
110

111 adoption of techniques developed for RF channels for VLC The rest of the paper is organized as follows. Section II 165
112 channels may not be straightforward since RF signals are presents the system model. Section III present the proposed 166
113 complex-valued, which is fundamentally different from the precoding schemes. Sections IV and V present the numerical 167
114 real and bounded VLC signals. Nevertheless, several studies results and the conclusion, respectively. 168
115 on precoding designs for MU-MISO VLC broadcast channels The following notations are adopted throughout the paper. 169
116 were proposed in [44]–[48]. In [44] and [45], the system Upper case bold characters denote matrices and lower case 170
117 was considered without an external eavesdropper, whereas bold characters denote column vectors. We use log(·), without 171
118 in [46] and [47], it was assumed that an external eavesdropper a base, to denote natural logarithms and information rates are 172
119 may exist within the same area. For both cases, the secrecy specified in (nats/s/Hz). All the mathematical operators and 173
120 performance of the systems was investigated, where only the parameters used in this paper are defined in Table I. 174
IEE

121 max-min fairness and the weighted fairness were used as

122 secrecy performance measures. Moreover, zero-forcing (ZF)
123 precoding was employed, to cancel the information leakage II. S YSTEM M ODEL 175

124 between users, in conjunction with uniform input signaling.

A. The VLC Channel Model 176
125 However, although it is a simple precoding scheme, ZF is
126 suboptimal in the sense that the secrecy performance of the We consider a DC-biased intensity-modulation direct- 177

127 system can be enhanced by searching for optimal precoding detection (IM-DD) scheme where the transmit element is 178

128 schemes. In [48], the same problem was considered for the an illumination LED driven by a fixed bias IDC ∈ R+ . 179

129 two-user MISO broadcast channel with confidential messages The DC-offset sets the average radiated optical power and, 180

130 under per-antenna amplitude constraint, per-antenna power consequently, settles the illumination level. The data signal 181

131 constraint and average power constraint were considered. s ∈ R is a zero-mean current signal superimposed on IDC 182

132 However, assuming only two active users in a VLC system to modulate the instantaneous optical power emitted from the 183

133 is not a realistic scenario especially for large geometric areas LED. In order to maintain linear current-light conversion and 184

134 or dense networks. avoid clipping distortion, the total current IDC + s must be 185

constrained within some range IDC ± νIDC where ν ∈ [0, 1] 186

is the modulation index [17]. Consequently, s must satisfy 187

135 C. Contributions an amplitude constraint expressed as |s| νIDC . After that, 188

136 In this paper, we consider a MU-MISO VLC broadcast the total current IDC + s is converted into an optical power 189

137 channel consisting of a transmitter (Alice), equipped with and transmitted by the LED, in which the conversion factor is 190

138 N fixtures of LEDs, aiming to transmit K confidential mes- denoted by η. 191

ARFAOUI et al.: SECRECY PERFORMANCE OF MU MISO VLC BROADCAST CHANNELS 3

TABLE I
TABLE OF N OTATIONS

of
Fig. 1. VLC path gain description.

Armed with the above description, the received signal is 216

expressed as 217

y = hs + n, (4) 218

ro where y represents the received signal, s represents the zero-

mean transmitted signal subject to the amplitude constraint
|s| ≤ A, where A = νIDC , n represents AWGN N (0, σ 2 )
distributed and h = ηRT g ∈ R+ represents the channel gain,
in which g denotes the path gain of the optical link. Assuming
that the considered LED has a Lambertian emission pattern,
the path gain is given as [51], [52]
219

220

221

222

223

224
EP
225
⎧
⎨ 1 ARX
(m+1) cosm (θ) 2 cos(ψ)R |ψ| ≤ ψF oV
192 At the receiver side, the receiver’s PD, with a responsiv- g = 2π d (5) 226

ity Rp , converts the incident optical power into a propor- ⎩0 |ψ| > ψ ,
193 F oV
194 tional current. Finally the DC-offset IDC is removed and a
− log(2)
195 transimpedance amplifier, with gain T , is used to produce a where m = log(cos(φ 1 )) is the order of the Lambertian 227

196 voltage signal y ∈ R, which is a scaled, but noisy, version of 2

emission with half irradiance at semi-angle φ 12 (measured 228
197 the transmitted signal s. The noise process is well-modeled from the optical axis of the LED). As shown in Fig 1, θ 229
198 in VLC channels as signal-independent, zero-mean, additive represents the angle of irradiance, d is the line-of-sight (LoS) 230
199 white Gaussian noise (AWGN) with variance σn2 , given by distance between the LED and the PD, ψ is the angle of 231

incidence, ψF oV is the receiver field of view (FoV) and

σn2 = σsh
2 2
IEE

+ σth
232
200 , (1) n2
ARX = sin2 (ψcF oV ) AP D is the receiver collection area, such 233
2 2
201 where σsh and σth are the variances of the shot noise and that nc is the refractive index of the optical concentrator and 234
202 thermal noise, respectively, [49]. The shot noise in an optical AP D is the PD area. 235
203 wireless channel is generated by the high rate physical photo- In most practical cases, the VLC channel is either constant 236
204 electronic conversion process (e.g., indoors VLC with no mobility) or varies very slowly 237

205
2
σsh = 2qB (Pr + Ibg I2 ), (2) compared to the transmission rate (mobility or outdoors VLC). 238

The channel coherence time is typically 0.1 to 10 ms whereas 239

206 where q is the electronic charge, B is the system bandwidth, the transmission rates are on the order of several tens of 240

207 Pr = Rp hIDC is the average received power, Ibg is the Mpbs to several Gbps. Thus, the channel remains constant 241

208 ambient current in the PD and I2 is the noise bandwidth factor. over thousands up to millions of consecutive bits, and hence, 242

209 The thermal noise is generated within the transimpedance it is considered quasi-static in the scale of interest [53]. 243

210 receiver circuitry and its variance is given by Various VLC channel estimation methods were introduced 244
in the literature, especially the receiver’s location and the 245
2 2 I2
211 σth = 8π K̄TK ARX B c + 2πΓF T ARX I3 Bc , (3) channel parameters in downlink VLC, as described in (5). For 246
G
the estimation of the receiver’s location, [54] and references 247

212 where K̄ is Boltzmann’s constant, TK is the absolute temper- therein proposed receiver positioning algorithms, whereas for 248

213 ature, c is the fixed capacitance of the PD per unit area, G is the channel estimation, [55] and [56] proposed estima- 249

214 the open-loop voltage gain, ΓF is the transimpedance channel tion methods using neural networks and statistical Bayesian 250

215 noise factor and I3 = 0.0868 [50]. MMSE, respectively. 251

4 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS

252 B. The MU-MISO VLC Broadcast Channel III. P ROPOSED P RECODING S CHEMES 298

253 As mentioned above, Alice is equipped with N fixtures

A. Single User Achievable Secrecy Rate
of LEDs and intends to transmit K (K ≤ N ) confidential
299
254

255 messages to K spatially dispersed users. For k ∈ [[1, K]], In this part, we derive an achievable secrecy rate of a single 300

256 we denote by uk the confidential message relative to the user existing within the MU framework. Since the K messages 301

257 kth user. The K messages are confidential and Alice has are confidential, when Alice wants to communicate with a 302

258 to communicate each message to its intended user while certain user in the network, the remaining users are treated 303

259 keeping each user unaware of the other messages. The N × 1 as eavesdroppers to this communication link. Therefore, for 304

260 transmitted signal is expressed as k ∈ [[1, K]], the received signal at the kth user and at the 305

K
remaining K − 1 users are expressed as 306

261 s = Wu = wk u k , (6)
k=1 yk = hTk wk uk + hTk W̄k ūk + nk
(12) 307
where W = [w1 , w2 , . . . , wK ] ∈ RN ×K is the precoding ȳk = H̄k wk uk + H̄k W̄k ūk + n̄k ,

of
262

263 matrix of the system, such that for k ∈ [[1, K]], wk ∈ RN

264 is the precoding vector relative to the kth message uk , and where ȳk , ūk and n̄k are the vectors y, u and n after removing 308
T
265 u = [u1 , u2 , . . . , uK ] is the zero-mean K × 1 vector of the kth element, respectively, and H̄k and W̄k are the matrices 309

266 confidential messages. The transmitted signal s is subject to a H and W after removing the kth row and the kth column, 310

267 peak-power constraint, i.e., amplitude constraint, expressed as respectively.1 In other words, the MISO VLC wiretap system 311

in (12) assumes that the remaining users are treated as a single 312
268 ||s||∞ ≤ A, (7) potential eavesdropper for the communication link between

ro
313

269 where A ∈ R∗+ .Without loss of generality, we assume that Alice and the kth user. Based on this discussion, an achievable 314

270 the K messages are independent and identically distributed secrecy rate of the kth Gaussian MISO VLC wiretap channel 315

271 (i.i.d) according to a generic continuous zero-mean random in (12) is given in the following theorem. 316

272 variable u that satisfies

|u| ≤ A. Consequently, ||u||∞ ≤ A, Theorem 1: An achievable secrecy rate of the MISO VLC 317
+
273 E (u) = 0 and E uuT = σu2 IK . On the other hand, in order Gaussian wiretap channel in (12) is equal to Rs,k , where 318

to satisfy the amplitude constraint in (7), we impose the ⎡ 2 ⎤

274

following constraint on the matrix W.

275 K T
⎢ 1 + au hk wi ⎥
EP
1 i=1
276 ||W||∞ ≤ 1. (8) Rs,k (pu , W) = log ⎢
⎣ 2 ⎥
⎦ 319
2 K T
1 + bu i=1 h w
k i
277 Based on the above, the signal received at the kth user, for i=k
⎡ ⎤
278 k ∈ [[1, K]], is expressed as
K
2
1 ⎢ ⎥
K
− log ⎣1 + bu hTi wk ⎦, (13) 320
279 yk = hTk wk uk + hTk wi ui + nk , (9) 2 i=1
i=k
i=1
i=k

σ2
280 where hk ∈ RN + is the channel gain vector of the kth user and where au = exp(2 hu )
2πeσ2 and bu = σu2 , such that hu denotes the 321

281 nk is a Gaussian noise sample which is N (0, σ 2 ) distributed. differential entropy of the random scalar variable u and σu2 322

As seen in (9), the first term hTk wk uk is the desired signal of

IEE

282 denotes its variance. 323

K
283 th kth user, while the second term i=1 hTk wi ui is the multi- Proof: See Appendix. 324
i=k
284 user interference (MUI) and the third term nk is the Gaussian Note that the achievable secrecy rate in (13) assumes that 325

285 noise superimposed at the reception. Consequently, the SINR all channels are perfectly known to Alice, which is a valid 326

286 at the kth received is expressed as assumption. In fact, since all users are active, Alice can 327

T 2
perfectly estimates the channel gain of each user through 328

hk wk σu2 feedbacks sent from each user. In addition, this result is valid 329
SINRk =
K T
287
2 . (10)
h w σ 2 + σ2 for any precoding matrix W and any continuous random vari- 330
i=1 k i u
i=k able u. In other words, the result of Theorem 1 is independent 331

288 In this case, the system model of the MU-MISO VLC broad- from the infinity constraint in (8) and from the choice of the 332

289 cast channel is expressed as probability distribution pu , but under the assumption that it is a 333

continuous probability distribution. However, the objective of 334

290 y = Hs + n = HWu + n, (11) the paper is developing well-structured designs of W, under 335
T T
291 where y = [y1 , y2 , . . . , yK ] , H = [h1 , h2 , . . . , hK ] and the infinity norm constraint in (8), that enhance the secrecy 336
T performance of the system.
292 n = [n1 , n2 , . . . , nK ] . 337

293 The objective of this paper is to construct linear precoding

294 schemes, represented by the precoding matrix W, that the 1 Note that ȳ represents the vector of received signals at the remaining
k
295 secrecy performance of the MU-MISO VLC broadcast channel K − 1 users before performing any decoding. However, the remaining users
are assumed to be able to work in a collaborative manner to jointly remove the
296 in (11) under the infinity norm constraint in (8). This will be interference caused for each other, which is in accordance with the worst-case
297 the focus of the next section. consideration in physical layer security studies.
ARFAOUI et al.: SECRECY PERFORMANCE OF MU MISO VLC BROADCAST CHANNELS 5

T
338 B. Secrecy Performance Measures xk = [xk,1 , xk,2 , . . . , xk,K ] . Consequently, for k ∈ [[1, K]], 384

339 Our objective here is to derive linear precoding schemes that the achievable secrecy rate Rs,k is a function of x and it is 385

340 maximize the secrecy performance with respect to certain per- re-expressed as 386

⎡ ⎤
341 formance measures under the amplitude constraint in (8). i.e., K
K
1 1 ⎢ ⎥
342 max S(pu , W) Rs,k (pu , x) = log 1+au xk,i − log ⎣1+bu xk,i⎦ 387
W 2 i=1
2 i=1
343 s.t. ||W||∞ ≤ 1, (14) ⎡ ⎤
i=k

K
344 where f (pu , W) is the objective function that represents the 1 ⎢ ⎥
− log ⎣1 + bu xi,k ⎦. (19) 388
345 secrecy performance measure of interest. Typical secrecy 2 i=1
346 performance measures include [57]: i=k

347 i) Max-min fairness (MMF): S = min Rs,k . Furthermore, we impose the following infinity norm constraint 389
1≤k≤K −1 on the matrix X. 390
K −1
ii) Harmonic mean (HM): S = K

of
348
k=1 Rs,k . √
K1 || X||∞ ≤ aH , (20) 391
K
349 iii) Proportional fairness (PF): S = k=1 Rs,k .
iv) Weighted fairness (WF): S =
K where aH = min ||H||∞ , ||H⊥1||∞ . In this case, the infinity 392
350
k=1 dk Rs,k , where
351 (dk )1≤k≤K ∈ R+ , norm constraint in (20) and the infinity norm constraint in (8) 393

352 with increasing order of achievable secrecy sum-rate and are equivalent, i.e., if one is satisfied, the other is automatically 394

353 decreasing order of user fairness. In fact, for the case where satisfied. In fact, if the constraint in (8) is satisfied, then 395

√
354

355

356

357

358

359
dk = 1 for all k ∈ [[1, K]], we have
MMF ≤ HM ≤ PF ≤ WF.

ro
However, in terms of user fairness, MMF is the best secrecy
performance measure [57]. In this subsection, we consider
all the secrecy performance measures discussed above while
assuming that the probability distribution pu is fixed and
(15)
|| X|| = ||HW||∞ ≤ ||H||∞ ||W||∞ ≤ ||H||∞ ≤ aH , (21)
and if the constraint in (20) is satisfied, then

≤ 1.
√
||W|| = ||H⊥ X||∞ ≤ ||H⊥ ||∞ || X||∞

≤ min ||H⊥ ||∞ ||H||∞ , 1
√

(22)
396

397

398

399

400
EP
360 known.
Based on the above, problem P1 can be re-written as 401
361 1) Max-Min Fairness: The max-min fairness measure aims
362 to maximize the minimum achievable secrecy rate among the P1 : Rs (pu ) = −min − z 402
x,z
363 K users, which leads to the following optimization problem. ⎧
⎪
⎪ c1 (pu , x, z) = z
364 P1 : Rs (pu ) = max min Rs,k (pu , W) ⎪
⎨ k−Rs,k (pu , x) ≤ 0, ∀k ∈ [[1, K]],
W k∈[[1,K]]
s.t. √
c2k (x) = || xk ||1
403
s.t. ||W||∞ ≤ 1. (16) ⎪
⎪
365
⎪
⎩ −a ≤ 0,
H ∀k ∈ [[1, K]].
366 Problem P1 is a max-min problem which involves two opti-
367 mization problems. The inner problem consists of finding the (23) 404

user with the lowest achievable secrecy rate for a fixed precod-
1 function (w,z)2 → −z is convex. However, the constraints
368
The 405
IEE

369 ing matrix W, whereas the outer problem involves finding the ck 1≤k≤K and ck 1≤k≤K are not convex and consequently 406
best precoding matrix that maximizes the achievable secrecy
the optimization problem P1 is not convex. However, one way
370
407
rate for a given user. Solving P1 is difficult due to the mutual
to solve problem P1 is using the convex-concave procedure
371
408
372 dependence between the optimization parameters in the inner
(CCP) [58]. CCP is a heuristic method used to find local 409
373 and outer problems and the non concavity of the achievable
solutions to problems involving the difference of convex (DC) 410
secrecy rate Rs,k . We reformulate problem P1 as
functions. Note that, for all k ∈ [[1, K]], we have
374
411

P1 : Rs (pu ) = −min − z
c1k (pu , x, z) = fk1 (pu , x, z) − gk1 (pu , x),
375
W,z (24) 412

z − Rs,k (pu , W) ≤ 0, ∀ k ∈ [[1, K]], where fk1 (w, z) and gk1 (w, z) are expressed, respectively, as
376 s.t. 413

||W||∞ − 1 ≤ 0, ⎧
K
⎪
⎪ 1
377 (17) ⎪f (pu , x, z) = z − log 1 + au
⎪ 1
xk,i ,
⎪
⎪ k
2
⎪
⎪ ⎡ i=1 ⎤
378 where z is simply an auxiliary variable. We perform the change ⎪
⎪
⎪
⎪ K
379 of optimization variable expressed as ⎪
⎨ 1 1 ⎢ ⎥
√ −1 √
gk (pu , x) = − log ⎣1 + bu xk,i ⎦
W = H⊥ X = HT HHT 2 (25) 414
380 X, (18) ⎪
⎪ i=1
⎪
⎪ ⎡ ⎤ i=k

where X = (xk,i )1≤k,i≤K is a K × K matrix, such that for ⎪

⎪
381 ⎪
⎪ 1 K
T ⎪
⎪ ⎢ ⎥
all (k, i) ∈ [[1, K]]2 , xk,i ∈ R+ . Let x = xT1 , xT2 , . . . , xTK ⎪
⎪ − log ⎣1 + bu xi,k ⎦.
382
⎪
⎩ 2
383 be the K 2 × 1 vector, such that for all k ∈ [[1, K]], i=1
i=k
6 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS

415 Clearly, the functions fk1 and gk1 are convex and, therefore, Consequently, armed with the above, the convex form of 456

416 the constraint c1k is a difference of two convex functions. problem P1 is given by 457

417 To tackle this problem, we convexify the constraint c1k through

418 a simple linearization of the function gk1 by applying the first- P1 (xl ) : Rs (pu ) = −min − z 458
x,z
419 order Taylor series approximation around a given feasible
point xl = [xTl,1 , xTl,2 , . . . , xTl,K ]T , Consequently, the convex c̃1k (pu , x, xl , z) ≤ 0, ∀k ∈ [[1, K]],
420 s.t. 459
421 form of c1k , denoted by c̃1k , is expressed as c̃2k (pu , x, xl ) ≤ 0, ∀k ∈ [[1, K]].
(35)
c̃1k (pu , x, xl , z) = fk1 (pu , x, z) − g̃k1 (pu , x, xl ),
460
422 (26)
423 where g̃K1
is the first-order Taylor series approximation of gk1 , Problem P1 (xl ) is a convex optimization problem that depends 461

424 that is expressed as on the linearization point xl and can be solved efficiently using 462

standard optimization packages [59], [60]. Finally, the detailed 463

425 g̃k1 (pu , x, xl ) = gk1 (pu , xl ) + ∇gk1 (xl )T (x − xl ), (27) iterative algorithm for solving P1 is given in Algorithm 1, 464

of
where the initial point x0 is a random feasible point that 465
426 where ∇gk1 denotes the gradient of the function gk1 with respect satisfies the constraints of problem P1 .
T 466

427 to (w.r.t) x. In addition, ∇gk1 (x) = vT1 , vT2 , . . . , vTK , where

428 for all i ∈ [[1, K]], vi = [vi,1 , vi,2 , . . . , vi,K ]T ∈ RK and it is Algorithm 1 Iterative Algorithm for Solving P1
429 expressed as
1. Initialization:
430 vi,j = −δi,k (1 − δj,k ) Ak − (1 − δi,k ) δj,k Bk , (28) i) Estimate H and σ 2 .
ii) Choose an Initial feasible point x0 .
431

432
such that
⎧
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎩
Ak =

Bk =
1
2
1 bu
21+ K
bu
K
1 + l=1 xk,l

l=1 xl,k
,
l=k

,
l=k

and δ is the Kronecker delta function defined, for all (i, j) ∈

ro (29)
2. Set: l = 0.
3. Repeat
i) Solve P1 (xl ).
ii) Assign the solution to xl+1 .
iii) Update iteration; l ← l + 1.
4. Termination: terminate step 3. when
i) |xl − xl−1 | ≤ , or
EP
433

434 N2 , as δi,j = 1, if i = j, and δi,j = 0, otherwise. ii) l = Lmax .

435 In the same spirit, and for all k ∈ [[1, K]], the constraint
436 c2k (pu , x) can be written as the difference of two convex
437 functions as 2) Harmonic Mean: When the optimization objective is to 467

maximize the harmonic mean of the overall system, the max- 468

438 c2k (pu , x) = fk2 (pu , x, z) − gk2 (pu , x), (30) imization problem becomes as follows. 469

where K −1
439

−1
fk2 (pu , x, z) = −aH , P2 : Rs (pu ) = max K Rs,k (pu , W) 470
W
440 √ (31) k=1
gk2 (pu , x) = −|| xk ||1 . s.t. ||W||∞ ≤ 1. (36)
IEE

471

441 Clearly, the functions fk2

and gk2
are convex and, therefore,
Problem P2 is a non-linear non-convex optimization problem
the constraint c2k is a difference of two convex functions.
472
442
that is difficult to solve due to the structure of the objective
To tackle this problem, we convexify the constraint c2k through
473
443
function and the expression of the achievable secrecy rate
a simple linearization of the function gk2 by applying the first-
474
444
Rs,k (pu , W). Therefore, we adopt a suboptimal approach in 475
445 order Taylor series approximation around xl . Consequently,
solving P2 , that is detailed as follows. In fact, since the
the convex form of c2k , denoted by c̃2k , is expressed as
476
446
harmonic mean is lower than the arithmetic mean for any 477

447 c̃2k (pu , x, xl ) = −aH − g̃k2 (pu , x, xl ), (32) positive valued set of reals, we have 478

K −1
448 where g̃k2 is the first-order Taylor series approximation of gk2 , K
−1
449 that is expressed as K Rs,k (pu , W) ≤ dk Rs,k (pu , W), (37) 479

k=1 k=1
450 g̃k2 (pu , x, xl ) = gk2 (pu , xl ) + ∇gk2 (xl )T (x − xl ), (33)
1
where dk = K , for all k ∈ [[1, K]]. Consequently, a suboptimal 480
451 where ∇gk2 denotes the gradient of the function gk w.r.t x. solution for problem P2 can be given through the following
T 481

452 In this context, ∇gk (x) = pT1 , pT2 , . . . , pTK , such that for optimization problem. 482
T
453 all i ∈ [[1, K]], pi = [pi,1 , pi,2 , . . . , pi,K ] ∈ RK and it is
K

expressed as ∗
P2 : W = argmax dk Rs,k (pu , W)
454
483
1 W k=1
455 pi,j = δi,k √ . (34)
2 xk,j s.t. ||W||∞ ≤ 1. (38) 484
ARFAOUI et al.: SECRECY PERFORMANCE OF MU MISO VLC BROADCAST CHANNELS 7

485 Adopting the change of variables used in the max-min fairness Finally, after determining a suboptimal solution x∗ for 521

486 measure, problem P2 can be reformulated as problem P2 , the harmonic mean of the system is expressed as 522

K −1
487 P2 : x∗ = argminf (pu , x)
∗ −1
x Rs (pu ) = K Rs,k (pu , x ) . (46) 523

488 s.t. c2k (x) ≤ 0, ∀k ∈ [[1, K]], (39) k=1

K 3) Proportional Fairness: For the proportional fairness 524

489 where f (pu , x) = − k=1 dk Rs,k (pu , x). The function measure, the maximization problem becomes as follows. 525
490 x → f (pu , x) can be expressed as the difference of two convex K K1
491 function as follows.
P3 : Rs (pu ) = max Rs,k (pu , W) 526

492 f1 (pu , x) = f 3 (pu , x) − f2 (pu , x), (40) W

k=1
s.t. ||W||∞ ≤ 1. (47) 527
493 where f1 (pu , x) and f2 (pu , x) are expressed, respectively, as
⎧ Similar to the previous secrecy performance measure, P3 is

of
528

⎪ K K
⎪
⎪ dk a non-linear non-convex optimization problem. To over- 529
⎪
⎪ f1 (pu , x) = − log 1 + au xk,i
⎪
⎪ 2 come this, as was done before, we resort to a suboptimal 530
⎪
⎪ k=1
⎡ i=1
⎤
⎪
⎪ approach. Knowing that the geometric mean is lower than the 531
⎪
⎪
⎪
⎨
K
dk ⎢ K
⎥
arithmetic mean for any positive valued set of reals, we have 532

f2 (pu , x) = − log ⎣1 + bu xk,i ⎦ K K1

494
⎪ 2 (41) K

⎪
⎪
k=1 i=1
Rs,k (pu , W) ≤ dk Rs,k (pu , W),
⎪
⎪ ⎡ ⎤ i=k (48) 533
⎪
⎪
⎪

ro
K K k=1 k=1
⎪
⎪ − dk ⎢ ⎥
⎪
⎪ log ⎣1 + bu xi,k ⎦. where dk = K 1
, for all k ∈ [[1, K]]. Consequently, a suboptimal
⎪
⎪ 2 534
⎩ k=1 i=1 solution for problem P3 can be given through the optimization 535
i=k

problem P2 discussed in the previous part. Finally, after 536

495 To overcome the problem of non-convexity of the objec- adopting the same approach and determining a suboptimal 537
496 tive function, we apply the CCP method, described above. solution x∗ for problem P2 , the proportional fairness of the 538
497 Consequently, we start by convexifying the function f through system is expressed as 539
498 a simple linearization of the function f2 by applying the first- K K1
EP
499 order Taylor series approximation around a given point xl . The
500 convex form of f , denoted by f˜, is expressed as Rs (pu ) = Rs,k (pu , x∗ ) . (49) 540

k=1
501 f˜ (pu , x, xl ) = f1 (pu , x) − f˜2 (pu , x, xl ), (42) 4) Weighted Fairness: When the optimization objective is to 541

maximize the weighted secrecy sum-rate of the overall system,

where f˜2 is the first-order Taylor series approximation of f2 ,
542
502
the maximization problem becomes as follows. 543
503 that is expressed as
K

504 f˜2 (pu , x, xl ) = f2 (xl ) + ∇f2 (xl )T (x − xl ), (43) P4 : Rs (pu ) = max dk Rs,k (pu , W) 544
W
k=1
505 in which ∇f2 denotes the gradient of the function f2 w.r.t x. s.t. ||W||∞ ≤ 1, (50) 545
In addition, ∇f2 is expressed as
IEE

506
where for all k ∈ [[1, K]], dk ∈ R+ is an arbitrary weight 546
K
for the kth user. Problem P4 is equivalent to P2 discussed in 547
507 ∇f2 = dk ∇gk1 , (44) the previous part, and therefore, the secrecy sum-rate of the 548
k=1 system for this case can be determined by following the same 549

508 where for all k ∈ [[1, K]], ∇gk1 is given in (28). Consequently, approach developed for the harmonic mean measure. 550

509 the convex form of problem P2 , denoted by P2 (xl ), is given
C. Complexity Analysis 551
510 by
In this part, we evaluate the computational complexity of 552

511 P2 (xl ) : Rs (pu ) = −minf˜ (pu , x, xl ) the proposed precoding schemes and we compare it to that 553
x
of conventional ZF. In Algorithm 1, we employ the well
s.t. c̃2k (pu , x, xl ) ≤ 0, ∀k ∈ [[1, K]].
554
512
known interior point algorithm (IPA) in solving the invoked 555
513 (45) convex problem. Therefore, we employ the number of Newton 556

steps, denoted by Ns , as a complexity measure. The number

Problem P2 (xl ) is a convex optimization problem that depends
557
514
of Newton steps denotes the number of recursive iterations 558
515 on the linearization point xl and can be solved efficiently
till convergence from a given starting point, i.e., the number 559
516 by using standard optimization packages [59], [60]. Finally,
of required recursive steps to reach a local solution. Based 560
517 based on the above analysis, the detailed iterative algorithm
on [61], the worst-case Ns to reach a local solution in a non-
for solving P2 is given in Algorithm 1, where it suffices to
561
518
linear convex problem is expressed as
replace P1 (xl ) by P2 (xl ) and the initial point x0 is a random
562
519

520 feasible point that satisfies the constraints of problem P2 . Ns ∼ problem size, (51) 563
8 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS

TABLE II
N UMBER OF N EWTON S TEPS OF A LGORITHM 1

564 where the problem size is the number of optimization

565 scalar variables. In Algorithm 1, we are supposed to solve
566 a non-linear convex problem at most L-times, and thus,
567 we are employing the IPA at most L-times. Based on this,
568 the worst-case complexity of our proposed schemes and of
569 the ZF precoding are presented in Table II.
570 Thus, for the adopted secrecy performance measures,

of
571 we have
Ns |proposed schemes √
572 ≈ K, (52)
Ns |ZF
573 where K is the number of active users. In other words,
the computational Fig. 2. MU-MISO VLC system with N = 16 LEDs fixtures and K = 2
574
√ complexity of our proposed schemes is users.
575 approximately K times higher than of ZF.

576

577

578

579

580
ro
D. The Truncated Generalized Normal (TGN) Distribution
The precoding schemes developed previously are valid for
any continuous input distribution pu with support [−A, A].
In our work, we adopt the TGN distribution as input signaling.
The TGN distribution is a class of real parametric continuous
probability distributions over a bounded interval, that adds a
respectively, as
⎧
⎪
⎪
⎪
⎨hu = log
⎪

⎪
⎪
⎪ 2 α

2
2αΓ( β1 )
β

⎩σu = Γ( 1 ) γ β , α
⎪
β
1 A β
3

A
+ η(α, β),
β
,

(54)
606

607
EP
β
γ β ,( α ) γ β1 +1,( A α)
581

582 shape parameter to the truncated normal distribution. A TGN where η(α, β) = log 1 + β
. 608
Γ( β ) γ β1 ,( A
α)
583 distribution over [−A, A], A ∈ R+ , with a position parameter
584 μ ∈ [−A, A], a scale parameter α ∈ R∗+ and a form In the above analysis, and since au = exp(2 2πeσ2
hu )
and 609
2
σu
585 parameter β ∈ R+ is denoted by TGN(−A, A, μ, α, β) and bu = σ2 , the secrecy rates Rs (pu ) and RZF (pu ) can be 610

586 its probability density function is given by enhanced by optimizing over the parameters α and β of 611

⎧ y−μ the probability distribution pu . However, since these maxi- 612

⎪ 1
⎪
⎪
⎨ α φ( α ) mizations are not straightforward, we adopt a discretization 613

−A ≤ y ≤ A approach as follows. Let D ∈ R+ and M1 , M2 ∈ N

pu (y) = Φ( A−μ ) − Φ( −A − μ )
614
587 (53)
⎪
⎪ α such that M1 = 0 and M2 = 0. Then we consider a 615
⎪
⎩ α
0 otherwise, two dimensional grid G (D) = (Gi,j (B))1≤i≤M1 , where 616
IEE

1≤j≤M2

588 where φ is the standard generalized normal distribution that Gi,j (D) = (αi , βj ) such that for all i ∈ [[1, M1 ]], αi = B×i
M1 , 617

589
β
is defined, ∀ x ∈ R, as φ(y) = 2Γ(β 1 ) exp−|y| , and Φ is and for all j ∈ [[1, M2 ]], βj = B×j
M2 . Based on this, for a given 618
β
secrecy performance measure, the highest secrecy sum-rate of
that is defined, ∀ y ∈ R,
619
590 its cumulative distribution function
y β
γ β1 ,( β ) our proposed schemes and of ZF precoding are expressed, 620
591 as Φ(y) = 12 + sign(y) 2Γ( 1 ) , where sign and γ denote respectively, as 621
β
592 the sign function and the incomplete gamma function, respec- ⎧ ∗
tively. The expected value of a TGN(−A, A, μ, α, β) is equal ⎨Rs = max Rs (pu ),
593
(α,β)∈G(B)
(55) 622
594 to μ. Furthermore, according to the parameters μ, α and β, ⎩RZF ∗
= max RZF (pu ).
595 the TGN class over [−A, A] includes: (α,β)∈G(B)

596 • The truncated Laplace distribution when β = 1,

597 • The truncated normal distribution when β = 2,
IV. S IMULATIONS R ESULTS 623
598 • The uniform distribution when μ = 0, α = A and
599 β → ∞. A. Simulations Settings 624

600 Hence, through an optimal design of the parameters μ, α To validate our proposed schemes, we consider a typical 625

601 and β, the secrecy performance can be improved. VLC system consisting of a single room as shown in Fig. 2. 626

602 In our scheme, we assume that random scalar variable u A Cartesian coordinate system, shown in Fig. 2, is used. 627

603 follows a TGN(−A, A, 0, α, β). The position parameter μ is The parameters of the room, the transmitter and the users 628

604 set to zero since u is zero-mean. In this case, the differ- are given in Table II. Alice is equipped with N = 16 629

605 ential entropy hu and the variance σu2 of u are expressed, fixture of LEDs, located at the ceiling of the room and at 630
ARFAOUI et al.: SECRECY PERFORMANCE OF MU MISO VLC BROADCAST CHANNELS 9

TABLE III
MU-MISO VLC S YSTEM PARAMETERS

of
Fig. 3. Average secrecy rates R∗s and R∗ZF versus A2 for the max-min
fairness (MMF), the harmonic mean (HM), the proportional fairness (PF) and
the weighted fairness (WF). The number of users is K = 4.

ro
EP
631 (x, y) ∈ {1, 2, 3, 4} × {1, 2, 3, 4}. The users height measured
632 from the room’s floor is 1m. Based on Table III and equations
633 (1), (2) and (3), the average noise variance at the receivers
634 is σ 2 = −98.82 dBm. The simulation results are obtained
635 through 105 independents Monte Carlo trials on the locations
636 of users within the room. In addition, the central processing
637 unit (CPU) of the machine on which all the simulations are
638 performed is an Intel Core i5 from the second generation
639 that has a dual-core, a basic frequency of 2.40 GHz and Fig. 4. Average secrecy rate R∗s for the max-min fairness (MMF),
640 a maximum turbo frequency of 3.40 GHz. Moreover, We the harmonic mean (HM), the proportional fairness (PF) and the weighted
fairness (WF) for the number of users K = 4 and K = 2.
= 10−3 and Lmax = 10 as stopping criterion for
IEE

641 use
642 Algorithm 1. Finally, The best input distribution used in all C. Complexity Analysis 659
643 simulations is TGN(−A, A, 0, A, 2).
Another metric that we can use to evaluate the complexity 660

of the proposed schemes is the execution time, i.e., the amount 661

644 B. Secrecy Performance of time required to obtain the best precoding matrix. Fig. 5 662

presents the average execution time in seconds of the proposed 663

645 In this subsection we evaluate the secrecy performance of precoding schemes and of ZF precoding versus A2 in dBm, 664
646 our proposed schemes. Fig. 3 presents the average secrecy rate where the number of active users is K = 4. This figure shows 665
647 Rs∗ and RZF∗
of the four secrecy performance measures con- that the complexity of our schemes is about two times that 666
648 sidered versus A2 in dBm, where the number of active users of ZF. This result is also expected since for a given number K 667
649 is K = 4. Fig. 3 shows that, for all the considered secrecy of active users, our proposed scheme consists of optimizing 668
650 performance measures, the proposed precoding schemes out- over K 2 + 1 variables whereas ZF precoding consists of 669
651 perform the conventional ZF precoding. This result is expected optimizing over K + 1 variables only. 670
652 since, as mentioned in subsection III-B, ZF precoding is a
653 special case of our proposed scheme. On the other hand,
654 Fig. 4 presents the average secrecy rate Rs∗ of the proposed D. Convergence Behavior 671

655 precoding schemes versus A2 in dBm for the four secrecy Fig. 6 presents the convergence behavior of Algorithm 1 672

656 performance measures when K = 2 and 4. As shown in the when applied to our precoding scheme, where the number 673

657 figure, the secrecy performance improves with the number of of active users is K = 4. The amplitude constraint is 674

658 users. A2 = 0 dBm. The convergence here is measured in terms 675

10 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS

techniques outperform the conventional ZF precoding scheme 692

in terms of secrecy performance. However, this improvement 693

comes at the expense of an increase of the execution time 694

of the proposed algorithms, which gives a trade-off between 695

complexity and performance improvement. 696

A PPENDIX 697

P ROOF OF T HEOREM 1 698

Based on the results of [33] and [34], a lower bound on 699

the secrecy capacity of the kth MISO VLC Gaussian wiretap 700

channel in (12) can be obtained as follows. 701

of
Ck ≥ max [I (uk ; yk ) − I (uk ; ȳk |ūk ) − I (uk ; ūk )]+ 702
p(uk )
⎡ ⎤+
a
≥ ⎣I (uk ; yk ) − I (uk ; ȳk |ūk ) − I (uk ; ūk )⎦
Fig. 5. Average execution time of the proposed scheme and of ZF precoding
versus A2 for the max-min fairness (MMF), the harmonic mean (HM), !
703

the proportional fairness (PF) and the weighted fairness (WF). The number =0
of users is K = 4. b +
= [h (yk )−h (yk |uk )−h (ȳk |ūk )+h (ȳk |uk , ūk )] , (56)

ro
704

where inequality a holds by choosing any probability dis- 705

tribution p(uk ), and I (uk ; ūk ) = 0, since all the messages 706

are statistically independent. Now, we develop each term of 707

equation (47)-b. Since for all i ∈ [[1, K]], ui is continuous, 708

we can use the entropy power inequality (EPI) to determine 709

a lower bound on h(yk ). Precisely, let (α)1≤i≤K ∈ R∗ and 710

EP
(x)1≤i≤K be K random scalar variables. Therefore, using the 711

EPI, we can obtain a lower bound on the differential entropy 712

K
of the random variable z = i=1 αi xi as follows. 713

K K
1
h (z) = h αi xi ≥ log e2h(αi xi ) 714

i=1
2 i=1
K
1
= log e2 log(|αi |)+2h(xi ) 715
2 i=1
Fig. 6. Convergence behavior of Algorithm 1 when K = 4 and K

IEE

A2 = 0 dBm. 1 2 2h(xi )
= log αi e . (57) 716
2 i=1
676 of the relative error between consecutive iterations and that
677 should be less than the adopted . This figure shows that Therefore, and based on (57), h(yk ) can be lower bounded as 717

678 on average, Algorithm 1 requires at most seven iterations to K

679 converge for the max-min fairness measure whereas it requires 1
e2h(hk wi ui ) + e2h(nk )
T
at most five iterations to converge for the other three measures. h(yk ) ≥ log 718
680
2 i=1
K
V. C ONCLUSION 1 T 2 2hu
681 2
= log h k wi e + 2πeσ 719
682 In this paper, we considered the MU-MISO VLC Broadcast 2 i=1
channel with confidential messages. We proposed new precod- K

e2hu T
683
1 2 1
684 ing schemes aiming to maximize typical secrecy performance = log 1 + h wi + log 2πeσ 2 ,
2πeσ 2 i=1 k
720

685 measures of the underlying system subject to a peak-amplitude 2 2

686 constraint. Moreover, due to the amplitude constraint imposed (58) 721

to the channel input, we adopted the TGN distribution as input

687
1 2
h (nk ) = 2 log 2πeσ
signaling and we optimized over its parameters to enhance where we used the fact that . On the

688 722

689 the secrecy performance of the system. We compared the K T

690 performance of our scheme to the conventional ZF precoding other hand, h (yk |uk ) = h i=k hk wi + nk and it can be 723
i=1
691 scheme and we showed through the numerical results that our upper bounded by the differential entropy of random Gaussian 724
ARFAOUI et al.: SECRECY PERFORMANCE OF MU MISO VLC BROADCAST CHANNELS 11

725 variable with the same variance as [11] S. Wu, H. Wang, and C. H. Youn, “Visible light communications for 5G 775
⎡ ⎤ wireless networking systems: From fixed to mobile communications,” 776

K
1 ⎢ T 2 ⎥
IEEE Netw., vol. 28, no. 6, pp. 41–45, Nov. 2014. 777

726 h(yk |uk ) ≤ log ⎣2πe hk wi σu2 + 2πeσ 2 ⎦ [12] Y. Liang, H. V. Poor, and S. Shamai (Shitz), “Physical layer security in 778
2 broadcast networks,” Secur. Commun. Netw., vol. 2, no. 3, pp. 227–238, 779
i=k
i=1 May/Jun. 2009. 780
⎡ ⎤ [13] T. Bilski, “New threats and innovative protection methods in wireless 781
K
1 ⎢ σ2 T 2⎥ 1 transmission systems,” J. Telecommun. Inf. Technol., vol. 3, no. 3, 782

727 = log ⎣1 + u2 hk wi ⎦ + log 2πeσ 2 . pp. 26–33, Mar. 2014. 783

2 σ 2 [14] Y.-S. Shiu, S. Y. Chang, H.-C. Wu, S. C.-H. Huang, and 784
i=k
i=1 H.-H. Chen, “Physical layer security in wireless networks: A tutorial,” 785

728 (59) IEEE Wireless Commun., vol. 18, no. 2, pp. 66–74, Apr. 2011. 786

[15] A. D. Wyner, “The wire-tap channel,” Bell Syst. Tech. J., vol. 54, no. 8, 787

729 In addition, h (ȳk |ūk ) = h H̄k wk + n̄k can be also upper pp. 1355–1387, Oct. 1975. 788

[16] I. Csiszár and J. Korner, “Broadcast channels with confidential mes- 789
730 bounded by the differential entropy of random Gaussian vector sages,” IEEE Trans. Inf. Theory, vol. IT-24, no. 3, pp. 339–348, 790
731 with the same covariance matrix as May 1978.

of
791

[17] A. Mostafa and L. Lampe, “Physical-layer security for MISO visible 792
732 h (ȳk |ūk )
" # #$ light communication channels,” IEEE J. Sel. Areas Commun., vol. 33, 793

1 #
2 K−1 #
T σu
2 # no. 9, pp. 1806–1818, Sep. 2015. 794
≤ log 2πeσ #
733
2 # H̄k wk H̄k wk σ 2 + IK−1 # [18] M.-A. Arfaoui, Z. Rezki, A. Ghrayeb, and M. S. Alouini, “On the 795

" $ input distribution and optimal beamforming for the MISO VLC wiretap 796

1 ## ##2 σ 2 K −1 channel,” in Proc. IEEE GlobalSIP, Washington, DC, USA, Dec. 2016,
= log 1 + ##H̄k wk ##2 u2 +
797
734 log 2πeσ 2 pp. 970–974.
2 2
798
σ
⎡ ⎤ [19] H. Zaid, Z. Rezki, A. Chaaban, and M. S. Alouini, “Improved achiev- 799
able secrecy rate of visible light communication with cooperative

ro
800
K
1 ⎢ σ2 T 2⎥ K −1 jamming,” in Proc. IEEE GlobalSIP, Orlando, FL, USA, Dec. 2015,
= log ⎣1 + u2 log 2πeσ 2 .
801
735 hi wk ⎦ + pp. 1165–1169.
2 σ 2 802
i=k [20] M. A. Arfaoui, Z. Rezki, A. Ghrayeb, and M. S. Alouini, “On the secrecy 803
i=1
capacity of MISO visible light communication channels,” in Proc. IEEE 804
736 (60) Globecom, Washington, DC, USA, Dec. 2016, pp. 1–7. 805

[21] F. Wang et al., “Optical jamming enhances the secrecy performance of 806
737 Furthermore, we have the generalized space-shift-keying-aided visible-light downlink,” IEEE 807

K −1 Trans. Commun., vol. 66, no. 9, pp. 4087–4102, Sep. 2018. 808

738 log 2πeσ 2 . (61)

h (ȳk |uk , ūk ) = h (ȳk |uk ) = h (n̄k ) = [22] A. Mostafa and L. Lampe, “Optimal and robust beamforming for secure 809
2
EP
transmission in MISO visible-light communication links,” IEEE Trans. 810

739 Finally, by substituting the different terms of equation Signal Process., vol. 64, no. 24, pp. 6501–6516, Dec. 2016. 811

740 (56)-b by their expressions, we obtain the expression of the [23] A. Mostafa and L. Lampe, “Securing visible light communications via 812
friendly jamming,” in Proc. IEEE Globecom Workshops, Austin, TX, 813
741 achievable secrecy rate Rs,k (pu ) given in Theorem 1, which USA, Dec. 2014, pp. 524–529. 814
742 completes the proof. [24] G. Pan, J. Ye, and Z. Ding, “On secure VLC systems with spatially 815
random terminals,” IEEE Commun. Lett., vol. 21, no. 3, pp. 492–495, 816
Mar. 2017. 817
743 R EFERENCES [25] S. Cho, G. Chen, and J. P. Coon, “Secrecy analysis in visible light 818
744 [1] A. Jovicic, J. Li, and T. Richardson, “Visible light communication: communication systems with randomly located eavesdroppers,” in Proc. 819
745 Opportunities, challenges and the path to market,” IEEE Commun. Mag., IEEE ICC, Paris, France, May 2017, pp. 475–480. 820

746 vol. 51, no. 12, pp. 26–32, Dec. 2013. [26] S. Cho, G. Chen, and J. P. Coon, “Securing visible light communi- 821
747 [2] S. Rajagopal, R. D. Roberts, and S.-K. Lim, “IEEE 802.15.7 visible cation systems by beamforming in the presence of randomly distrib- 822
light communication: Modulation schemes and dimming support,” IEEE uted eavesdroppers,” IEEE Trans. Wireless Commun., vol. 17, no. 5,
IEE

748 823
749 Commun. Mag., vol. 50, no. 3, pp. 72–82, Mar. 2012. pp. 2918–2931, May 2018. 824
750 [3] A. R. Ndjiongue, H. C. Ferreira, and T. Ngatched, “Visible light [27] M.-A. Arfaoui, A. Ghrayeb, and C. Assi, “Enhancing the secrecy 825
751 communications (VLC) technology,” in Wiley Encyclopedia of Electrical performance of Gaussian MISO VLC wiretap channels with randomly 826
752 and Electronics Engineering. Hoboken, NJ, USA: Wiley, Jun. 2015. located eavesdroppers,” in Proc. IEEE ICC, Kansas City, MO, USA, 827
753 [4] S. Gao, “Performance study for indoor visible light communication May 2018, pp. 1–6. 828
754 systems,” Ph.D. dissertation, Fac. Eng., School Elect. Eng. Comput. Sci., [28] M. A. Arfaoui, A. Ghrayeb, and C. Assi, “Secrecy rate closed-form 829
755 Univ. Ottawa, Ottawa, ON, Canada, 2013. expressions for the SISO VLC wiretap channel with discrete input sig- 830
756 [5] D. K. Borah, A. C. Boucouvalas, C. C. Davis, S. Hranilovic, and naling,” IEEE Commun. Lett., vol. 22, no. 7, pp. 1382–1385, Jul. 2018. 831
757 K. Yiannopoulos, “A review of communication-oriented optical wireless [29] S. Cho, G. Chen, and J. P. Coon, “Physical layer security in 832
758 systems,” EURASIP J. Wireless Commun. Netw., vol. 2012, pp. 1–28, visible light communication systems with randomly located collud- 833
759 Dec. 2012. ing eavesdroppers,” IEEE Wireless Commun. Lett., to be published, 834
760 [6] M. A. Khalighi and M. Uysal, “Survey on free space optical communi- doi: 10.1109/LWC.2018.2820709. 835
761 cation: A communication theory perspective,” IEEE Commun. Surveys
762 Tuts., vol. 16, no. 4, pp. 2231–2258, 4th Quart., 2014. [30] O. Ozel, E. Ekrem, and S. Ulukus, “Gaussian wiretap channel with 836

763 [7] D. Karunatilaka, F. Zafar, V. Kalavally, and R. Parthiban, “LED based amplitude and variance constraints,” IEEE Trans. Inf. Theory, vol. 61, 837

764 indoor visible light communications: State of the art,” IEEE Commun. no. 10, pp. 5553–5563, Oct. 2015. 838

765 Surveys Tuts., vol. 17, no. 3, pp. 1649–1678, Aug. 2015. [31] A. Lapidoth, S. M. Moser, and M. A. Wigger, “On the capacity of 839

766 [8] H. Elgala, R. Mesleh, and H. Haas, “Indoor optical wireless commu- free-space optical intensity channels,” IEEE Trans. Inf. Theory, vol. 55, 840

767 nication: Potential and state-of-the-art,” IEEE Commun. Mag., vol. 49, no. 10, pp. 4449–4461, Oct. 2009. 841

768 no. 9, pp. 56–62, Sep. 2011. [32] M.-A. Arfaoui, A. Ghrayeb, and C. Assi, “Achievable secrecy sum-rate 842

769 [9] A. Chaaban, Z. Rezki, and M.-S. Alouini, “Fundamental limits of paral- of the MISO VLC broadcast channel with confidential messages,” in 843

770 lel optical wireless channels: Capacity results and outage formulation,” Proc. IEEE Globecom, Singapore, Dec. 2016, pp. 1–6. 844

771 IEEE Trans. Commun., vol. 65, no. 1, pp. 296–311, Jan. 2017. [33] R. Liu, I. Maric, P. Spasojević, and R. D. Yates, “Discrete memoryless 845
772 [10] A. A. Farid and S. Hranilovic, “Capacity bounds for wireless optical interference and broadcast channels with confidential messages: Secrecy 846
773 intensity channels with Gaussian noise,” IEEE Trans. Inf. Theory, rate regions,” IEEE Trans. Inf. Theory, vol. 54, no. 6, pp. 2493–2507, 847
774 vol. 56, no. 12, pp. 6066–6077, Dec. 2010. Jun. 2008. 848
12 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS

849 [34] R. Liu and H. V. Poor, “Secrecy capacity region of a multiple-antenna [57] Z.-Q. Luo and S. Zhang, “Dynamic spectrum management: Complex- 921
850 Gaussian broadcast channel with confidential messages,” IEEE Trans. ity and duality,” IEEE J. Sel. Topics Signal Process., vol. 2, no. 1, 922
851 Inf. Theory, vol. 55, no. 3, pp. 1235–1249, Mar. 2009. pp. 57–73, Feb. 2008. 923
852 [35] G. Geraci, M. Egan, J. Yuan, A. Razi, and I. B. Collings, “Secrecy sum- [58] T. Lipp and S. Boyd, “Variations and extension of the convex–concave 924
853 rates for multi-user MIMO regularized channel inversion precoding,” procedure,” Optim. Eng., vol. 17, no. 2, pp. 263–287, Nov. 2015. 925
854 IEEE Trans. Commun., vol. 60, no. 11, pp. 3472–3482, Nov. 2012. [59] M. Grant, S. Boyd, and Y. Ye. (Dec. 2017). CVX: MATLAB 926
855 [36] G. Geraci, R. Couillet, J. Yuan, M. Debbah, and I. B. Collings, “Large Software for Disciplined Convex Programming. [Online]. Available: 927
856 system analysis of linear precoding in MISO broadcast channels with http://cvxr.com/cvx/ 928
857 confidential messages,” IEEE J. Sel. Areas Commun., vol. 31, no. 9, [60] J. Lofberg, “YALMIP: A toolbox for modeling and optimization in 929
858 pp. 1660–1671, Sep. 2013. MATLAB,” in Proc. IEEE ICRA, New Orleans, LA, USA, Sep. 2004, 930
859 [37] G. Geraci, A. Y. Al-Nahari, J. Yuan, and I. B. Collings, “Linear pp. 284–289. 931
860 precoding for broadcast channels with confidential messages under [61] Y. Nesterov and A. Nemirovskii, Interior-Point Polynomial Algorithms 932
861 transmit-side channel correlation,” IEEE Commun. Lett., vol. 17, no. 6, in Convex Programming, vol. 13. Philadelphia, PA, USA: SIAM, 1994. 933
862 pp. 1164–1167, Jun. 2013.
863 [38] G. Geraci, H. S. Dhillon, J. G. Andrews, J. Yuan, and I. B. Collings,
864 “Physical layer security in downlink multi-antenna cellular networks,”
865 IEEE Trans. Commun., vol. 62, no. 6, pp. 2006–2021, Jun. 2014.

of
866 [39] G. Geraci, S. Singh, J. G. Andrews, J. Yuan, and I. B. Collings,
867 “Secrecy rates in broadcast channels with confidential messages and
868 external eavesdroppers,” IEEE Trans. Wireless Commun., vol. 13, no. 5, Mohamed Amine Arfaoui received the B.E. degree 934
869 pp. 2931–2943, May 2014. in electrical and computer engineering from the 935
870 [40] A. Wiesel, Y. C. Eldar, and S. Shamai (Shitz), “Zero-forcing precoding École Polytechnique de Tunisie, Tunisia, in 2015, 936
871 and generalized inverses,” IEEE Trans. Signal Process., vol. 56, no. 9, and the M.Sc. degree in information systems engi- 937
872 pp. 4409–4418, Sep. 2008. neering from Concordia University, Montreal, QC, 938
873 [41] Q. H. Spencer, A. L. Swindlehurst, and M. Haardt, “Zero-forcing meth- Canada, in 2017, where he is currently pursuing the 939
874 ods for downlink spatial multiplexing in multiuser MIMO channels,” Ph.D. degree in information systems engineering. 940
875 IEEE Trans. Signal Process., vol. 52, no. 2, pp. 461–471, Feb. 2004. His current research interests include communication

ro
941
876 [42] C. B. Peel, B. M. Hochwald, and A. L. Swindlehurst, “A vector- theory, optical communications, and physical layer 942
877 perturbation technique for near-capacity multiantenna multiuser security. 943
878 communication—Part I: Channel inversion and regularization,” IEEE
879 Trans. Commun., vol. 53, no. 1, pp. 195–202, Jan. 2005.
880 [43] V. Stankovic and M. Haardt, “Generalized design of multi-user MIMO
881 precoding matrices,” IEEE Trans. Wireless Commun., vol. 7, no. 3,
882 pp. 953–961, Mar. 2008.
883 [44] T. V. Pham and A. T. Pham, “Max-min fairness and sum-rate maximiza-
884 tion of MU-VLC local networks,” in Proc. IEEE Globecom Workshops,
San Diego, CA, USA, Dec. 2015, pp. 1–6. Ali Ghrayeb received the Ph.D. degree in electrical
EP
885 944
886 [45] T. V. Pham, H. Le-Minh, and A. T. Pham, “Multi-user visible light engineering from The University of Arizona, 945
887 communication broadcast channels with zero-forcing precoding,” IEEE Tucson, AZ, USA, in 2000. He was with Concordia 946
888 Trans. Commun., vol. 65, no. 6, pp. 2509–2521, Jun. 2017. University, Montreal, QC, Canada. He is currently 947
889 [46] T. V. Pham and A. T. Pham, “On the secrecy sum-rate of MU-VLC a Professor with the Department of Electrical and 948
890 broadcast systems with confidential messages,” in Proc. IEEE CSNDSP, Computer Engineering, Texas A&M University 949
891 Prague, Czech Republic, Jul. 2016, pp. 1–6. at Qatar. He has co-authored the book Coding 950
892 [47] T. V. Pham and A. T. Pham, “Secrecy sum-rate of multi-user MISO for MIMO Communication Systems (Wiley, 2008). 951
893 visible light communication systems with confidential messages,” Opt.- His research interests include wireless and mobile 952
894 Int. J. Light Electron Opt., vol. 151, pp. 65–76, Dec. 2017. communications, physical layer security, massive 953
895 [48] A. Mostafa and L. Lampe, “On linear precoding for the two- MIMO, wireless cooperative networks, and ICT 954
896 user MISO broadcast channel with confidential messages and per- for health applications. He was a co-recipient of the IEEE GLOBECOM 955
897 antenna constraints,” IEEE Trans. Signal Process., vol. 65, no. 22, 2010 Best Paper Award. He served as an instructor or a co-instructor in 956
898 pp. 6053–6068, Nov. 2017. technical tutorials at several major IEEE conferences. He served as the 957
[49] S. Hranilovic and F. R. Kschischang, “Optical intensity-modulated direct Executive Chair for the 2016 IEEE WCNC Conference and the TPC Co-Chair
IEE

899 958
900 detection channels: Signal space and lattice codes,” IEEE Trans. Inf. for the Communications Theory Symposium at the 2011 IEEE GLOBECOM. 959
901 Theory, vol. 49, no. 6, pp. 1385–1399, Jun. 2003. He has served on the editorial board of several IEEE and non-IEEE journals. 960
902 [50] T. Komine and M. Nakagawa, “Fundamental analysis for visible-
903 light communication system using LED lights,” IEEE Trans. Consum.
904 Electron., vol. 50, no. 1, pp. 100–107, Feb. 2004.
905 [51] J. M. Kahn and J. R. Barry, “Wireless infrared communications,” Proc.
906 IEEE, vol. 85, no. 2, pp. 265–298, Feb. 1997.
907 [52] L. Zeng et al., “High data rate multiple input multiple output (MIMO)
908 optical wireless communications using white led lighting,” IEEE J. Sel. Chadi M. Assi received the Ph.D. degree from 961
909 Areas Commun., vol. 27, no. 9, pp. 1654–1662, Dec. 2009. The City University of New York (CUNY) in 2003. 962
910 [53] V. W. S. Chan, “Free-space optical communications,” J. Lightw. Technol., He is currently a Full Professor with Concordia 963
911 vol. 24, no. 12, pp. 4750–4762, Dec. 2006. University. His current research interests are in the 964
912 [54] W. Guan et al., “A novel three-dimensional indoor positioning algorithm areas of network design and optimization, network 965
913 design based on visible light communication,” Opt. Commun., vol. 392, modeling, and network reliability. He was a recipient 966
914 pp. 282–293, Jun. 2017. of the prestigious Mina Rees Dissertation Award 967
915 [55] A. Yesilkaya, O. Karatalay, A. S. Ogrenci, and E. Panayirci, “Channel from CUNY in 2002 for his research on wavelength- 968
916 estimation for visible light communications using neural networks,” in division multiplexing optical networks. He is on the 969
917 Proc. IEEE IJCNN, Vancouver, BC, Canada, Jul. 2016, pp. 320–325. Editorial Board of the IEEE C OMMUNICATIONS 970
918 [56] X. Chen and M. Jiang, “Adaptive statistical Bayesian MMSE chan- S URVEYS AND T UTORIALS , the IEEE T RANSAC - 971
919 nel estimation for visible light communication,” IEEE Trans. Signal TIONS ON C OMMUNICATIONS, and the IEEE T RANSACTIONS ON V EHICU - 972
920 Process., vol. 65, no. 5, pp. 1287–1299, Mar. 2016. LAR T ECHNOLOGIES . 973
IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS 1