Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

Covert Communication with Power Uncertainty for

D2D Content Sharing


Cheng Wan, Dan Wu, Meng Wang, Xin Shi, and Xinrong Guan
Army Engineering University of PLA, 210007, Nanjing, China
E-mail: chengwan1996@126.com, wujing1958725@126.com, njnu wangmeng@sina.com, shi1995@pku.edu.cn,
geniusg2017@gmail,com.

Abstract—With the rapid growth of mobile data traffic, device-


to-device (D2D) content sharing has attracted wide attention. Warden
However, due to the openness of wireless channel, D2D content Does D1 send covert
sharing may be detected by a malicious warden. In this paper, content?
covert communication is used to address the security issue of D1
D2D content sharing in dedicated mode. Specifically, when D2D
pair wants to share covert content without being detected , other
D2D pairs who reuse the same resource block (RB) transmit
with random power to produce co-channel interference confusing D2 D3
the warden. We derive the optimal threshold and the expected
detection error rate of warden under the cases where the number
of D2D pairs reusing the same RB is various. Extensive numerical
results show that increasing the number and transmission power DN
of D2D pair can reduce the probability of content sharing being Fig. 1: D2D content sharing in dedicated mode.
detected and better guarantee D2D content sharing security in
dedicated mode.
Index Terms—D2D content sharing, D2D dedicated mode, noise uncertain [5]. iii) power uncertainty uses random trans-
transmission detection, covert communication performance. mission power to puzzle the warden [6]. Since covert commu-
nication can guarantee the security of transmission behavior,
it has broad application prospects in D2D content sharing.
I. I NTRODUCTION
Authors in [7] use channel uncertainty to influence warden’s
In recent years, device-to-device (D2D) content sharing has detection to help D2D pairs implement covert communication
attracted wide attention because it can reduce the pressure for D2D underlaying cellular networks. However, after the
of base station (BS) and improve the spectral efficiency [1]. introduction of the D2D technology in the cellular system, the
However, due to the openness of the wireless channel, the D2D pairs may also operate in a dedicated mode according to
security of D2D content sharing is a necessary concern. On the radio resources allocated by the communication system
one hand, traditional encryption technology ensures that the to the D2D pairs. In this case, channel uncertainty is not
content is not obtained by setting the cryptographic keys. To applicable because the spectra are orthogonal to each other.
solve this problem, on the other hand, physical layer security In this paper, we use power uncertainty to help D2D pairs
realizes security protection from aspect of wireless channel, to implement covert communication in dedicated mode, where
which also ensures the security of content [2, 3]. However, one resource block (RB) is reused by multiple D2D pairs.
neither of these technologies guarantees that D2D content When a D2D pair wants to share covert contents, other D2D
sharing will not be detected. And then, covert communication, pairs reusing the same RB send contents with random trans-
a cutting-edge security technology interferes with warden mission power, which causes the uncertainty of interference
to ensure that transmission behavior is not detected, which for warden. We consider the cases where the number of D2D
provides a new secure idea for D2D content sharing. pairs reusing the same RB is various, and then the optimal
In covert communication, the uncertainty of interference threshold and the expected detection error rate of warden are
makes warden confused about the existence of transmission derived, respectively. Finally, comparing the numerical results,
behavior. In general, it is mainly divided into the following we find that the security performance is improved when the
three categories. i) Channel uncertainty is studied in fading number and transmission power of D2D pair increase.
channels, where warden is uncertain about the channel knowl-
edge [4]. ii) Noise uncertainty utilizes aggregate interference II. S YSTEM M ODEL
generated by a passion field of interferers, which makes the
A. Network Model
This work is supported by the National Natural Science Foundation of As shown in Fig 1, we consider a D2D content sharing
China (Grant No. 61671474), the Jiangsu Provincial Natural Science Fund
for Excellent Young Scholars (No. BK20170089), and the Jiangsu Provincial scenario, where N D2D pairs are randomly distributed in the
Natural Science Fund for Outstanding Young Scholars (No. BK20180028). network and they are in dedicated mode with reusing the same

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on June 02,2020 at 10:53:55 UTC from IEEE Xplore. Restrictions apply.
 N
 ∑ √

 pDm hDT m ,W d−α/2
DT m ,W xDm + nW , H0 ,
yW = m=2
(2)

 √ ∑
N √
 pD1 hDT 1 ,W d−α/2
DT 1 ,W xD1 +
−α/2
PDm hDT m ,W dDT m ,W xDm + nW , H1 ,
m=2

 N
 ∑

 pDm |hDT m ,W |2 d−α 2
H0 ,
DT m ,W + σW ,
m=2
TW = ∑
(4)


N
 pD1 |hDT 1 ,W |2 d−α
DT 1 ,W +
pDm |hDT m ,W |2 d−α
DT m ,W + σW , H1 .
2
m=2

RB. Each D2D pair contains a D2D content transmitter (DT) by [9]
and a D2D content receiver (DR). The N D2D pairs comprise D1
D = {D1 , · · ·, Dn , · · ·, DN }. Accordingly, the N DTs and TW ≷ γ, (3)
D0
DRs comprise DT = {DT1 , · · ·, DTn , · · ·, DTN } and DR =
{DR1 , · · ·, DRn , · · ·, DRN }. For a period of time, D1 wants l

y (k) 2
where TW = 1
l W defines as the average power
to send covert contents, meanwhile, there is a warden in the k=1
network who wants to detect whether the D1 transmits covert received by the warden in a time slot, decision D0 and D1
contents with a radiometer. support hypothesis H0 and H1 , γ is a detection threshold
In order to ensure the covertness of D1 content sharing, predetermined by the warden. In this paper, we take into
other D2D pairs Dm (m = 2, 3...N ) without covert demand account an infinite length of channel use [10]. Thus, TW is
are of great importance. Specifically, when D1 wants to given by Eq. (4) when l → ∞.
transmit covert contents, Dm transmits contents at the same There are two kinds of detection errors for warden. One is
time. Because all D2D pairs reuse the same resource block in the false alarm (FA) when the warden makes decision D1 while
D2D dedicated mode, Dm will cause co-channel interference H0 is true. The other is the miss detection (MD) when the war-
to D1 , and this kind of co-channel interference confuses the den makes decision D0 while H1 is true. Their probabilities
warden. are defined as PF A = P(D1 | H0 ) and PM D = P(D0 | H1 ).
All the channels are independent quasi-static Rayleigh fad- If hypotheses H0 and H1 are equal priori probabilities, the
ing, where the channel coefficients are constant in a time slot detection error rate of warden will be given as
with l channel uses while it changes with time slot [8]. We
define transmission power of D1 and other D2D pairs Dm as ξ = PF A + PM D . (5)
pD1 and pDm , and it is important to note that pD1 is fixed.
Remark 1: When D1 wants to send covert contents, too
On the contrary, pDm changes with the time slot and[ follows]
many Dm will affect the transmission rate of D1 and waste
a continuous uniform distribution over the interval 0, pmax Dm
energy. Therefore, we mainly study the case of N = 2 and
with the probability density function (PDF) given as
{ N = 3, and explore the effect of increasing the number of D2D
1
pmax , 0 ≤ x ≤ pmax
Dm ,
pair in dedicated mode on covert communication performance.
fpDm (x) = Dm (1) Finally, the conclusion we have got can be extended to the case
0, otherwise.
of N.

B. Detection Methods of Warden


III. C OVERT P ERFORMANCE W HEN N = 2
In one time slot, the warden must decide whether D1 trans-
mits covert contents between the transmitter and the receiver. A. Probability of False Alarm and Miss Detection
As a result, the warden faces a binary hypothesis testing When N = 2, D1 transmits covert contents and D2 makes
problem in which H0 indicates that D1 does not transmit co-channel interference to confuse the warden.
covert contents, while alternative hypothesis H1 indicates that
Lemma 1. Probabilities of false alarm and miss detection
D1 transmits. Under these assumptions, the signal received by
are calculated as
the warden is given as Eq. (2), where hDT1 ,W and hDTm ,W 
donate the channels from DT1 and DTm to warden, the 
 1, γ < σW 2 ,
γ−σW 2
distance from DT1 and DTm to warden are donated as dDT1 ,W PF A = 1 − pmax , σW 2 ≤ γ ≤ φ1 ,
D2 |hDT2 ,W | dDT2 ,W
2 −α
and dDTm ,W , α donates the path-loss exponent, xD1 and xDm 

are the covert contents transmitted by DT1 and DTm , nW is 0, γ > φ1 ,
 (6)
the additive white gaussian noise (AWGN) at warden and its

 0, γ < φ2 ,
variance is σW 2 . The warden aims to minimize his detection γ−φ2
PM D = , φ2 ≤ γ ≤ φ3 , (7)
| 2 ,W | dDT2 ,W
2 −α
error by applying the Likelihood Ratio test and the optimal  pmax h
 D2 DT

decision rule for him with Neyman-Pearson criterion is given 1, γ > φ3 ,

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on June 02,2020 at 10:53:55 UTC from IEEE Xplore. Restrictions apply.
dα dα dα dα α
2 dDT2 ,W
ξ(t) ∗ =
DT1 ,W DT2 ,W DT1 ,W DT2 ,W
+ t( − ) + t ln t − t , (14)

DT2 ,W dα
DT1 ,W dα
DT2 ,W dα
DT1 ,W dα
DT1 ,W



 1, γ < σW 2 ,

 2

 1−
(γ−σW 2 )
, σW 2 ≤ γ < ψ1
+ σW 2 ,
|hDT2 ,W | dDT ,W pD |hDT |
2 −α 2 −α 2
2pmax max
3 ,W dDT3 ,W
PF A = D 2 2 3
(18)

2
(γ−σW 2 − ψ1 ) ψ1

 , + σW 2 ≤ γ < ψ1 + σW 2 ,

 2pmax |hDT2 ,W |
2 −α
dDT ,W pmax |hDT |
2 −α
dDT3 ,W 2
 D 2 2 D 3 3 ,W
0, γ ≥ ψ1 + σW 2 ,



 0, γ < ψ2 + σW 2 ,

 2

 max
(γ−σW 2 − ψ2 )
, σW 2 + ψ2 ≤ γ < ψ1
+ ψ2 + σ W 2 ,
|hDT2 ,W | dDT ,W pD |hDT |
2 −α max 2 −α 2
2pD 3 ,W dDT3 ,W
PM D = 2 2 3 (19)

2
(γ−σW − ψ1 − ψ2 )
2
ψ1

 1− , + ψ2 + σ W 2 ≤ γ < ψ 1 + ψ2 + σ W 2 ,

 2pmax |
2 −α
|
hDT2 ,W dDT2 ,W pmax
2 −α
hDT3 ,W dDT3 ,W | | 2
 D2 D3
1, γ ≥ ψ1 + ψ2 + σW 2 ,

where the detection error rate is


−α

φ1 = pmax 2
D2 |hDT2 ,W | dDT2 ,W + σW ,
2

 1, γ < σW 2 ,

 2
2 −α
φ2 = pD1 |hDT1 ,W | dDT ,W + σW ,2 
 1 − pmax γ−σW 2 −α , σW 2 ≤ γ ≤ φ1 ,
2 −α
1
2 −α
 D2 | hDT2 ,W | dDT2 ,W
D2 |hDT2 ,W | dDT2 ,W + D1 |hDT1 ,W | dDT1 ,W
φ3 = pmax p ξ= 0, φ1 < γ < φ2 , (12)
+ σW 2 . 



γ−φ2
, φ2 ≤ γ ≤ φ3 ,

 pmax | h 2 ,W | dDT2 ,W
2 −α
 D 2
DT

Proof: According to Eq. (3), the probability of false alarm 1, γ > φ3 ,


and miss detection are shown as
the warden can easily set φ1 < γ < φ2 and get the minimum
= P[pD2 |hDT2 ,W | d−α value ξ ∗ = 0. When φ1 > φ2 , the detection error rate is given
2
PF A DT 2 ,W
+ σW 2 > γ]
γ − σW 2 (8) as
= P[pD2 > 2 −α
], 
|hDT2 ,W | dDT2 ,W 
 1, γ < σE 2 ,

 1− 2


γ−σ W
, σw 2 ≤ γ ≤ φ2 ,
2 −α 
 D2 |hDT2 ,W | dDT2 ,w
pmax 2 −α

PM D = P[pD2 |hDT2 ,W | dDT2 ,W + φ2 < γ]  pD1 |hDT ,W |2 d−α


γ − φ2 (9) ξ= 1 − pmax 1 DT1 ,W
, φ2 < γ < φ1 , (13)
= P[pD2 < ]. 
 |h |
2 −α
d
|hDT2 ,W | d−α
2 

D2
γ−φ2
DT 2 ,W DT2 ,W

 , φ1 ≤ γ ≤ φ3 ,
DT2 ,W

 D2 |hDT2 ,W | dDT2 ,W
pmax 2 −α

With the pdf of pDm in Eq. (1), we can obtain Eqs. (6) and 1, γ > φ3 ,
(7) by calculating Eqs. (8) and (9).
when γ < φ2 , ξ continuously decreases. When γ > φ1 , ξ
continuously increases. When φ2 < γ < φ1 , ξ is a fixed value.
B. Optimal Detection Threshold and Expected Detection Er- And because ξ is a continuous function about γ , the warden
ror Rate can easily set φ2 < γ < φ1 and get the minimum ξ ∗ =
pD1 |hDT ,W |2 d−α
Theorem 1. Utilizing a radiometer as its detector, the 1 − pmax 1 DT1 ,W

D2 |hDT2 ,W | dDT2 ,W
2 −α

optimal threshold for the warden is denoted as Because D1 does not know the channel from itself to
{ warden, we consider the expected detection error rate ξ ∗ as
γ∗ = [φ1 , φ2 ], φ1 < φ2 ,
(10) the measure of covertness from D1 perspective.
[φ2 , φ1 ], φ1 ≥ φ2 ,
Theorem 2. With the optimal threshold, the expected de-
and the minimum detection error rate at the warden is denoted tection error rate ξ ∗ is shown as Eq. (14), where t =
pD1 λ1
as pD1 λ1 + PDmax
λ2
.
 2

 0, φ1 < φ2 , Proof: The expected detection error rate can be obtained


ξ∗ = pD1 |hDT ,W |2 d−α (11) from Eq. (10)
 1 − pmax , φ1 ≥ φ2 .
1 DT 1 ,W

( D |hDT2 ,W | d−α
2

ξ ∗ = P[ φ1 < φ2 ] × 0 + P[ φ1 ≥ φ2 ] × E[ξ ∗ |φ1 ≥ φ2 ]


DT2 ,W )
2

Proof: Comparing φ1 , φ2 and φ3 , we find that φ3 is = P[ φ1 ≥ φ2 ] × E[ξ ∗ |φ1 ≥ φ2 ],


always the largest. When φ1 < φ2 , we can easily obtain that (15)

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on June 02,2020 at 10:53:55 UTC from IEEE Xplore. Restrictions apply.
−α 2 −α 2
PF A = P[pD2 |hDT2 ,W | dDT2 ,W + pD3 |hDT3 ,W | dDT3 ,W + σW > γ],
2
(20)

2 −α 2 −α 2 −α
PM D = P[pD1 |hDT1 ,W | dDT1 ,W + pD2 |hDT2 ,W | dDT2 ,W + pD3 |hDT3 ,W | dDT3 ,W + σW < γ].
2
(21)

{
0, ψ1 < ψ2 ,

ξ = (ψ1 − ψ2 )2 (23)
, ψ1 ≥ ψ2 ,
|hDT3 | pmax |hDT |
2 −α 2 −α
4 pmax
D ,W dDT ,W D 2 ,W dDT2 ,W
3 3 2



 1, γ < σW 2 ,

 2

 1 − max (γ−σW 2 )
, σW 2 ≤ γ < ψ1
+ σW 2 ,

 2pD |hDT2 ,W | d−α
2
DT2 ,W D3 |hDT3 ,W | dDT3 ,W
p max 2 −α 2

 2

 (γ−σW − ψ1 )
2 2
ψ1

 , + σW 2 ≤ γ < ψ1 + σW 2 ,
D2 |hDT2 ,W | dDT2 ,W D3 |hDT3 ,W | dDT3 ,W
 2pmax
2 −α 2 −α 2
pmax
ξ= 0, ψ1 + σ W 2 ≤ γ < ψ 2 + σ W 2 , (24)



2
(γ−σW 2 − ψ2 ) ψ1

 , ψ2 + σ W ≤ γ < 2 2
+ ψ2 + σ W ,

 2pD |hDT2 ,W | dDT ,W D |hDT3 ,W | d−α
max 2 −α
p max 2 2


2 2 3 DT3 ,W


2
(γ−σW 2 − ψ1 − ψ2 ) ψ1

 1 − max , + ψ2 + σW 2 ≤ γ < ψ1 + ψ2 + σW 2 ,

 2p |h | d−α
2
DT2 ,W D3 |hDT3 ,W | dDT3 ,W
pmax 2 −α 2
 D2 DT 2 ,W

1, γ ≥ ψ1 + ψ2 + σW 2 .

and then B. Optimal Detection Threshold and Expected Detection Er-


 α
 ror Rate
 2
D |hDT2 ,W | d
pmax 
2 DT1 ,W Theorem 3. With the radiometer, the warden can set an
P[ φ1 ≥ φ2 ] = P |hDT1 ,W | ≤
2

 pD1 dα
DT 2 ,W
 optimal threshold which is given by
{
dα pmax
DT1 D2 λ2 [ψ1 , ψ2 ], ψ1 < ψ2 ,
= , γ∗ = (22)
α max
dDT2 (pD2 λ2 + pD1 λ1 ) [ψ2 , ψ1 ], ψ1 ≥ ψ2 ,
(16) and the warden can minimize the detection error rate with the
optimal threshold given by Eq.(23).
|hDT1 ,W | φ
2
pD1 dα
E[ξ ∗ |φ1 ≥ φ2 ]=1 − pmax DT2 ,W
E[ 2 | 1 ≥
φ2 ] Proof: Comparing ψ1 , ψ21 and ψ2 , we find that the rela-
| hDT1 ,W |
α
D2 dDT1 ,W
α
pD1 dDT ,W
{ max
pD λ2 max
pD λ 2
} tionship of size between the three determines the expression
= 1 − pmax α
2
ln(1 + p
2
) − p +
2
pmax , of ξ, and the detection error rate is presented as the following
D2 dDT1 ,W D1 λ1 D1 λ 1 D2 λ 2
(17) cases.
2
where λi donates the mean value of |hDT i,W | . Finally, the Case 1: When ψ1 ≤ ψ 2 , we can easily obtain that the
proof is finished by substituting Eqs. (16) and (17) into Eq. detection error rate is Eq. (24).
(15). Case 2: When ψ21 < ψ2 < ψ1 , the detection error rate is
Eq. (25).
Case 3: When ψ21 ≥ ψ2 , the detection error rate is Eq. (26).
IV. C OVERT P ERFORMANCE W HEN N = 3
The functions in all three cases are first reduced and then
A. Probability of False Alarm and Miss Detection increased. When case 1, the warden can easily set ψ1 + σW 2 ≤
γ < ψ2 + σW 2 and get the minimum value ξ ∗ = 0. When case
When N = 3, D1 transmits covert contents while D2 and
2 and 3, the warden can set ψ2 + σW 2 ≤ γ < ψ1 + σW 2
D3 make co-channel interference to confuse the warden.
and ψ21 + σW 2 ≤ γ < ψ2 + ψ21 + σW 2 to get the same
Lemma 2. Probabilities of false alarm and miss de- 2
(γ−σW 2 − ψ1 ) + (γ−σW 2 − ψ2 )
2

tection are calculated as Eqs. (18) and (19), where minimum value .
| DT2 ,W | dDT ,W pD |hDT3 ,W | d−α
2 −α 2
2pmax h max
2 −α 2 −α D2 DT3 ,W

D2 |hDT2 ,W | dDT2 ,W + D3 |hDT3 ,W | dDT3 ,W ,


2 3
ψ1 = pmax pmax So we combine case 2 and 3 as ψ1 ≥ ψ2 . Next, we take the
ψ2 = pD1 |hDT1 ,W | d−α derivative of that, when γ = ψ1 +
2 ψ2
DT1 ,W 2 + σW 2 , the derivative
ψ1 + ψ2 2
Proof: According to Eq. (5), the probability of false alarm is 0. When γ < 2 + σW , the derivative is less than
ψ1 + ψ2
and miss detection are shown as Eqs.(20) and (21). With the 0. When γ > 2 + W 2 , the derivative is more than 0.
σ
joint probability density function of pD2 and pD3 , we can ∗
So we get ξ = pmax (ψ1 − ψ2 )2
4 D |hDT3 ,W | dDT ,W pmax
D2 |hDT2 ,W | dDT2 ,W
2 −α 2 −α
obtain Eqs. (18) and (19) by substituting Eqs. (20) and (21). 3 3
ψ1 + ψ2 2
when γ = 2 + σW .

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on June 02,2020 at 10:53:55 UTC from IEEE Xplore. Restrictions apply.

 1, γ < σW 2 ,



2
(γ−σW 2 )

 1 − max , σW 2 ≤ γ < ψ1
+ σW 2 ,

 2pD |hDT2 ,W | dDT ,W pD |hDT3 ,W | d−α
2 −α max 2 2


2 2 3 DT3 ,W

 (γ−σW 2 − ψ1 )
2
ψ1
+ σW 2 ≤ γ < ψ2 + σW 2 ,

 | |
2 −α
| hDT3 ,W | dDT3 ,W
2 −α , 2
 2p max p max
h d
 D2 DT 2 ,W DT2 ,W D3
2 2
(γ−σW − ψ1 ) + (γ−σW − ψ2 )
2 2
ξ= , ψ2 + σ W 2 ≤ γ < ψ 1 + σ W 2 , (25)
D2 |hDT2 ,W | dDT2 ,W D3 |hDT3 ,W | dDT3 ,W
2 −α 2 −α


2pmax pmax

 (γ−σW − ψ2 )
2 2

 , ψ1 + σ W 2 ≤ γ < ψ1
+ ψ2 + σ W 2 ,

 2pmax | DT2 ,W | dDT ,W pD
2 −α max
|hDT3 ,W | d−α
2 2


D 2
h 2 3 DT3 ,W

 (γ−σW 2 − ψ1 − ψ2 )
2
ψ1

 1 − max , + ψ2 + σW 2 ≤ γ < ψ1 + ψ2 + σW 2 ,
 2pD |hDT2 ,W | d−α | 3 ,W | dDT3 ,W
2 max 2 −α 2
p
 2 DT2 ,W D3 h DT

1, γ ≥ ψ1 + ψ2 + σW 2 .


 1, γ < σW 2 ,



2
(γ−σW 2 )

 1− , σW 2 ≤ γ < ψ2 + σW 2 ,

 2pmax |hDT2 ,W | dDT ,W pD
2 −α max
|hDT
2 −α
dDT3 ,W |


D 2 3 ,W
2 3

 1−
2 2
(γ−σW ) − (γ−σW − ψ1 ) 2 2
, ψ2 + σW 2 ≤ γ < ψ1
+ σW 2 ,

 |
2 −α
|
2 −α
| | 2


2pmax
D2 hDT2 ,W dDT2 ,W pmax
2
D3 hDT3 ,W dDT3 ,W
2
(γ−σW − ψ1 ) + (γ−σW − ψ2 )
2 2
ψ1
ξ= , + σW 2 ≤ γ < λ2 + ψ21 + σW 2 , (26)
| | | |
2 −α 2 −α


2pmax hDT2 ,W dDT2 ,W pmax hDT3 ,W dDT3 ,W 2

D2 D3

 −
2 2
(γ−σW − ψ1 − ψ2 ) − (γ−σW 2 − ψ1 )
2
, σW 2 + ψ2 + ψ21 ≤ γ < ψ1 + σW 2 ,

 1
|
2 −α
|
2 −α
| |


max
2pD max
hDT2 ,W dDT2 ,W pD3 hDT3 ,W dDT3 ,W


2
(γ−σW 2 − ψ1 − ψ2 )
2

 1− , ψ1 + σW 2 ≤ γ < ψ1 + ψ2 + σW 2 ,

 2pmax |
2 −α
p max
|
2 −α
| |
 D2 hDT2 ,W d DT2 ,W D3 hDT3 ,W dDT3 ,W
1, γ ≥ ψ1 + ψ2 + σW 2 .

∫ ∞ ∫ ∞ ∫ ψ1 (ψ1 − ψ2 )2
E[ξ ∗ |ψ1 ≥ ψ2 ] = 0 0 0 4 xypmax −α −α f|h | (x) f|hDT3 ,W | (y) f|hDT1 ,W | (z)dzdydx.
2 2 2 (29)
D3 dDT3 ,W pmax
D2 dDT2 ,W DT2 ,W


m2 ∑ √ ∑
Na Na
∗ m m m m
E[ξ |ψ1 ≥ ψ2 ] = wj 1− j
y 2 wi 1 − x2i G( xi + , yj + ), (30)
4 j=1 i=1
2 2 2 2

−α pmax −α
λ1 (pmax
Theorem 4. The expected detection error rate ξ ∗ is given 2 (pD1 d−α
2 −
D2 dDT2 ,W x+ D3 dDT3 ,W y)
pD d−α
DT1 ,W )
by Eq. (28) and (30) λ31
e ],
1 DT1 ,W and
Proof: The expected detection error rate can be got from yj = cos(π 2j−1
wj = wi = Nπa , Na ), x i = cos(π 2i−1
Na ). Na
Eq. (22) and m denote the parameter of the gauss-chebyshev formula.
ξ ∗ = P[ ψ1 < ψ2 ] × 0 + P[ ψ1 ≥ ψ2 ] × E[ξ ∗ |ψ1 ≥ ψ2 ]
= P[ ψ1 ≥ ψ2 ] × E[ξ ∗ |ψ1 ≥ ψ2 ], V. N UMERICAL R ESULTS
(27) In this section, numerical results are presented to analyze
the performance of covert communication and covert rate of
P[ ψ1 ≥ ψ2 ] D1 . In general, we set the distance from DT to warden as 5m,
pD |hDT ,W |2 pmax |hDT |
2
pmax |hDT |
2 the path-loss exponent α is 4, the transmission power of DT1
= P{ 1dα 1 < D 2

2 ,W
+ D3

3 ,W
} is 17 dBm, the maximum transmission power of DT2 and DT3
DT1 ,W DT2 ,W DT3 ,W
=1− (
λ1 pmax α
D2 dDT 1,W
1
)(
λ1 pmax α
D3 dDT 1,W
). is 20 dBm, noise power at warden is -174dBm, the mean value
2
1+ λ2 pD1 dα
DT 2,W
1+ λ3 pD1 dα
DT 3,W of |hDT i,W | , λi is 1, the parameter of the gauss-chebyshev
(28) formula, N a and m are 20 and 1500.
E[ξ ∗ |ψ1 ≥ ψ2 ] is given by Eq. (29). For this transcendant Fig. 2 and Fig. 3 show the probabilities of FA, MD and
integral, we first simplify the triple integral into a double total error probability ξ ∗ versus the detection threshold γ of
integral, and then make an approximation using the gauss- eavesdropper when N = 2 and N = 3, respectively. As
chebyshev formula [11]. Finally we have the Eq. (30), where expected, we can see that the the simulated curves precisely
pmax −α pmax −α 2
e−λ2 x e−λ3 y ( D2 dDT2 ,W x+ D3 dDT3 ,W y) fit with the theoretical ones, which confirms the correctness
G(x, y) = x y [ λ1

2 pD1 −α max −α max −α
dDT1 ,W (pD2 dDT2 ,W x+pD3 dDT 3,E y) 2 (pD1 d−α
2 of our Lemma 1 and Lemma 2. We also observe that there
DT1 ,W )
λ21
+ λ31
− is indeed an optimal value of γ that minimizes ξ and this

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on June 02,2020 at 10:53:55 UTC from IEEE Xplore. Restrictions apply.
1 1 1
FA, theoretical
0.9 0.9 MD, theoretical
, theoretical 0.9
0.8 0.8 FA, simulated

Expected detection error rate


FA, theoretical MD, simulated
0.7 0.7 , simulated N=3,p =0dBm
D1
MD, theoretical 0.8
, theoretical N=3,p = -10dBm
D1
0.6 0.6
Probability

FA, simulated

Probability
N=2,p =0dBm
D1
MD, simulated
0.5 0.5 0.7 N=2,p = -10dBm
, simulated D1

0.4 0.4
0.6
0.3 0.3

0.2 0.2 0.5

0.1 0.1
0.4
0 0 0 2 4 6 8 10 12 14 16 18 20
0 0.5 1 1.5 2 2.5 3 0 1 2 3 4 5 6
Detection Threshold 10 -4 Detection Threshold 10 -4 p max
m
[dBm]

Fig. 2: FA, MD and total error Fig. 3: FA, MD and total error Fig. 4: Expected detection error rate
probability when N = 2. probability when N = 3. when N = 2 and 3.

minimum value satisfies Eq. (11) and Eq. (23), which verifies [2] W. Wang, K. C. Teh, and K. H. Li, “Enhanced physical
the correctness of our Theorem 1 and Theorem 3. In addition layer security in d2d spectrum sharing networks,” IEEE
to this, comparing the two minimum value shows that N = 3 Wireless Commun., vol. 6, no. 1, pp. 106–109, Feb. 2016.
has higher minimum detection error rate of warden. Because [3] D. Wu, L. Zhou, Y. Cai, H. Chao, and Y. Qian, “Physical-
increasing the number of N is equivalent to increasing co- social-aware d2d content sharing networks: A provider-
channel interference, which makes Warden’s decision more demander matching game,” IEEE Trans. Veh. Technol.,
difficult. vol. 67, no. 8, pp. 7538–7549, Aug. 2018.
Fig. 4 shows the expected detection error rate versus the [4] J. Wang, W. Tang, Q. Zhu, X. Li, H. Rao, and S. Li,
transmission power of Dm for pD1 = 0dBm and −10dBm. “Covert communication with the help of relay and chan-
Firstly, we can observe that the expected detection error nel uncertainty,” IEEE Wireless Commun. Lett., vol. 8,
rate increases with the transmission power of Dm . When no. 1, pp. 317–320, Feb. 2019.
pmax
Dm increases sufficiently, the expected detection error rate [5] B. He, S. Yan, X. Zhou, and H. Jafarkhani, “Covert wire-
approaches 1 and the D2D content sharing is harder to be less communication with a poisson field of interferers,”
detected. Then, we also observe that the expected detection IEEE Trans. Wireless Commun., vol. 17, no. 9, pp. 6005–
error rate increases with the number of Dm , which shows 6017, Sep. 2018.
that the D2D content sharing is harder to be detected when [6] K. Shahzad, X. Zhou, S. Yan, J. Hu, F. Shu, and J. Li,
the number of Dm increases. Last but not least, when N “Achieving covert wireless communications using a full-
is fixed, the transmission power of D1 is smaller and the duplex receiver,” IEEE Trans. Wireless Commun., vol. 17,
less likely content sharing is to be detected, since decreasing no. 12, pp. 8517–8530, Dec. 2018.
the transmission power of D1 means that the uncertainty of [7] X. Shi, D. Wu, C. Yue, C. Wan, and X. Guan, “Resource
interference caused by pDm is relatively greater. allocation for covert communication in d2d content shar-
ing: A matching game approach,” IEEE Access, vol. 7,
no. 2, pp. 72 835–72 849, May. 2019.
VI. C ONCLUSIONS [8] J. Hu, K. Shahzad, S. Yan, X. Zhou, F. Shu, and J. Li,
In this paper, we mainly study the security problem of D2D “Covert communications with a full-duplex receiver over
content sharing in dedicated mode with power uncertainty. wireless fading channels,” in Proc. IEEE Int. Conf.
Multiple D2D pairs reuse the same RB to produce the co- Commun. (ICC), May 2018, pp. 1–6.
channel interference to interfere with the warden, which makes [9] K. Shahzad, X. Zhou, and S. Yan, “Covert communi-
the covert content sharing difficult to be detected. In particular, cation in fading channels under channel uncertainty,” in
we study the case of two D2D pairs and three D2D pairs Proc. IEEE Veh. Technol. Conf. (VTC Spring), Jun. 2017,
reusing the same RB respectively and come to the conclusion pp. 1–5.
that when increasing the number and the transmission power [10] J. Hu, S. Yan, X. Zhou, F. Shu, and J. Wang, “Covert
of D2D pairs, D2D pair which needs covert content sharing communication in wireless relay networks,” in Proc.
has a lower probability to be detected and better guarantees IEEE Global Commun. Conf. (GLOBECOM), Dec. 2017,
D2D content sharing security in dedicated mode. pp. 1–6.
[11] Z. Ding, Z. Zhao, M. Peng, J. Hu, and H. Vincent,
“On the spectral efficiency and security enhancements of
R EFERENCES noma assisted multicast-unicast streaming,” IEEE Trans.
[1] D. Wu, L. Zhou, Y. Cai, and Y. Qian, “Collaborative Wireless Commun., vol. 65, no. 7, pp. 3151–3163, Jul.
caching and matching for d2d content sharing,” IEEE 2017.
Wireless Commun., vol. 25, no. 3, pp. 43–49, Jun. 2018.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on June 02,2020 at 10:53:55 UTC from IEEE Xplore. Restrictions apply.

You might also like