Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LCOMM.2018.2808180, IEEE
Communications Letters

Coverage and Interference in D2D Networks with Poisson Cluster Process


Sandeep Joshi, Student Member, IEEE, and Ranjan K. Mallik, Fellow, IEEE

Abstract—In this paper, we present an analytical framework with parameter K, and no fading, respectively, b) we derive
for the coverage probability analysis in a device-to-device (D2D) Laplace transform of intra-cluster and inter-cluster interference
network with the location of devices modeled as a Poisson cluster powers. Using Laplace transform-based approach we analyze
process (PCP). We consider Nakagami-m fading between the D2D
communication links which provides a more realistic scenario the performance in terms of the signal-to-interference-plus-
for the performance analysis. We assume a standard singular noise ratio (SINR) coverage probability considering that the
path loss model and use stochastic geometry as a tool for the D2D transmitter, which is chosen at random, is uniformly
interference and coverage probability analysis. Furthermore, we distributed in the same cluster where the typical D2D receiver
derive a closed form approximate expression and a bound on the is located, c) we extend the coverage probability analysis to
coverage probability, and also include the interference-limited
case. Numerical results corroborate our analysis. include interference-limited D2D networks and provide the
signal-to-interference ratio (SIR) coverage probability analy-
Index Terms– Coverage probability, Poisson cluster process sis, and d) we also provide an approximate bound on the cov-
(PCP), stochastic geometry. erage probability. Our analysis demonstrates that the wireless
I. I NTRODUCTION channel, density of users, and the random device placement
Device-to-device (D2D) communication, described as direct affect the SINR and SIR distributions.
communication between devices in proximity, is envisaged
to be an important part of 5G [1], [2]. In most of the II. S YSTEM M ODEL
studies on the performance characterization of the wireless We consider that the location of devices are modeled as
D2D networks, the locations of the devices and base stations PCPs, with the parent point process modeled as a PPP denoted
(BSs) are modeled as homogeneous Poisson point processes by Φ having density λ, and the daughter point processes being
(PPPs) [3]–[5], which provides the advantage of tractability but independent and identically distributed (i.i.d.) as a symmetric
are insufficient in characterising the interference management normal distribution with probability density function (pdf) of
techniques with spatial correlations between the devices. The the device location X relative to the parent points given by
network deployment scenarios due to the densification of wire- 2
!
less networks have changed drastically, which is also evident 1 ||x||
fX (x) = exp − , (1)
from the third generation partnership project (3GPP) proposed 2 π σ2 2 σ2
models [6] which suggest that the PPP models can capture the
irregularity in locations of the devices by taking into account where x ∈ R2 , σ 2 is the variance, and ||·|| represents the
Euclidean norm. Due to homogeneous independent clustering
the clustering effect in the device locations. Thus, a more between parent and daughter processes, the statistics of only
realistic modeling of device locations in the D2D network is to one cluster, called the representative cluster, Φ0 with density
model their distribution as a Poisson cluster process (PCP) [7]– λ0 , needs to be specified. The set of devices in each cluster
[9], which incorporates spatial configurations of the devices. is denoted by y ∈ Φ where all the N devices in a cluster
In this work, we consider a D2D-enabled downlink cellular are capable of communication, i.e., transmission and reception
which are called as transmitting and receiving devices based on
network with the location of devices modeled as PCPs. As per their transmit and receive operation, respectively. The cluster
the best of the authors’ knowledge, we provide novel analytical centres y0 ∈ Φ correspond to the parent points and the devices
expressions for the coverage probability analysis in a dedicated are the daughter points. The simultaneously transmitting de-
D2D network considering the effect of thermal noise and vices in the cluster are assumed to be Poisson distributed with
interference power. D2D communication being a short range mean λ1 , conditioned on the total simultaneously transmitting
devices which are less than or equal to the total transmitting
communication it experiences different propagation conditions devices. The set of simultaneously transmitting devices inside
due to low antenna heights which can be well incorporated the cluster is denoted by N . We consider a typical device
by Nakagami-m fading channel model as compared to the (D2D receiver) which is located at the origin for the analysis,
conventional heterogeneous cellular network which experi- which is chosen from Φ0 and both the typical device and the
ences diffuse propagation environment with fading channel representative cluster are chosen randomly. The number of
interfering devices within the cluster is N − 1 and is Poisson
modeled as Rayleigh distributed. Specifically, our main con- distributed with mean λ1 − 1. The D2D transmitter is assumed
tributions are: a) using tools from stochastic geometry, we to be located at x0 inside the cluster and the standard singular
develop an analytical framework for the performance analysis path loss model is used i.e., f (r) = r−α , where α(> 2)
of a clustered dedicated D2D network considering a more is the path loss exponent. Furthermore, the fading channel
general Nakagami-m fading channel model which for m = 1, between the D2D transmitter and the typical device is assumed
to be Nakagami-m distributed; therefore, the fading power
m = (K + 1)2 /(2K + 1), and m → ∞ models Rayleigh has gamma distribution and the complementary cumulative
fading, provides a close parametric fit for Ricean distribution distribution function (CCDF) using the sum representation is
given by
Sandeep Joshi and Ranjan K. Mallik are with the Department of Elec-
trical Engineering, Indian Institute of Technology - Delhi, Hauz Khas, m−1
New Delhi 110016, India (e-mail: sandeep85joshi@gmail.com,
X (mr)k
F̄|h0 |2 (r) = exp(−mr) , r ≥ 0, (2)
rkmallik@ee.iitd.ernet.in). k=0
k!

1089-7798 (c) 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LCOMM.2018.2808180, IEEE
Communications Letters

where m is the fading parameter which is assumed to be a The expectation over interference is expressed in terms of the
positive integer. The received power is given as Laplace transform, using the fact that intra-cluster and the
2 −α inter-cluster interferences are independent, is given as
Pr = P |h0 | ||y0 + x0 || , (3)
 
where P denotes the transmit power, h0 denotes the complex LI (s) = E [exp(−sIcluster )] E exp(−sIN \cluster
channel fading coefficient between the D2D transmitter and = LIcluster (s) LIN \cluster (s) , (9)
the typical device. The typical device experiences interference
from D2D transmitters within the representative cluster, re- where the parameter s = mβP −1 rα . For simplification, we
ferred to as intra-cluster interference, and expressed as further assume that the intra-cluster distances are uncorrelated;
hence the distance r is not conditioned on the distance
P |hx |2 ||y0 + x||−α ,
X
Icluster = (4)
v0 . Therefore, we can express (8) in terms of the Laplace
x∈Φ\x0
transform as
while interference from D2D transmitters outside the represen- ∞ m−1 k (k−g) k
σn2 mβP −1 r α
Z
tative cluster, referred to as inter-cluster interference, is given PSINR =
XX
as 0 g! (k − g)!
P |hx |2 ||y + x||−α .
k=0 g=0
X X
IN \cluster = (5)
(g)
×exp −mβP −1 r α σn2 (−1)g LI (s)fR (r) dr ,

y ∈Φ\y0 x ∈N (10)
The total interference experienced by the typical user is denoted (g)
as I = Icluster + IN \cluster . We consider a dedicated D2D where LI (s) represents the gth derivative of the Laplace
network, wherein the frequency resources are orthogonally transform of the D2D interference.
allocated between D2D and cellular users; therefore, there is
no interference between D2D and cellular users. The SINR at A. Intra-Cluster Laplace Transform
the typical device is given by
The expectation over the intra-cluster interference as given
Pr by (4), is expressed in terms of the Laplace transform as
SINR = , (6)
I + σn2  
m
where σn2
denotes the noise power. In interference-limited (a)
LIcluster (s) = E 
Y m
(m + s P ||y0 + x||−α )m

networks the noise power is very small as compared to the in- x∈Φ\x0
terference power, and therefore, can be ignored; thus, the SINR N−1 j
given by (6) with σn2 → 0 reduces to the SIR. To characterize
(b) X Z mm
= f
−α )m X
(x) dx
the distance distributions, let us denote the distance between j=0 R2 (m + s P ||y0 + x||

(λ1 − 1)j exp (− (λ1 − 1)) j!



the typical device and the possible transmitting devices in the
cluster as {Ri }i=1:N , with ri = ||y0 +x|| being the realization × PN−1 
i
i=0 (λ1 − 1) exp (− (λ1 − 1)) i!
of Ri . As y0 and x are i.i.d. Gaussian random variables having (c)
 Z 
mm

variance σ 2 , (y0 + x) is a Gaussian random variable with = exp − 1−
R2 (m + s P ||y0 + x||−α )m
variance 2σ 2 ; thus, Ri is Rayleigh distributed with pdf denoted × (λ1 − 1) fX (x) dx]
by fR (r). Since the daughter points are i.i.d. around the cluster " m
!Z k
(d) m ∞
s P w−α /m
center, R, conditioned on the distance v0 = ||y0 ||, has pdf
X
= exp − (λ1 − 1)
given by the Rician distribution as in [9] and is denoted as k=1
k 0 (1 + s P w−α /m)m
fR (r|v0 ). Therefore, the distance distribution in the typical ×fW (w|v0 ) dw]
" m
!Z k
cluster and the distance from the intra-cluster interferers will (e) X m ∞
s P w−α /m
also follow the Rician distribution. Similarly, for the intra- = exp − (λ1 − 1)
k 0 (1 + s P w−α /m)m
k=1
cluster case, the distance between the interfering devices and
×fW (w)dw] , (11)
the typical device, denoted by w = ||y0 + x||, ∀x ∈ Φ\x0 ,
conditioned on v0 = ||y0 ||, follows the Rician distribution with where (a) follows from the fading power being gamma
pdf fW (w|v0 ). The inter-cluster interferer distance denoted by distributed, (b) follows from the fact that the locations of
u = ||y + x||, ∀x ∈ Φ, conditioned on the distance v = ||y||, active intra-cluster transmitters are independent, and taking
also has a Rician distribution with pdf fU (u|v). the expectation over Poisson distributed interfering devices, (c)
follows from the simplifying assumption that, in comparison
III. P ERFORMANCE A NALYSIS
to the size of the cluster, the number of devices which are
The coverage probability is the probability that the SINR active simultaneously in the cluster is very small, λ1 << N ,
of a typical device exceeds a predefined threshold β, is given (d) follows from the change of variable ||y0 + x|| to w with
as PSINR = Pr(SINR > β); using (6), it can be written as change of Cartesian to polar coordinates, and (e) follows from
PSINR = ER Pr |h0 |2 > βP −1 Rα (I + σn2 ) | R = r , (7) the assumption that the intra-cluster distances are uncorrelated
 
with fW (w) being the Rayleigh pdf.
where E[·] is the expectation operator. Using (2), the fact that
noise and interference power are independent, and applying 1) Lower Bound: The intra-cluster Laplace transform,
binomial expansion, we can express (7) as given by (11), is simplified by unconditioning separately and
applying Jensen’s inequality [9] to give the lower bound as
X X σn2 (k−g) mβP −1 r α k
"m−1 k 
PSINR = ER 2 !
g! (k − g)! (λ1 − 1)

k=0 g=0 Ps α
LIcluster (s) ≥ exp − A1 , (12)
×exp −mβP r σn E I exp(−mβP −1 r α I) .
−1 α 2 2 σ2 m
  g 
(8)

1089-7798 (c) 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LCOMM.2018.2808180, IEEE
Communications Letters

1

where the constant A1 is and for m > 1, the constant c = (Γ(m + 1)) m . We can also
m−1
!    express the coverage probability as
1 X m 2 2
A1 = B + u, m − +u , (13) "  #
EI γ m, s I + σn2

α u=0 u α α
PSINR = ER 1 − . (21)
Γ(a) Γ(b) Γ(m)
where B(a, b) = Γ(a+b) represents the Beta function.
Using (20) and further simplification, we can express the
B. Inter-Cluster Laplace Transform
second term in the right hand side of (21) as
The expectation over the inter-cluster interference as given m  
EI γ m, s I + σn2
 
by (5), can be expressed in terms of the Laplace transform
X m
(−1)k exp −c s kσn2

>
similar to the case of intra-cluster Laplace transform and is Γ(m) k
k=0
given by (14) at the top of the next page, where (a) follows ×LI (c s k) . (22)
from taking the expectation over the number of interfering
devices in the cluster, and (b) follows from the probability Using (22) in (21), we can express the approximate tight closed
form lower bound for the coverage probability as
generating functional (PGFL) of the PPP [7], conversion of

m X
! −v
Cartesian into polar coordinates, and from the assumption that X (−1)k+v+1 m P (2 σ)−α
PSINR ≥
λ1 << N . v! k m β σn2 c k
k=1 v=0
1) Lower Bound: The inter-cluster Laplace transform given
Γ 1 + α2v

by (14) can be simplified using exponential Taylor series ×
expansion, and the Rician distribution property [9] to give the 2
1+ α v . (23)
2
lower bound as 1 + A2 (c k) α
 2!
Ps α
2) Interference-Limited Case: The coverage probability is
LIN \cluster (s) ≥ exp −2πλ1 λ A1 . (15)
m given as PSIR = Pr(SIR > β) and is computed by proceeding
in a way similar to the SINR coverage analysis which gives
h Pk i
C. Coverage Probability: Approximation m−1 Γ z=1 ℓz
+1
X X 2
In this subsection, we derive a closed form approximate PSIR = (−1)k Pk !
z=1 ℓz
expression for the coverage probability. The gth derivative k=0 (ℓ1 ,...,ℓk ) 2
+1
(g) (1 + A2 )
LI (s) of the Laplace transform is obtained by evaluating 0≤ℓ1 ,...,ℓk ≤k
the derivatives of H(s) = ln LI (s), where LI (s) is obtained ℓ1 +2ℓ2 +···+kℓk =k
by using (12) and (15) in (9), and then applying Faa di Bruno’s k  ! ℓi
formula for the derivatives of a composite function as in [5]. Y 1 −A2 Γ α2 + 1
× , (24)
The ℓth derivative of H(s) is given by i! Γ α2 − i + 1

i=1
ℓi !

(λ1 −1)
 2
2 Ps
s−ℓ k tuples

−A1 2 σ2
+ 2πλ1 λ Γ α
+1 m
α where the composite summation is carried over Pall
(ℓ) k
H (s) = 2
 , (16) (ℓ1 , . . . , ℓk ) of integers in range [0, k] satisfying i=1 iℓi = k.
Γ α
−ℓ+1
and the gth derivative of LI (s) is expressed by applying Faa IV. N UMERICAL R ESULTS
di Bruno’s formula as
! ℓi In this section, we numerically evaluate the analytical
g
(g)
X Y 1 H (i) (s) expressions with location of devices as PCP and the cluster
LI (s) = g! LI (s) . (17)
ℓ i! i! centres distributed as PPP with the system parameter values
i=1
(ℓ1 ,...,ℓg )
as: P = 1 W, σn2 = −104 dBm, and λ = 150 cluster/km2 .
0≤ℓ1 ,...,ℓg ≤g
ℓ1 +2ℓ2 +···+gℓg =g
Fig. 1 shows the variation of the coverage probability, the
coverage bound, and the SIR coverage versus the pre-defined
Using (16) in (17) and substituting the result in (10) along threshold for Nakagami-m fading environment for different
with further simplification gives the approximate coverage values of m with α = 4. The values of m = 1 and m = 17
probability expression as given by (18) at the top of the next correspond to the Rayleigh fading and the Rician fading with
page, where the composite summation in (18) is carried over K = 15 dB, respectively. Fig. 2 shows the variation of the
Pgg tuples (ℓ1 , . . . , ℓg ) of integers in range [0, g] satisfying
all coverage probability versus the pre-defined threshold with
i=1 iℓi = g, and the parameter A2 is given as α = 3 and 4. It can be seen that the coverage probability
2 is affected by the values of α and m; the smaller value of α
A2 = A1 β α 8 π λ σ 2 λ1 + 2 (λ1 − 1) .

(19)
indicates stronger interference and hence low coverage while
1) Approximate Bound: A tractable bound for the coverage lower m values correspond to higher outage. Fig. 3 shows
probability which simplifies computation of expectation over the behavior of the coverage probability versus the average
the interference can be obtained using the Alzer’s inequality number of simultaneously active D2D transmitters for different
[10] which gives σ and β values with m = 2 and α = 4. Fig. 4 shows
γ(m, z) the behavior of the coverage probability versus the average
> (1 − exp(−cz))m , (20) number of simultaneously active D2D transmitters for different
Γ(m)
R∞ λ and β values. It can be seen from the plots that with increase
where γ(a, x) = x ta−1 exp(−t) dt, denotes the lower in the value of σ, the coverage probability decreases, because
incomplete gamma function, Γ(·) denotes the gamma function, with increase in value of σ the link distances increase which

1089-7798 (c) 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LCOMM.2018.2808180, IEEE
Communications Letters

 
N Z
λj1 exp (−λ1 ) j!
j 
(a) Y X fX (x) dx
LIN \cluster (s) = EΦ  m PN −1 i  
(1 + ||y + x||−α s P/m) R2 i=0 λ1 exp (−λ1 ) i!
y∈Φ\y0 j=0
 Z ∞   Z ∞  
(b) 1
= exp −2 π λ 1 − exp −λ1 1 − m fU (u/v) du v dv (14)
0 0 (1 + u−α s P/m)

hP i
∞ g g α(v+k−g)  ! ℓi
m−1
XX k X
(−1) v+g

P (2 σ) −α
g−k−v X Γ z=1 ℓz + 1 + Y 1 2 −A2 Γ α2 + 1
PSINR = (18)
m β σn2 i! Γ α2 − i + 1

(k − g)! α(v+k−g)
Pg
g=0 v=0
v! (
(1 + A2 ) z=1
ℓ z +1+ 2 ) ℓ
i=1 i
!
k=0 (ℓ1 ,...,ℓg )
0≤ℓ1 ,...,ℓg ≤g
ℓ1 +2ℓ2 +···+gℓg =g

1
= 2, = 4, = 10 = 150, = 4, m = 2
0.7 1
m = 1, SINR = 5, = 0 dB
m = 1, SINR bound = 5, = 10 dB
0.9
m = 2, SINR = 5, = 20 dB
0.6
m = 2, SINR bound = 10, = 0 dB
0.8
m = 5, SINR bound = 10, = 10 dB
0.5 m = 17, SINR bound = 10, = 20 dB
m = 2, SIR
0.7 Simulation
Coverage probability

Coverage probability
Simulation
0.6
0.4
0.5
0.3
0.4

0.2 0.3

0.2
0.1
0.1

0 0
0 5 10 15 20 25 30 35 40 1 2 3 4 5 6 7 8 9 10
SINR threshold (dB) Average number of simultaneously active transmitters
Fig. 1. Coverage probability versus the SINR threshold with λ = 150 Fig. 3. Coverage probability versus the average number of simultaneously
clusters/km 2 . active D2D transmitters for different values of σ.
1
= 2, = 10 = 20, = 4, m = 2
0.5 1
m = 1, = 4, SINR = 15, = 0 dB
m = 1, = 3, SINR 0.9 = 15, = 10 dB
0.45
m = 2, = 4, SINR
= 15, = 20 dB
m = 2, = 3, SINR 0.8
0.4 = 30, = 0 dB
m = 2, = 4, SIR
m = 2, = 3, SIR 0.7 = 30, = 10 dB
0.35
Coverage probability
Coverage probability

= 30, = 20 dB
0.3 0.6 Simulation

0.25 0.5

0.2 0.4

0.15 0.3

0.1 0.2

0.05 0.1

0 0
0 5 10 15 20 25 30 35 40 1 2 3 4 5 6 7 8 9 10
SINR threshold (dB) Average number of simultaneously active transmitters

Fig. 2. Coverage probability versus the SINR threshold for α = 3 and 4 Fig. 4. Coverage probability versus the average number of simultaneously
and λ = 150 clusters/km 2 . active D2D transmitters for different values of λ.

[4] M. Peng, Y. Li, T. Q. S. Quek, and C. Wang, “Device-to-device underlaid


results in the degradation of the link quality. Similarly, increase cellular networks under rician fading channels,” IEEE Trans. Wireless
in value of λ increases the interference which results in lower Commun., vol. 13, no. 8, pp. 4247–4259, Aug. 2014.
coverage. Also, with the increase in threshold the coverage [5] S. Joshi and R. K. Mallik, “Analysis of dedicated and shared device-
to-device communication in cellular networks over Nakagami-m fading
reduces as lower threshold value implies that the system has channels,” IET Commun., vol. 11, no. 10, pp. 1600–1609, Jul. 2017.
more number of simultaneously active transmitters, i.e., more [6] 3GPP TR 36.814, “Further advancements for E-UTRA physical layer
interference can be tolerated. It can also be observed from the aspects,” Tech. Rep., 2010.
[7] M. Haenggi, Stochastic Geometry for Wireless Networks, New York,
analytical expressions that the coverage is inversely related to NY, USA: Cambridge Univ. Press, 2013.
the average number of simultaneously active users. [8] R. K. Ganti and M. Haenggi, “Interference and outage in clustered
R EFERENCES wireless ad hoc networks,” IEEE Trans. Inf. Theory, vol. 55, no. 9,
pp. 4067–4086, Sep. 2009.
[1] J. G. Andrews, S. Buzzi, W. Choi, S. V. Hanly, A. Lozano, A. C. [9] M. Afshang, H. S. Dhillon, and P. H. J. Chong, “Modeling and
K. Soong, and J. C. Zhang, “What will 5G be?,” IEEE J. Sel. Areas performance analysis of clustered device-to-device networks,” IEEE
Commun., vol. 32, no. 6, pp. 1065–1082, Jun. 2014. Trans. Wireless Commun., vol. 15, no. 7, pp. 4957–4972, Jul. 2016.
[2] A. Asadi, Q. Wang, and V. Mancuso, “A survey on device-to-device [10] H. Alzer, “On some inequalities for the incomplete Gamma function,”
communication in cellular networks,” IEEE Commun. Surv. Tuts., vol. AMS Mathematics of Computation, vol. 66, no. 218, pp. 771–778, Apr.
16, no. 4, pp. 1801–1819, Fourth Quart. 2014. 1997.
[3] J. G. Andrews, F. Baccelli, and R. K. Ganti, “A tractable approach to
coverage and rate in cellular networks,” IEEE Trans. Commun., vol. 59,
no. 11, pp. 3122–3134, Nov. 2011.

1089-7798 (c) 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.

You might also like