You are on page 1of 6

Improving Spectral Efficiency of D2D Cellular

Networks Through RF Energy Harvesting


Rachad Atat∗ , Lingjia Liu∗ , Jonathan Ashdown† , Michael Medley† , John Matyjas† , and Yang Yi∗
∗ Department of Electrical Engineering and Computer Science, University of Kansas, Lawrence, KS 66045, USA
† Information Directorate, Air Force Research Laboratory, Rome, NY 13441, USA

Abstract—In this paper, we consider device-to-device (D2D) conversion circuit to extract DC power from the received RF
communication underlaying cellular networks, where D2D users signals. We only consider energy harvesting for D2D users,
harvest radio frequency (RF) power from uplink cellular trans- especially with the increasing energy demands of D2D social
missions. This paper addresses two important issues for energy
harvesting-based D2D cellular networks. The first is how the networking applications, such as video streaming and media
energy harvested from cellular ambient RF signals affects the sharing. We assume that users cannot use the same channels
D2D spectral efficiency. Using tools from stochastic geometry, for harvesting ambient RF energy and transmitting information
we investigate this issue by first characterizing the transmission simultaneously, a valid assumption for small low-complex de-
probability of D2D transmitter as the probability of having vices1 . Based on this assumption, as we limit the proportion of
enough battery power, and then obtaining analytic expressions
of the spectral efficiency for D2D and cellular networks. The the spectrum available to D2D users, the D2D spectrum access
second issue is related to the cellular spectral efficiency: the more probability decreases; however the average D2D transmission
RF energy D2D users can harvest, the higher their transmission probability increases due to being able to harvest more RF
probability becomes, which leads to generating more interference power from ambient cellular signals. On the other hand, if
in the cellular network. We carry out simulations to understand D2D users are active for longer time, they generate more
this impact of RF energy harvesting on the spectral efficiency
of both D2D and cellular networks. Results show promising interference in cellular network, which negatively impacts the
improvement to the whole network, in terms of the weighted cellular spectral efficiency. The objective of this paper is to
spectral efficiency, when employing RF harvesting technology investigate the impact of energy harvesting on the spectral
and when there are enough available channels in the network. efficiency of D2D users and the underlying cellular network.
The major contributions of our paper can be summarized as
I. I NTRODUCTION the following:

E Nergy harvesting will be one of the important technolo-


gies that will help increase energy efficiency as the world
prepares to move towards 5G. More specific, energy harvesting
∙ First, to the best of our knowledge, this is the first work to
analyze the impact of RF energy harvesting on the spec-
tral efficiency of D2D and cellular networks. Using tools
from ambient RF signals will be among the forefront technolo- from stochastic geometry, we obtain analytical expression
gies to power small low-power devices, especially with future of the expected RF energy harvesting rate by making
advancements in RF circuitry. As a matter of fact, 3.5 mW full use of spatial locations of cellular transmitters. Then,
harvested power was achieved at a distance of 0.6 meters and we obtain the average energy utilization rate by defining
1 uW at a distance of 11 meters using Powercast RF energy- a transmission region for D2D transmitter. Finally, we
harvester operating at 915 MHz [1]. characterize the average transmission probability of D2D
Device-to-Device (D2D) communications [2], [3], [4] have transmitter in terms of the expected energy harvesting
recently grown in popularity due to their powerful benefits, rate and the average energy utilization rate.
many of which have been validated in scientific literature. D2D ∙ Second, we analytically characterize the spectral effi-
communication allows users in close proximity to communi- ciency of both D2D and cellular networks.
cate directly with each other bypassing the base station (BS). ∙ Third, our simulation results provide insights on the trade-
This helps offload part of the organic cellular traffic to D2D off between increasing D2D transmission probability by
networks, thereby improving spectral utilization, increasing harvesting more RF energy and the spectral efficiency of
capacity, decreasing delay, improving cellular coverage, and cellular networks.
increasing energy efficiency. The remainder of this paper is organized as follows. In
In this paper, we consider a D2D communication in under- Section II, a system model for D2D network underlaid with
lay cellular networks, where D2D users reuse the spectrum cellular network is presented. Section III introduces the RF
occupied by cellular users. We define a spectrum partition energy harvesting technology and model. Section IV presents
factor, which measures the fraction of spectrum available to
D2D users. Each D2D user will be equipped with an RF power 1 Note that we assume that D2D users cannot perform RF harvesting on a
channel while using another channel for data transmission at the same time,
Approved for Public Release; Distribution Unlimited: 88ABW-2016-4210. as this incurs technological limitations.

978-1-5090-1328-9/16/$31.00 ©2016 IEEE


the spectral efficiency analysis of both D2D and cellular ∙ Cellular user: The UEs in cellular mode form a thinning
networks. Section V presents the simulations and analytical PPP Φ𝐶 from Φ𝑈 , with intensity 𝜆𝐶 = 𝒫 (𝛿𝑖 = 𝛿𝐶 ) 𝜆𝑈 .
results. Finally, conclusions are drawn in Section VI. Note that Φ𝑈 = Φ𝐶 ∪ Φ𝐷 , and Φ𝐷 ∩ Φ𝐶 = ∅.
We assume that each D2D receiver is uniformly distributed
II. S YSTEM M ODEL with probability density function (PDF):
2𝑟
𝑓D (𝑟) = , 0 ≤ 𝑟 ≤ 𝜇, (1)
𝜇2
where 𝜇 is the maximum allowable distance between a D2D
pair defined in Eq. (9). In this paper, we assume that all users
use constant transmit powers, and we leave channel inversion
power control for future research. Denoting by ∣𝐵∣2 the
number of channels, D2D transmitters can access 𝜅∣𝐵∣ of them
randomly, where 𝜅 ∈ [0, 1] measures the fraction of spectrum
available to D2D users [5]. The D2D transmit power 𝑃𝐷 is
split among the 𝜅∣𝐵∣ subchannels as 𝑃ˆ𝐷 = (1/(𝜅∣𝐵∣))𝑃𝐷 .
Let us denote the channel SINR between nodes 𝑖 and 𝑗 by
𝛾𝑖,𝑗 . Accordingly, the probability that a source node 𝑠 is
Fig. 1: An example of a hybrid network with D2D and cellular in D2D mode is given by: 𝒫(𝛿𝑠 = 𝛿𝐷 ) = 𝒫 (𝛾𝑠,𝑑 ≥ 𝜃𝐷 ),
links with RF energy harvesting. with 𝛾𝑠,𝑑 denoting the channel SINR between source node
𝑠 and destination node 𝑑. If a source node 𝑠 cannot be
in D2D mode, then it becomes a cellular node, such that
Fig. 1 shows an example of a hybrid network that consists 𝒫 (𝛿𝑠 = 𝛿𝐶 ) = 1 − 𝒫 (𝛿𝑠 = 𝛿𝐷 ).
of D2D and cellular user equipments (UEs) with RF energy The distance between any two nodes 𝑖 and 𝑗 is denoted by
harvesting. We consider an uplink cellular network, where ∥𝑖−𝑗∥. We assume that the power of the signal transmitted by
D2D users and cellular users (CU) share the licensed spectrum. UEs decays at a rate of 𝑙(𝑖, 𝑗) = ∥𝑖 − 𝑗∥−𝛼 , where 𝛼 > 2 is
By using an uplink channel, the D2D transmitters experience the path-loss exponent of both cellular and D2D transmitters.
less interference when compared to using the downlink, due Rayleigh fading with mean one is used to model the small-
to the lower transmit power of cellular users when compared scale fading over each channel, with ℎ𝑖𝑗 denoting the channel
to that of the BS. Considering the in-band underlay system coefficient between nodes 𝑖 and 𝑗.
model, D2D transmitters randomly access the spectrum that Let 𝑞 be the probability that a BS assigns any 𝜅∣𝐵∣ channels
may be occupied by cellular users. Different D2D deployment to serve its users. Then 𝑞 is given by [7]:
scenarios including overlay in-band D2D, underlay in-band
(1−𝜅)∣𝐵∣
D2D and out-of-band D2D have been thoroughly analyzed and ∑
compared in [5] in terms of optimal spectrum partitioning, 𝑞 =1− 𝒫(𝑁𝑢 = 𝑛), (2)
𝑛=0
spectral efficiency and rates of D2D and cellular users. It
was shown that the underlay model can achieve a higher where 𝑁𝑢 is the number of users associated to a BS, and
throughput because it utilizes the spectrum more efficiently 𝒫(𝑁𝑢 = 𝑛) is given in [Eq. (6) in [7]]. Then, Φ̃𝐶 is a
compared to the overlay model where D2D and cellular users point process representing the set of CUs not using 𝜅∣𝐵∣
subchannels with intensity 𝜆 ˜ 𝐶 = (1 − 𝑞)𝜆𝐶 . Note that Φ̃𝐶
access orthogonal in-band spectrum. The locations of the
macro BSs are modeled by a homogeneous Poisson Point is not a PPP due to the correlation among uplink CUs, where
Process (PPP), Φ𝐵 with intensity 𝜆𝐵 . Let 𝒜(𝑘, 𝑅𝐵 ) denotes each CU is assigned a unique channel by the BS. However
the coverage region of a macrocell, approximated by a disk this dependency is weak and can be ignored as shown in [6]
−1/2 in order to provide a tractable analysis.
with radius 𝑅𝐵 = (𝜋𝜆𝐵 ) centered at a generic BS 𝑘.
UEs are uniformly distributed in the coverage region of the III. RF E NERGY H ARVESTING
corresponding BS and form a homogeneous PPP, Φ𝑈 with
A. Expected RF Energy Harvesting Rate
intensity 𝜆𝑈 [6].
Types of Nodes: We differentiate between two different Each D2D UE is equipped with an RF power conversion
types of nodes 𝛿𝑖 with 𝑖 = 𝐷, 𝐶, for D2D user and cellular circuit that extracts DC power from the received electromag-
user, respectively. netic waves [8]. Without loss of generality, assume the D2D
transmitter is located at the origin. It is important to note that
∙ D2D user: Here, a D2D user can either be a D2D we are only interested in cellular transmitters that are not
transmitter or a D2D receiver. The UEs in D2D mode using 𝜅∣𝐵∣ subchannels (i.e., Φ̃𝐶 ), since D2D users cannot
form a thinning PPP Φ𝐷 from Φ𝑈 , with intensity 𝜆𝐷 = use 𝜅∣𝐵∣ subchannels for harvesting energy and transmitting
𝒫 (𝛿𝑖 = 𝛿𝐷 ) 𝜆𝑈 . A UE 𝑖 is in D2D mode if the end- information simultaneously to simplify the system operation
to-end signal-to-interference-plus-noise (SINR) ratio be-
tween transmitter and receiver is greater than 𝜃𝐷 . 2 ∣.∣ denotes the set cardinality.
and analysis. The cellular uplink RF interference samples B. Energy Utilization Rate
received by D2D user inside the macrocell region 𝒜 and The energy utilization rate 𝜐 is defined as the number of
outside of it can be expressed as: energy units required per second by a D2D user [11].
∑ ∑
𝑦[𝑛] = 𝑠𝑖 [𝑛] + 𝑠𝑖 [𝑛] + 𝑧[𝑛], Theorem 2. The energy utilization rate can be expressed as:
˜ ∩𝒜 ˜ ∩𝒜𝑐 (3) ⎛ ( )2/𝛼 ( ⎞
𝑖∈𝜙𝐶 𝑖∈𝜙 𝐶
ˆ𝐷 )
Φ̃𝐶 ∩𝒜∕=∅ 𝑃 2
𝜐 = 𝜅𝜆𝐷 exp ⎝−𝜆𝐷 𝜋 Γ 1+ ⎠ . (8)
𝜖 𝛼
where 𝑛 = 0, 1, ..., 𝑁 − 1 is the sample index with 𝑁 being
the total number of samples; 𝑠𝑖 [𝑛] = (𝑃𝐶 ℎ𝑖0 𝑙(𝑖, 0)) [𝑛] is the
Proof. Without loss of generality, the transmission region
𝑛th sample of the received signal from cellular transmitter 𝑖
𝑅0 (𝜖) ⊂ R2 around a typical D2D transmitter is random and
by a typical D2D user; 𝑧[𝑛] is the Gaussian noise (𝑧[𝑛] ∼
defined as the range within which other nodes can receive its
𝒩 (0, 𝜎𝑛2 )); 𝜙˜𝐶 is a realization of Φ̃𝐶 denoting the set of
signal with a power above a specified decoding threshold 𝜖.
cellular transmitters’ locations; and 𝒜𝑐 is the complementary
This allows a receiver to satisfy a minimum SINR so that the
set of 𝒜.∑ The average received power can be expressed as
𝑁 −1 two nodes are connected [12]. We can define the transmission
𝜉 = 1/𝑁 𝑛=0 𝑦[𝑛]. When 𝑁 is large, by central limit the-
region as [7]:
orem, the distribution of 𝜉 approaches Gaussian distribution. { }
Thus, we can characterize the mean and variance of 𝜉 as [9]: 𝑅0 (𝜖) = 𝑥 ∈ R2 : 𝑃ˆ𝐷 𝑙(𝑥, 0)ℎ𝑥0 > 𝜖
∑ ⎧ ( )1/𝛼 ⎫
E(𝜉) = 𝑃𝐶 ℎ𝑖0 𝑙(𝑖, 0) = 𝐼RF , ⎨ ˆ𝐷 ℎ𝑥0
𝑃 ⎬ (9)
˜𝐶 ,∣𝒜∣>0
𝑖∈𝜙 = 𝑥 ∈ R2 : ∥𝑥∥ < 𝜇, 𝜇 = ,
(4) ⎩ 𝜖 ⎭
1 ( )
𝑉 𝑎𝑟(𝜉) = 2
𝐼RF + 𝜎𝑛2 ,
𝑁 where 𝜖 is the minimum power level required to be success-
where ∣𝒜∣ denotes the number of cellular transmitters inside fully heard (may or may not decode successfully). The area
the macrocell region. As can be seen in Eq. (4), the random of transmission region of a typical D2D transmitter is defined
variable 𝐼RF depends on the spatial distribution of cellular as [11]:

users. Because the amount of harvested RF power depends
∣𝑅0 (𝜖)∣ = Π𝑦∈Φ𝐷 (𝑃ˆ𝑑 𝑙(𝑦, 0)ℎ𝑦0 > 𝜖)𝑑𝑥, (10)
heavily on 𝐼RF , we can express the expected RF energy R2
harvesting rate 𝜂 as:
where the indicator function (𝐴) for event 𝐴 is equal to one
∫ ∞
if 𝐴 occurs or else is zero. The average transmission area can
𝜂 = 𝜆𝐷 𝑓𝐼RF (𝑥)𝒫 (𝜉 > 0∣𝐼RF ) 𝑑𝑥, (5) be expressed as:
0
E[∣𝑅𝑦 (𝜖)∣] = EE0Φ𝐷 [∣𝑅0 (𝜖)∣] = EEΦ𝐷 [∣𝑅0 (𝜖)∣]
( 𝑓𝐼RF (𝑥)
where ) is the PDF of 𝐼RF ; and 𝒫 (𝜉 > 0∣𝐼RF ) = ∫
𝑄 √−𝑁 𝑥 2 , where 𝑄-function is the tail probability of = Eℎ EΦ𝐷 Π𝑦∈Φ𝐷 (𝑃ˆ𝐷 ∥𝑥∥−𝛼 ℎ0 > 𝜖)𝑑𝑥
𝑥+𝜎𝑛 2
∫R
the standard normal distribution. Next, we will obtain an ˆ −𝛼

expression for 𝑓𝐼RF (𝑥) in order to characterize 𝜂. = Eℎ 𝑒−𝜆𝐷 (𝑃𝐷 ∥𝑥∥ ℎ0 )>𝜖 𝑑𝑥
2
∫R
( Laplace transform of 𝐼RF is given
The ) by [10]: ℒ𝐼RF (𝑠) =
exp −2𝜆 ˜ 𝐶 𝜋 2 𝑠2/𝛼 𝑃 2/𝛼 / (𝛼 sin(2𝜋/𝛼)) . For the special = Eℎ 𝑒−𝜆𝐷 (∥𝑥∥<𝜇) 𝑑𝑥
𝐶
( √ ) ( R 2
)
case of 𝛼 = 4, we have: ℒ𝐼RF (𝑠) = exp −𝜆 ˜ 𝐶 𝜋 2 𝑠𝑃𝐶 /2 . 2
= Eℎ 𝑒−𝜆𝐷 𝜋𝜇 = 𝑒−𝜆𝐷 𝜋Eℎ [𝜇 ]
2

The following lemma provides an expression for the PDF of ⎛ ( )2/𝛼 ( ⎞


ˆ )
𝐼RF for 𝛼 = 4 using the inverse Laplace transform. (𝑎) 𝑃𝐷 2 ⎠
= exp ⎝−𝜆𝐷 𝜋 Γ 1+ ,
𝜖 𝛼
Lemma 1.
√ ( ) (11)
˜𝐶
𝜋 3/2 𝑃𝐶 𝜆 ˜ 2 𝑃𝐶
𝜋4 𝜆
−3/2 𝐶
𝑓𝐼RF (𝑥) = 𝑥 exp − . (6) where (a) ]comes from the fact that: Eℎ [𝜇2 ] =
4 16𝑥 [ 2
ˆ 2
𝑃 𝐷 ℎ0 𝛼 𝑃ˆ𝐷 𝛼 2
𝑚
Eℎ ( 𝜖 ) = ( 𝜖 ) Eℎ (ℎ0 ), and Eℎ [ℎ𝑥 ] = Γ(1 + 𝑚).
𝛼

Theorem 1. The expected RF energy harvesting rate, when The energy utilization rate is expressed as:
the path loss exponent 𝛼 = 4, is expressed as: 𝜐 = 𝜅𝜆𝐷 E[∣𝑅𝑦 (𝜖)∣]. Substituting in Eq.(11) completes
√ ( )
˜ 𝐶 𝜆𝐷 ∫ ∞
the proof.
𝜋 3/2 𝑃𝐶 𝜆 −𝑁 𝑥 𝜋4 𝜆
˜2 𝑃
𝐶 𝐶
𝜂= 𝑄 √ 𝑥−3/2 𝑒− 16𝑥 𝑑𝑥. C. Transmission Probability of D2D User
4 0 𝑥 + 𝜎𝑛2

(7) Let 𝑆𝑡,𝑘 denotes the average battery level at time 𝑡 for
D2D user 𝑘. The dynamics of the average battery level can be
Proof : The result is obtained by substituting (6) into (5). captured as: 𝑆𝑡,𝑘 = min {𝑆𝑡−1,𝑘 + 𝜂 − 𝜐 (𝑆𝑡−1,𝑘 ≥ 𝜐) , 𝐺},
where 𝐺 is the battery capacity identical for all users. We Lemma 3. The Laplace transform of 𝐼tot,D can be character-
define the transmission probability 𝜌 of D2D user as [13]: ized as:
1∑
𝑛 ( )
𝜌 = lim E [ (𝑆𝑡,𝑘 ≥ 𝜐)] . (12)
2/𝛼
−𝜆𝐶 𝜋𝑃𝐶 − 𝜆 ˜ 𝐷 𝜋 𝑃ˆ 2/𝛼
𝑛→∞ 𝑛 ℒ𝐼tot,D (𝑠) = exp 𝐷
𝑠2/𝛼 . (18)
𝑡=1 sinc( 𝛼2 )
Assuming infinite batter capacity3 and using calculations
in [13], we can express the transmission probability of D2D The following theorem provides an expression for the
user as: ( 𝜂) average SIR probability at a typical D2D receiver.
𝜌 = min 1, (13)
𝜐 Theorem 3. In interference-limited network (𝜎 2 = 0), we
IV. S PECTRAL E FFICIENCY A NALYSIS have:
The aggregate interference at a typical BS comes from 2
1 − 𝑒−𝜇 (𝑎2 +𝑎3 )
cellular transmitters in other cells (𝐼C ) and D2D transmitters 𝒫 (𝛾𝑖,0 ≥ 𝜃𝐷 ) = , (19)
(𝐼D ). It can be expressed as [14], [15]: (𝑎2 + 𝑎3 ) 𝜇2
∑ ∑ (( ) (
𝐼BS = 𝑃𝐶 ℎ𝑘0 𝑙(𝑘, 0) + 𝑃ˆ𝐷 ℎ𝑘0 𝑙(𝑘, 0), (14) 2 )) 2 − 2
where 𝑎2 = 𝜋𝜆𝐶 𝜃𝐷𝛼
/ sinc( 𝛼2 ) 𝑃𝐶𝛼 𝑃ˆ𝐷 𝛼 , and 𝑎3 =
𝑘∈Φ𝐶 ∩𝒜𝐶 𝑘∈Φ̃𝐷 2
˜ 𝐷 𝜃 𝛼 )/(sinc( 2 )).
(𝜋 𝜆
where Φ̃𝐷 is a PPP representing the set of effective D2D 𝐷 𝛼
˜ 𝐷 = 𝜅𝜌𝜆𝐷 ,
transmitters that are transmitting with intensity 𝜆
with 𝜌 defined in (13). Proof. Since ℎ0 ∼ exp(1), then 𝒫 (ℎ0 ≥ 𝑥) = 𝑒−𝑥 :
( )
Lemma 2. The Laplace transform of 𝐼𝐵𝑆 can be character-
𝒫 (𝛾𝑖,0 ≥ 𝜃𝐷 ) = 𝒫 ℎ0 ≥ 𝜃𝐷 𝑙(𝑖, 0)−1 𝑃ˆ𝐷 −1
(𝐼tot,D + 𝜎 2 )
ized as: [ ( )]
( )
˜ 𝐷 𝜋 𝑃ˆ 2/𝛼
𝜆 = E exp −𝜃𝐷 𝑙(𝑖, 0)−1 𝑃ˆ𝐷 −1
(𝐼tot,D + 𝜎 2
ℒ𝐼BS (𝑠) = exp − 𝐷
𝑠2/𝛼 ⋅ [ ( )]
sinc( 𝛼2 ) (𝑎)
( ( )) = E𝑋 exp −𝑎1 𝑋 − 𝑎2 𝑋 2/𝛼 − 𝑎3 𝑋 2/𝛼
2𝜋𝜆𝐵 𝑃𝐶 𝑠 2 2 𝑠𝑃𝐶 ( )
exp − 𝛼−2 2 𝐹1 1, 1 − 𝛼 ; 2 − 𝛼 ; − 𝑅𝛼 , ∫ 𝜇 2
1 − 𝑒−𝜇 (𝑎2 +𝑎3 )
(𝛼 − 2)𝑅𝐵 𝐵 (𝑏) ( ) 2𝑟
= exp −(𝑎2 + 𝑎3 )𝑟2 2 𝑑𝑟 = ,
(15) 0 𝜇 (𝑎2 + 𝑎3 ) 𝜇2
Γ(𝑐) ∫ 1 𝑏−1
2 𝐹1 (𝑎, 𝑏; 𝑐, 𝑧) = Γ(𝑏)Γ(𝑐−𝑏) 0 𝑡 (1 − 𝑡)𝑐−𝑏−1 (1 − 𝑡𝑧)−𝑎 𝑑𝑡
is the hypergeometric function. where in (a), 𝑋 = ∥𝑖 − 𝑗∥𝛼 , 𝑎1 = 𝜎 2 𝜃𝐷 𝑃ˆ𝐷 −1
; and (b) comes
2
from the fact of ignoring the noise 𝜎 = 0 and using the PDF
Proof. First we find the Laplace transform of the aggregated of the D2D distance in Eq.(1).
interference from cellular transmitters outside the cell.
⎡ ⎤
∏ ∏ Theorem 4. The spatially averaged spectral efficiency of
ℒ𝐼C (𝑠) = EΦ𝐶 ,Φ𝐵 ,ℎ ⎣ 𝑒−𝑠𝑃𝐶 ℎ𝑥𝑖 0 𝑙(𝑥𝑖 ,0) (∣𝑥𝑖 ∥≥𝑅𝐵 ) ⎦ cellular transmitters, 𝑅𝐶 , and D2D transmitters, 𝑅𝐷 can be
𝑥𝑖 ∈𝜙𝐶 𝑦𝑖 ∈𝜙𝐵 characterized as:
( ∫ ∞( ) )
1 √ 7/2 [( )−5/2 ( )−5/2 ]
≈ exp −2𝜋𝜆𝐵 1− −𝛼
𝑟𝑑𝑟 343 7𝜆 7𝜆 7𝜆
𝑅𝐵 1 + 𝑠𝑃𝐶 𝑟 𝑅𝐶 = √ 𝐵 𝐵

𝐵
+ 𝜆𝐶 ⋅
( ( )) 2 2
2𝜋𝜆𝐵 𝑃𝐶 𝑠 2 2 𝑠𝑃𝐶 20 2𝜆𝐶
= exp − 𝛼−2 2 𝐹1 1, 1 − 𝛼 ; 2 − 𝛼 ; − 𝑅𝛼 , ∫ ∫
(𝛼 − 2)𝑅𝐵 2 2 𝛼 𝑡 ( ( ))
𝐵 2𝜋𝑟𝜆𝐵 𝑒−𝜋𝜆𝐵 𝑟 𝑒−𝜎𝑛 𝑟 (2 −1) ℒ𝐼𝐵𝑆 𝑟𝛼 2𝑡 − 1 𝑑𝑡𝑑𝑟,
(16) 𝑟>0 𝑡>0
For (the special case of(√𝛼 = 4, we have: ℒ𝐼𝐶 (𝑠) = (20)
−2 √ 2
))
exp −𝑅𝐵 𝑠𝑃𝐶 arctan 𝑠𝑃𝐶 /(𝑅𝐵 ) . The Laplace trans- ∫ 𝜇 ∫
form of the aggregate interference
( from D2D transmitters ) is 2𝑟 ∞ −𝜎𝑛2 𝑟𝛼 (2𝑡 −1) ( ( ))
𝑅 𝐷 = 𝜅𝜌 𝑒 ℒ𝐼tot,D 𝑟𝛼 2𝑡 − 1 𝑑𝑡𝑑𝑟,
ˆ 2/𝛼 2/𝛼
given by: ℒ𝐼𝐷 (𝑠) = exp −𝜅𝜌𝜆𝐷 𝜋 𝑃𝐷 𝑠 / sinc( 𝛼 ) . 2
0 𝜇
2
0
(21)
The aggregate interference at a typical D2D user is the inter-
ference from cellular transmitters and other D2D transmitters, Proof : See Appendix VI.
and is given by:
∑ ∑ We show in our simulations the weighted proportional-
𝐼tot,D = 𝑃𝐶 ℎ𝑘0 𝑙(𝑘, 0) + 𝑃ˆ𝐷 ℎ𝑘0 𝑙(𝑘, 0). (17) fair spectral efficiency, 𝑅𝑝 , which helps maintain a balance
𝑘∈Φ𝐶 𝑘∈Φ̃𝐷 ∖0 between efficiency and fairness among D2D and cellular
3 The limit of exceeding the battery capacity is negligible when the capacity
transmitters that are sharing the spectrum. It can be expressed
is much larger than the average stored energy. Thus, the infinite battery as [5]: 𝑅𝑝 = 𝑤𝐶 log 𝑅𝐶 + (1 − 𝑤𝐶 ) log 𝑅𝐷 , where 0 ≤ 𝑤𝐶 ≤
capacity assumption can be regarded as an approximation. 1.
V. S IMULATION R ESULTS AND A NALYSIS validated our theoretical results with simulations and showed
In this section, we present Monte Carlo simulation and that employing RF energy harvesting technology in a dense
analytical results to study the impact of RF energy harvesting cellular network can significantly increase the overall weighted
on the spectral efficiency of D2D users and cellular networks. proportional-fair spectral efficiency of the whole network
Unless otherwise stated, we set the following system param- when the number of available channels is sufficiently large.
eters: 𝑅𝐵 = 788 meters (which corresponds to an inter-BS
ˆ ACKNOWLEDGMENT OF S UPPORT AND D ISCLAIMER
distance of 1500 meters), 𝑁 = 5000, 𝑃𝐶( = 200 ) mW, 𝑃𝐷 = 2
2
mW, 𝜖 = −70 dBm, 𝛼 = 4, 𝜆𝐵 = 1/ 𝜋𝑅𝐵 , 𝜆𝑈 = 10𝜆𝐵 , R. Atat, and L. Liu acknowledge the U.S. Government’s
𝜃𝐷 = 10 dB, 𝜅 = 0.5, and ∣𝐵∣ = 10. support in the publication of this paper. This material is based
Fig. 2 shows the weighted proportional-fair spectral effi- upon work funded by U.S. Air Force Research Laboratory
ciency, 𝑅𝑝 , in terms of the cellular intensity, 𝜆𝐶 , for different (AFRL), under AFRL Grant No. FA8750-14-1-0077, and by
values of the total number of channels, ∣𝐵∣. This figure National Science Foundation under Grant ECCS-1509514.
suggests that as ∣𝐵∣ increases, the proportional-fair spectral Any opinions, findings and conclusions or recommendations
efficiency of the whole network increases. This is explained expressed in this material are those of the authors and do not
by the fact that more available channels means the intensity necessarily reflect the views of AFRL.
of cellular users that are not using 𝜅∣𝐵∣ subchannels, Φ̃𝐶 ,
A PPENDIX : P ROOF OF T HEOREM 4
increases. In this way, D2D users can harvest more RF power
from ambient cellular signals, thereby increasing their trans- The spectral efficiency of cellular users is defined by Propo-
mission probability. When ∣𝐵∣ = 5, we can see that 𝑅𝑝 starts sition 3 in [5] as:
decreasing after reaching its optimal value at 𝜆𝐶 ≈ 5.8𝜆𝐵 , [ ]
1
due to a reduction in Φ̃𝐶 and number of cellular transmitters. 𝑅𝐶 = E0Φ𝐵 ,Φ𝐶 E [log (1 + SINR)] , (22)
𝑁
On the other hand, it is interesting to note that this is not the
case for the case where ∣𝐵∣ ≥ 10. In fact, when ∣𝐵∣ ≥ 10 it where 𝑁 is a random variable representing the number of
is a win-win situation for both D2D users and cellular users cellular transmitters in a cell. Considering a cell area 𝑆, and
as compared to a network without energy harvesting (such as having Φ𝐵 independent from Φ𝐶 , we have:
in [5]). In those networks, 𝑅𝑝 first increases and then decreases [ ] [ [ ]]
1 1
as 𝜆𝐶 increases. This shows that RF energy harvesting can be E0Φ𝐵 ,Φ𝐶 = E𝑆 E0Φ𝐶 ∣𝑆
𝑁 𝑁 (𝑆)
beneficial to the whole network when ∣𝐵∣ is sufficiently large. [∞ ] [( )]
To explain the interesting results of Fig 2, we demonstrate ∑ (𝑆𝜆𝐶 ) 𝑛−1
𝑒 −𝑆𝜆 𝐶 1 − 𝑒−𝑆𝜆𝐶
= E𝑆 = E𝑆 .
in the following results how different system parameters can 𝑛=1
𝑛(𝑛 − 1) 𝑆𝜆𝐶
affect the spectral efficiency. For instance, Fig 3 shows the (23)
D2D average spectral efficiency in terms of 𝜆𝐶 for different
values of ∣𝐵∣. As 𝜆𝐶 increases, 𝑅𝐷 increases due to an To obtain a closed-form expression for the PDF of 𝑆, we use
increase in D2D transmission probability 𝜌 for ∣𝐵∣ ≥ 10 (see a sufficiently accurate approximation as defined in [16]:
√ ( )
Fig. 4), without negatively impacting 𝑅𝐷 from an increase in 343 7 7
cellular interference at D2D users. However, this is not true 𝑓𝑆 (𝑥) = (𝑥𝜆𝐵 )5/2 exp − 𝑥𝜆𝐵 𝜆𝐵 . (24)
15 2𝜋 2
when the number of channels is small and D2D users cannot
harvest enough power nor increase their spectrum access. Also Substituting (24) in (23) gives;
note that for ∣𝐵∣ = 10, 𝑅𝐷 becomes stable beyond 𝜆𝐶 ≈ 8𝜆𝐵 [ ] √ 7/2 [( )−5/2 ( )−5/2 ]
1 343 7𝜆𝐵 7𝜆𝐵 7𝜆𝐵
and does not improve anymore, unlike the case of ∣𝐵∣ = 40, 0
EΦ𝐵 ,Φ𝐶 = √ − + 𝜆𝐶 .
mainly due to limiting the amount of harvested RF power. 𝑁 20 2𝜆𝐶 2 2
Finally, Fig. 5 shows the impact of 𝜌 on 𝑅𝐶 . As the
To calculate E [log (1 + SINR)], we use Theorem 3 in [17]:
transmission probability, 𝜌, of D2D users increases, 𝑅𝐶 de-
creases. This is because more active D2D users means more E [log (1 + SINR)] =
∫ ∫
interference around cellular transmitters. As the spectrum −𝜋𝜆𝐵 𝑟 2 2 𝛼
(2𝑡 −1)
( ( ))
partition factor, 𝜅, increases, more D2D users are accessing 2𝜋𝑟𝜆𝐵 𝑒 𝑒−𝜎𝑛 𝑟 ℒ𝐼𝐵𝑆 𝑟𝛼 2𝑡 − 1 𝑑𝑡𝑑𝑟,
𝑟>0 𝑡>0
the spectrum thereby creating more interference in the cellular
−𝜋𝜆𝐵 𝑟 2
network. However, note that it is difficult to achieve 𝜌 = 1 as where 2𝜋𝑟𝜆𝐵 𝑒 is the PDF of the transceiver distance.
𝜅 increases, since D2D users will harvest less RF power from Similarly, we characterize the spectral efficiency of D2D.
(1 − 𝜅)∣𝐵∣ subchannels. Thus, this figure is just to show the
potential impact that 𝜌 can have on 𝑅𝐶 . R EFERENCES
[1] A. M. Zungeru, L. Ang, S. R. S. Prabaharan, and K. P. Seng, “Ra-
VI. C ONCLUSIONS dio frequency energy harvesting and management for wireless sensor
Using tools from stochastic geometry, we analytically char- networks,” CoRR, vol. abs/1208.4439, 2012.
[2] H. Chen, L. Liu, Y. Li, B. Yu, L. Shi, and Y. Yi, “Device-to-device
acterized the D2D transmission probability in order to study communications in cellular networks,” IEEE COMSOC MMTC E-Letter,
its impact on the cellular and D2D spectral efficiency. We vol. 9, no. 1, pp. 29–33, Jan. 2014.
Average D2D Spectral Efficiency (bits/sec/Hz)
Weighted proportional fair spectrum efficiency
|B|=5 analytical
0.6 |B|=40 10
|B|=5 simulation
|B|=10 9 |B|=10 analytical
0.5 |B|=5 |B|=10 simulation
8
|B|=40 analytical
0.4 7 |B|=40 simulation
6
0.3
5

0.2 4
3
0.1
2

3 4 5 6 7 8 9 10 3 4 5 6 7 8 9 10
Intensity of cellular users λC −6
x 10 Intensity of cellular users λC x 10
−6

Fig. 2: The weighted proportional-fairness spectral efficiency Fig. 3: The D2D average spectral efficiency in terms of 𝜆𝐶
for different values of ∣𝐵∣ (𝜆𝐷 = 20𝜆𝐵 ; 𝑤𝐶 = 0.65). for different values of ∣𝐵∣.

Average Cellular Spectral Efficiency (bits/sec/Hz)


0.8
|B|=5 analytical κ=0.2
0.7 |B|=5 simulation 1.45 κ=0.5
D2D transmission probability ρ

|B|=10 analytical κ=0.8


1.4
0.6 |B|=10 simulation
|B|=40 analytical 1.35
0.5 |B|=40 simulation
1.3
0.4
1.25
0.3
1.2
0.2
1.15
0.1 1.1

1.05
3 4 5 6 7 8 9 10 0.2 0.4 0.6 0.8 1
Intensity of cellular users λC x 10
−6 ρ

Fig. 4: The D2D average transmission probability 𝜌 in terms Fig. 5: The average cellular spectral efficiency, 𝑅𝐶 , in terms
of 𝜆𝐶 for different values of ∣𝐵∣. of the transmission probability, 𝜌, for different values of 𝜅.

[3] H. Chen, L. Liu, J. Matyjas, and M. Medley, “Optimal resource York, NY, USA: Cambridge University Press, 2012.
allocation for sensing based spectrum sharing D2D networks,” Elsevier [11] H. Dhillon, Y. Li, P. Nuggehalli, Z. Pi, and J. G. Andrews, “Fundamen-
J. of Computers & Electr. Eng., vol. 44, pp. 107–121, May 2015. tals of heterogeneous cellular networks with energy harvesting,” IEEE
[4] Q. Du, W. Li, L. Liu, P. Ren, Y. Wang, and L. Sun, “Dynamic RACH Trans on Wireless Comm., vol. 13, no. 5, pp. 2782–2797, May 2014.
partition for massive access of differentiated M2M services,” Sensors, [12] C. Capar, D. Goeckel, and D. Towsley, “Broadcast analysis for extended
vol. 16, no. 4, p. 455, 2016. cooperative wireless networks,” IEEE Trans on Information Theory,
[5] X. Lin, J. G. Andrews, and A. Ghosh, “Spectrum sharing for device- vol. 59, no. 9, pp. 5805–5810, Sept 2013.
to-device communication in cellular networks,” IEEE Trans on Wireless [13] K. Huang, “Spatial throughput of mobile ad hoc networks powered by
Commun., vol. 13, no. 12, pp. 6727–6740, Dec 2014. energy harvesting,” IEEE Trans on Information Theory, vol. 59, no. 11,
[6] T. Novlan, H. Dhillon, and J. Andrews, “Analytical modeling of uplink pp. 7597–7612, Nov 2013.
cellular networks,” IEEE Trans on Wireless Commun., vol. 12, no. 6, [14] H. Chen and L. Liu, “Resource allocation for sensing-based D2D
pp. 2669–2679, June 2013. networks,” in 2015 IEEE 49th Asilomar Conference on Signals, Systems
and Computers, Nov. 2015, pp. 1058–1062.
[7] A. H. Sakr and E. Hossain, “Cognitive and energy harvesting-based D2D
[15] H. Chen, L. Liu, T. Novlan, J. Matyjas, B. L. Ng, and J. Zhang, “Spatial
communication in cellular networks: Stochastic geometry modeling and
spectrum sensing based device-to-device (D2D) cellular networks,”
analysis,” IEEE Trans on Commun., vol. 63, no. 5, pp. 1867–1880, May
IEEE Trans on Wireless Comm., July 2016, accepted.
2015.
[16] J.-S. Ferenc and Z. Neda, “On the size distribution of poisson-voronoi
[8] T. Le, K. Mayaram, and T. Fiez, “Efficient far-field radio frequency cells,” Physica A-Statistical Mechanics And Its Applications, vol. 385,
energy harvesting for passively powered sensor networks,” IEEE Journal no. 2, pp. 518–526, 2007.
of Solid-State Circuits, vol. 43, no. 5, pp. 1287–1302, May 2008. [17] J. G. Andrews, F. Baccelli, and R. K. Ganti, “A tractable approach
[9] X. Kang, Y. C. Liang, H. Garg, and L. Zhang, “Sensing-based spec- to coverage and rate in cellular networks,” IEEE Trans on Commun.,
trum sharing in cognitive radio networks,” IEEE Trans. on Vehicular vol. 59, no. 11, pp. 3122–3134, November 2011.
Technology, vol. 58, no. 8, pp. 4649–4654, Oct 2009.
[10] M. Haenggi, Stochastic Geometry for Wireless Networks, 1st ed. New

You might also like