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.2017.2725259, IEEE
Communications Letters

IEEE COMMUNICATIONS LETTERS, VOL. X, NO. X, XXX XXXX 1

Interference Analysis in Wireless Power Transfer

Abstract— This letter studies the effect of co-channel interfer- and SWIPT-based wireless power. Analytical expressions are
ence (CCI) generated by wireless power transfer on information obtained. Numerical results show that the interference from
delivery for three widely used setups, i.e., power beacon (PB), hy- WPT indeed degrades the performance and that SWIPT has
brid access point (HAP), and simultaneous wireless information
and power transfer (SWIPT). It focuses on the comparison of CCI the largest amount of interference among the three considered
from these three schemes in terms of the average interference setups. For some conditions, PB scheme has the same per-
power, the ergodic capacity, and the bit error rate. Numerical formance as HAP one. The main contribution of this work
results show that the extra interference caused by wireless power lies in the fact that it provides useful and important guidance
transfer indeed degrades the system performance. Moreover, on the design of WPT in wireless systems by quantifying the
SWIPT has the largest degradation, while PB and HAP have
the same degradation when proper parameters are chosen. effect of CCI from WPT using analytical expressions and by
examining the WPT schemes under different conditions. This
Index Terms— Bit error rate, co-channel interference, ergodic contribution allows system designers to choose the best WPT
capacity, wireless power transfer.
scheme and the best parameters for the chosen scheme for
the concerned applications. Previous works on the CCI from
I. I NTRODUCTION information delivery do not provide such guidance.
Radio-frequency (RF) energy harvesting can use intentional
wireless power transfer (WPT) for enabling perpetual opera- II. S YSTEM M ODEL
tions of wireless networks [1]–[3]. There are three main forms We consider the downlink of a multi-user multi-cell (or
of WPT, namely, power beacon (PB) [1], hybrid access point multi-cluster) network, in which the desired user receives
(HAP) [2], and simultaneous wireless information and power information from its allocated base station or access point
transfer (SWIPT) [3]. When WPT is performed in the same (AP). If PB is adopted for WPT, the interference at the desired
frequency as information delivery, co-channel interference user can be expressed as
(CCI) occurs. √ √
Previous works have mainly focused on the effect of CCI ∑ J1
Pj ∑I
Qj
Ip = v hj xj αj + gj wj βj , (1)
from information delivery on WPT. For example, in [4], the
j=1
d j dvj
j=J1 +1
effect of CCI on wireless powered relaying systems was
studied, while [5] examined the effect of CCI on HAP systems. where I is the total number of undesired APs and PBs, J1
The authors in [6] derived the average system throughput of a is the number of undesired APs transmitting data, Pj is the
PB scheme which powered the source node. Also, the effect transmission power, dj is the distance, hj is the fading coeffi-
of CCI in a K-user multiple-input multiple-output interference cient, xj is the transmitted symbol of the j-th data-transmitting
channel was investigated in [7]. None of these works has AP, respectively, αj accounts for the asynchronization with
considered the effect of CCI from WPT on information data-transmitting APs, I − J1 is the number of undesired
delivery. Owing to the use of wireless power, the amount PBs transferring power, Qj is the transmission power, gj is
of CCI in wireless powered networks will inevitably increase the fading coefficient, wj is the transmitted symbol of the
compared with the conventional networks that rely on battery j-th power-transferring AP, respectively, βj accounts for the
power. Even worse, since the power receiver often has a lower asynchronization with power-transferring APs, and v is the
sensitivity than the information receiver, a larger transmission path loss exponent. The first term in (1) represents CCI from
power has to be used for WPT. Thus, more and stronger CCI conventional data transmission, while the second term in (1)
will be seen in wireless powered networks. represents CCI from wireless power transfer.
In this work, we propose a new analytical framework to In this work, we assume that the distances are uniformly
compare the performances of different WPT schemes in a distributed over a circular ring such that the probability density
wireless powered downlink network. The average interference function (PDF) of dj can be written as
power, ergodic capacity, and bit error rate (BER) of the 2x
wireless system are derived with CCI from the PB-, HAP- fd (x) = , rl ≤ x ≤ rh (2)
rh2 − rl2
Yunfei Chen is with the College of Computer and Information, Ho- where rh denotes the radius of the outer ring and rl stands for
hai University, Nanjing, China 210098. He is also with the School of the radius of the inner ring. The fading coefficients hj and gj
Engineering, Warwick University, Coventry, UK CV4 7AL (e-mail: Yun-
[email protected]). are assumed to be complex Gaussian random variables with
D. B. da Costa is with the Department of Computer Engineering, Federal zero-mean and variance 2α2 such that the channel follows a
University of Ceará, Sobral, CE, Brazil (email: [email protected]). Rayleigh model. In this paper, without any loss of generality,
H. Ding is with the State Key Laboratory of Integrated Service Networks,
Xidian University, Xi’an, China and with Xi’an Communication Institute, we consider a M -ary phase shift keying (PSK) modulation
Xi’an, China (email: [email protected]). such that |xj | = 1. Also, assume that |wj | = 1 and that the

1089-7798 (c) 2017 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.2017.2725259, IEEE
Communications Letters

IEEE COMMUNICATIONS LETTERS, VOL. X, NO. X, XXX XXXX 2

frame length is T . The interfering APs start their transmission where (7) uses the average power of PSK signals, (8) uses
randomly at τ , where τ is uniformly distributed between 0 the average power of Rayleigh fading, (9) uses the average
and T . Thus, their interfering power over the period of T is of a uniform distribution, and (10) uses the average of (2).
determined by Tτ , which is uniformly distributed between 0 Similarly, for HAP and SWIPT schemes, one has respectively
and 1 and represented by αj and βj . [I ]
2α 2 ∑ 1 ∑I
Similarly, if HAP is adopted, the interference at the desired E{|Ih |2 } = d Pi + Qi , (11)
user can be expressed as 3 i=1 i=I1 +1
√ √ [ I ]
∑I1
Pi ∑I
Qi 2α2 ∑ ∑I
Ih = v hi xi αi + gi wi βi , (3) E{|Is | } =
2
d Pi + Qi . (12)
i=1
di dvi 3
i=I1 +1 i=1 i=1
where all the symbols are similar to before, except that I1 From the above, PB and HAP have the same average
denotes now the number of APs that transmit data and I − I1 interference power when J1 = I1 . Also, SWIPT always has
stands for the number of APs that transfer power, for HAP. larger average interference power than HAP and PB.
Finally, considering a SWIPT scheme, the interference at
the desired user can be written as
√ B. Ergodic Capacity
∑I
Pi + Qi
Is = hi xi αi , (4) Let the instantaneous signal-to-interference ratio (SIR) for
i=1
dvi the PB scheme be
W |h0 |2
where Pi is used for information decoding and Qi is used γp = , (13)
|Ip |2
for energy harvesting. Herein, we adopt a power-splitting
receiver for SWIPT with an equivalent power splitting factor where W = Pdv0 |x0 |2 is the fixed part of the signal power and
0
of ρ = PiP+Qi
i
. The SWIPT scheme receives information and |Ip |2 is the interference power. We need the PDF of γp . This
energy simultaneously but to maintain the same quality of the can be derived by using the PDFs of |h0 |2 and |Ip |2 . It can be
information signal after splitting at the receiver, an aggregate shown that |h0 |2 is exponentially √ distributed. Also, Ip in (1)
power of Pi + Qi is required at the transmitter. can be rewritten as Ip = IpR + −1IpI , where
Using the interference expressions above, a unified expres- √ √
sion for the received signal at the desired user is given by ∑J1
Pj ∑ I
Qj
√ R
Ip = R
v xj αj hj +
R
v wj βj gj ,(14)
P0 j=1
d j d j
j=J1 +1
y= h0 x0 + z, (5) √ √
dv0 ∑ Pj
J 1 ∑I
Qj
IpI = v xj αj hIj + I
v wj βj gj , (15)
where z can be replaced by Ip , Ih and Is depending on dj dj
j=1 j=J1 +1
the scheme used, P0 denotes the transmission power, d0 √ √
with hj = hR j + −1hj and gj = gj + −1gj . In a Rayleigh
I R I
represents the distance, h0 is the fading coefficient and x0
R I R I
is the transmitted symbol of the desired AP. In this work, we fading channel, hj , hj , gj , gj are independent real Gaussian
assume an interference-dominant scenario such that noise is random variables with mean zero and variance α2 . Since IpR
ignored in the received signal. Also, the distance d0 between and IpI are linear sums of Gaussian random variables, they are
the desired user and its AP is assumed to be constant. Note also Gaussian with mean zero and common variance α2 up ,
∑J1 Pj ∑I Q
that we assume power transfer and information delivery in the where up = j=1 dv |xj |2 αj2 + j=J1 +1 dvj |wj |2 βj2 . Thus,
j j
same frequency. The PB could operate at a different frequency |Ip |2 is an exponential random variable, conditioned on up .
for power transfer, but this will invalidate the comparison. Thus, the conditional PDF of γp given up can be derived as
III. P ERFORMANCE A NALYSIS W up W u1p
A. Average Interference Power fγp |up (z) = = , (16)
(up z + W )2 (z + W u1p )2
For the PB scheme, the average interference power can be
derived from (1) as by using (13), the exponential PDFs of |h0 |2 and |Ip |2 , and [8,
  eq. (3.351.3)]. The capacity is given by C(γp ) = log2 (1+γp ),
2α 2 ∑ J1 ∑
I
which holds for Gaussian inputs in additive white Gaussian
E{|Ip |2 } = d Pj + Qj  , (6)
3 noise channels. However, for other inputs and channel condi-
j=1 j=J1 +1
tions, it is also useful and can be used to calculate the ergodic
by using capacity [9]. Thus, we consider it here. By averaging it over
E{|xj |2 } = E{|wj |2 } = 1 (7) the conditional PDF of γp in (16) and using [8, eq. (4.291.15)],
the conditional ergodic capacity can be derived as
E{|hj | } = E{|gj | } = 2α
2 2 2
(8)
1 W
1 W ln up − ln W W up ln up
E{|αj |2 } = E{|βj |2 } = (9) C(up ) = = . (17)
3
{ }  2(ln rh −ln rl ) ln 2 up − W ln 2 W u1p − 1
1  2 −r 2 , v=2
rh
d=E = 2−v
−r
l
2−v (10) Now, we need to find the PDF of up or u1p for deriving the
v
dj  2l 2 h ,
2(r )
v>2 unconditional ergodic capacity. Since a closed-form analytical
(rh −rl )(v−2)

1089-7798 (c) 2017 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.2017.2725259, IEEE
Communications Letters

IEEE COMMUNICATIONS LETTERS, VOL. X, NO. X, XXX XXXX 3

expression is hard to be attained, we resort to approximations. parameters of λ, µ, k, θ, p and q are calculated in different
After trying several distributions for up , we have found that the ways for different schemes.
inverse Gaussian distribution fits up the best for large values
of I and J. For u1p , the Gamma and Weibull distributions C. Bit Error Rate
provide excellent approximations. Using the inverse Gaussian
distribution, the PDF of up can be approximated by For illustration purposes, we use the binary PSK signal as
√ an example. However, other PSK schemes can be examined in
2
λ − λ(x−µ) a similar way. From [8, eq. (3.383.10)], the conditional BER
fup (x) ≈ 3
e 2µ2 x , (18) can be calculated as
2πx
∫ ∞ W u1p
where λ and µ are determined by moment-matching as 1 √
  P (up ) = erf c( z) dz
∑
J1 ∑ I 0 2 (z + W u1p )2
d µ3 c d2 √
µ=  Pj + Qj  , + µ2 = ( − ) ( )
3 j=1 λ 5 9 1 1 W uWp 1 W
j=J1 +1 = − e Γ , , (25)
 2 2 2 up 2 up
∑J1 ∑ I
d2 ∑
J1 ∑I
·[ Pj2 + Q2j ] + Pj + Qj (19)
. where Γ(·, ·) denotes the upper incomplete
∫ ∞ Gamma function
j=1
9 j=1 [8, eq. (8.350.2)] and erf c(x) = √2π x e−t dt is the com-
2
j=J1 +1 j=J1 +1

1
plementary error function. Thus, the unconditional BER are
Also, using the Gamma approximation, the PDF of up is derived as
√ ( )
tk−1 − t mW n λ µλ λ
f u1 G (t) ≈ e θ, (20) P1 = n+ 1 e K−n− 21 , (26)
p Γ(k)k θ µ 2 2π µ
where Γ(·) denotes the Gamma function [8, eq. (8.310.1)], k 1 θkW 3
P2 = − 2 F1 (1, k + 1, k + ; 1 − θW ),
stands for the shape parameter, and θ is the scale parameter of 2 2(k + 12 ) 2
n
( )
the Gamma distribution. On the other hand, using the Weibull m(qW ) p+n
approximation, the PDF of u1p can be approximated by P3 = Γ , (27)
2 p
( )p−1 which hold for inverse Gaussian, Gamma and Weibull approx-
p t
e−( q ) ,
t p
f u1 W (t) ≈ (21) imations, respectively, where 2 F1 (·, ·, ·; ·) denotes the hyper-
p q q
geometric function √[8, eq. (9.100)] and [8, eq. (3.471.9)], [8,
where p and q are the shape parameter and scale parame- eq. (3.478.1)], 1 − xex Γ( 12 , x) ≈ mxn have been used with
ter of the Weibull distribution, respectively. Using moment- m = 0.16 and n = −0.52 from curve-fitting.
matching, one has kθ = E{ u1p }, k 2 θ2 + kθ2 = E{ u12 },
p
qΓ(1 + p1 ) = E{ u1p }, q 2 Γ(1 + p2 ) = E{ u12 }. However, due IV. N UMERICAL R ESULTS AND D ISCUSSION
p
to the intricacy in calculating the moments of u1p , analytical In this section, numerical examples in terms of ergodic
expressions for k, θ, p and q cannot be obtained. In this capacity and BER are presented by using the expressions
work, we use simulation to find the moments of u1p . In the derived in the previous section, where the values of λ, µ, k, θ,
∑I1 Pi
above results, by replacing up with uh = i=1 dv |xi |2 αi2 + p and q are derived from moment-matching, the moments of
∑I ∑ I
i
up , uh and uh are calculated using their expressions, and the
Qi
i=I1 +1 dv |wi |2 βi2 for HAP, with us = i=1 Pid+Q v
i
|xi |2 αi2
i i moments of u1p , u1h and u1s are simulated using 100000 runs.
for SWIPT, and setting J1 = I, results for HAP, SWIPT
For the plots, we set Pi = P = 20 W , Qj = Q = 40 W ,
and no wireless power can also be obtained. Since these
rh = 100 m, rl = 1 m, and v = 3. Also, let γ̂ = dW be the
substitutions are trivial, they are not presented here. 3 IP

Using the PDF in (18), (20) or (21) and the approximation average SIR when there is no wireless power.
Fig. 1 shows the ergodic capacity versus γ̂ using different
1−t ≈ at , where a = 1.3 and b = −0.7 from curve-fitting,
ln t b

[8, eq. (3.471.9)], [8, eq. (3.381.4)] and [8, eq. (3.478.1)], the approximations for the PB-based WPT. The upper part em-
unconditional ergodic capacity can be derived as ploys I = 20 and J1 = 10, while the lower part uses I = 2000
√ ( ) and J1 = 1000. One sees from Fig. 1 that the Gamma and
2a λ µ b− 1 λ Weibull approximations are very close to the simulated results,
C1 = ( ) 2 Kb− 21 , (22)
ln 2 2πW W µ while the inverse Gaussian approximation only works when
a(θW )b+1 Γ(k + b + 1) the values of I and J1 are large. It can be shown that similar
C2 = , (23) conclusions hold for the HAP and SWIPT schemes, as well
Γ(k) ln 2
( ) as for the BER performance. Fig. 2 plots the ergodic capacity
a p+b+1
C3 = (qW )b+1 Γ , (24) versus γ̂ for different wireless powered networks, assuming
ln 2 p the Gamma approximation. One sees that the case without
which hold for the inverse Gaussian, Gamma and Weibull wireless power has the highest ergodic capacity, as expected.
approximations, respectively, and where K(·) is the modified However, this case has to be supported by conventional battery
Bessel function of the second kind [8, eq. (8.432.1)]. These power for operation. One also sees that wireless power in
expressions are applicable to all schemes, except that the general degrades the ergodic capacity performance due to

1089-7798 (c) 2017 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.2017.2725259, IEEE
Communications Letters

IEEE COMMUNICATIONS LETTERS, VOL. X, NO. X, XXX XXXX 4

8 100
Gamma No wireless power
Ergodic capacity 6 Weibull PB/HAP, I=10, I 1 =J 1 =8
Inverse Gaussian PB/HAP, I=10, I 1 =J 1 =2
4 Simulation PB/HAP, I=20, I 1 =J 1 =8
PS, I=10
2 10-1 PS, I=20

BER
-10 -8 -6 -4 -2 0 2 4 6 8 10
γ̂ (dB)

4
10-2
Ergodic capacity

Gamma
3 Weibull v=3
Inverse Gaussian
2 Simulation

1
10-3
0 -10 -8 -6 -4 -2 0 2 4 6 8 10
-10 -8 -6 -4 -2 0 2 4 6 8 10 γ̂ (dB)
γ̂ (dB)

Fig. 1. Ergodic capacity of PB versus γ̂ for different approximations. Fig. 3. BER versus γ̂ for different wireless powered networks.

7
No wireless power
PB/HAP, I=10, I 1 =J 1 =8
works. Note also that we only studied the impact of CCI
PB/HAP, I=10, I 1 =J 1 =2 on the desired user’s information delivery. From a system’s
PB/HAP, I=20, I =J =8
6
PS, I=10
1 1 point of view, if the WPT scheme degrades the desired user’s
PS, I=20 information delivery, it may benefit other users’ power transfer.
Ergodic capacity

v=3
For example, a large transmission power of PB will cause
4
large interference to the desired user’ information delivery but
3 at the same time will benefit other users by harvesting more
2 power. Thus, considering the whole system, there is a tradeoff
between information and energy at different users. This is very
1
similar to the interference management techniques, such as
0
-10 -8 -6 -4 -2 0 2 4 6 8 10
interference alignment, where one uses precoding to remove
γ̂ (dB) the CCI at one user while maximizing the signal power at the
Fig. 2. Ergodic capacity versus γ̂ for different wireless powered networks.
other user. Also, optimization problems can be formulated to
consider the whole system. For example, in PB, if one wants
to minimize the average interference power at the desired user
while maximizing the average ∑Iharvested power at other users,
2
increased interference. The performance degrades when I one can maximize a 2α3 d j=J1 +1 Qj − (1 − a)E{|Ip |2 },
increases, as more interfering APs are transmitting, but the where a is a weighting coefficient. The studies of interference
degradation decreases with I. For fixed I, the performance management techniques and system-wide optimizations are
degrades when I1 and J1 decrease, because power transfer beyond the limit of this letter but are interesting topics for
has a larger interference than information delivery and the future works.
number of power-transferring APs increases. Also, PB has
the same interference as HAP when J1 = I1 1 , as can also R EFERENCES
be seen from (1) and (3). SWIPT has the smallest ergodic
[1] K. Huang and V. K. N. Lau, “Enabling wireless power transfer in cellular
capacity due to the largest interference caused. Our results networks: architecture, modeling and deployment,” IEEE Trans. Wireless
give a quantitative analysis of this effect. Fig. 3 shows the Commun., vol. 13, no. 2, pp. 902-912, Feb. 2014.
BER. Similar observations can be made. [2] H. Ju, R. Zhang, “Throughput maximization in wireless powered com-
munication networks,” IEEE Trans. Wireless Commun., vol. 13, no. 1,
Note that the performance of a practical communication pp. 418-428, Jan. 2014.
system can be affected by many factors, including interfer- [3] L. R.Varshney, “Transporting information and energy simultaneously,”
ence. Even with interference, the system performance is not in Proc. IEEE Int. Symp. Inf. Theory, pp. 1612-1616, Jun. 2008.
[4] I. Krikidis, “Simultaneous information and energy transfer in large-scale
necessarily degraded, as one can use advanced interference networks with/without relaying,” IEEE Trans. Commun., vol. 62, no. 3,
management techniques to alleviate it. However, this work pp. 900 - 912, Mar. 2014.
focuses on the effect of CCI from WPT in a generic commu- [5] Y. Chen, D. B. da Costa, and H. Ding, “Effect of CCI on WPC with time-
division energy and information transmission,” IEEE Wireless Commun.
nication system to quantify possible degradation before other Lett., vol. 5, no. 2, pp. 168-171, Apr. 2016.
techniques are adopted. This can be used as a benchmark to [6] C. Zhong, X. Chen, Z. Zhang, and G. K. Karagiannidis, “Wireless-
evaluate these advanced techniques. The use of these tech- powered communications: Performance analysis and optimization,”
IEEE Trans. Commun., vol. 63, no. 2, pp. 5178-5190, Dec. 2015.
niques and the consideration of all parts of a communication [7] Z. Zong, H. Feng, F.R. Yu, N. Zhao, T. Yang, and B. Hu, “Optimal
system are beyond the scope of this letter and can be future transceiver design for SWIPT in K-user MIMO interference channels,
IEEE Trans. Wireless Commun., vol. 15, no. 1, pp. 430-445, Jan. 2016.
1 Note that our formulations also incorporate the case when J ̸= I , which
[8] I.S. Gradshteyn, I.M. Ryzhik, Table of Integrals, Series, and Products,
1 1 6th Ed. Academic Press: London, 2000.
depends on the specific network setup and AP/HAP association for the desired [9] M.K. Simon, M.-S. Alouini, Digital Communication over Fading Chan-
user. The equations for PB and HAP are very similar but we separate them nels, Wiley: London, 2dn Ed. 2005.
for clarity purpose.

1089-7798 (c) 2017 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.