You are on page 1of 6

On the Performance of Non-Orthogonal Multiple

Access Systems with Imperfect Successive

Interference Cancellation
Lina Bariah* , Arafat Al-Dweik*† and Sami Muhaidat* ‡
Department of Electrical and Computer Engineering, Khalifa University, Abu Dhabi, UAE.
Emails: {lina.bariah, dweik}
Western University, London, Canada. Email:
University of Surrey, Guildford, U.K. Email:

Abstract—Non-orthogonal multiple access (NOMA) technique the outage probability and the ergodic sum rates performance
has sparked a growing research interest due to its ability to in downlink NOMA systems with randomly deployed users.
enhance the overall spectral efficiency of wireless systems. In The derived analytical results in [3] show that the outage
this paper, we investigate the pairwise error probability (PEP)
performance of conventional NOMA systems, where an exact probability of NOMA systems highly depends on the targeted
closed form expression for the PEP is derived for different users, data rates and the allocated power for each user. Ding et al.
to give some insight about the reliability of the far and near users. [4] studied the effect of user pairing on the outage probability
Through the derivation of PEP expressions, we demonstrate that performance and the sum rate for two scenarios, fixed power
the maximum achievable diversity order is proportional to the allocation and cognitive-radio inspired NOMA. As reported in
user’s order. The obtained error probability expressions are used
to formulate an optimization problem that minimizes the overall [4], selecting users with distinctive channel gains can enhance
bit error rate (BER) under power and error rate threshold the achieved sum rate.
constrains. The derived analytical results, corroborated by Monte Dynamic power allocation for uplink and downlink NOMA
Carlo simulations, are presented to show the diversity order and systems is presented in [5] with guaranteed QoS for dif-
error rate performance of each individual user. ferent users. Unlike conventional techniques, such as fixed
Index terms— NOMA, pairwise error probability, reliabil- power allocation and cognitive-radio inspired NOMA, dy-
ity, diversity gain, optimization. namic power allocation provides more flexibility by allowing
tradeoffs between user fairness and overall system throughput.
Performance analysis of NOMA systems is evaluated in [6]
Non-orthogonal multiple access (NOMA) is a promising from users’ fairness standpoint. In particular, the authors
technique for the upcoming fifth generation (5G) wireless investigate the outage probability and the sum rate of different
communications, and it has attracted an increased research power allocation scenarios, where instantaneous and average
interests in recent years. Enhanced latency, spectral efficiency channel gains are considered.
and connectivity are the main factors that stimulated the emer- Although performance analysis of NOMA systems is well
gence of NOMA systems, in which multiple users are allowed investigated in the literature [7]–[12], most of the reported
to share the same time and frequency resources [1]. The key work concentrates on evaluating the system’s performance in
point of NOMA systems is to permit a constrained level of terms of outage probability, individual sum rate and average
interference from other users that allows the receiver to per- sum rate. To the best of the authors knowledge, none of the
form successive interference cancellation (SIC) for the other reported work addressed the error rate performance analysis
users’ signals before detecting its own signal. NOMA systems of NOMA systems. Emphasizing on this, studying the error
rely on exploiting the power domain multiplexing to control rate performance of different users while considering imperfect
interference and maintain user fairness, in a way that grants SIC is crucial, to have some insightful results about the QoS
the far users higher power coefficients and assign low power of each individual user. Accurate bit error rate (BER) analysis
coefficients to near users [2]. Although NOMA technique of NOMA systems is intractable due to the SIC process,
enhances users’ fairness, in comparison with the conventional however, pairwise error probability (PEP) can be analyzed.
systems such as orthogonal multiple access (OMA) systems, It is worth noting that PEP gives a valuable indicator for the
quality of service (QoS) of far users is relatively low, which is BER performance, since it is considered as an upper bound
considered as a performance limiting factor in many scenarios for the BER.
due to error propagation. Based on the aforementioned discussion, the main contri-
Extensive research efforts have been conducted to study the butions of this paper are summarized as follows:
performance of NOMA systems from different perspectives • In this work, the PEP performance analysis of conven-
and under different scenarios. In [3], the authors investigated tional NOMA systems with imperfect SIC is considered,

978-1-5386-4328-0/18/$31.00 ©2018 IEEE

where an exact closed form PEP expression is derived for where xl is the transmitted signal of
Pthe lth user and αl is the
each user individually. The derived PEP expressions are power allocation coefficient, where l=1 αl = 1. The received
verified by Monte Carlo simulations. signal at the lth user is,
• Building on the obtained PEP formulae, asymptotic PEP
rl = hl s + nl (2)
is derived to analyze the achieved effective diversity gain,
which represents the performance of the system at high where hl ∼ CN (0, 2σh2 ) is the channel frequency response
SNR regime. and nl is the additive white Gaussian noise (AWGN) with
• Using the derived asymptotic expression of the PEP, an zero mean and variance σn2 . Power allocation coefficients are
optimization problem is formulated and solved to obtain sorted in descending order, α1 > α2 > ... > αL , given that
the optimum power allocation coefficients that minimize |h1 |2 < |h2 |2 < ... < |hL |2 . The first user decodes only
the BER, under power and users’ individual error rate its signal x1 , while treating the signals of all other users as
constrains. interference. The rest of the users should employ SIC to be
The rest of the paper is organized as follows. Adopted system able to detect their signals. For the lth user, it should perform
and channel models are presented in Sec. II followed by exact SIC for the higher power users, i.e., U1 , · · · Ul−1 , and treat the
and asymptotic PEP analysis for each individual user in Sec. rest of users signals as interference, i.e., Ul+1 , · · · UL .
III. Power allocation coefficients optimization is addressed in
Sec. IV. Numerical and simulation results are presented in III. PAIRWISE E RROR P ROBABILITY A NALYSIS FOR
Sec. V and the paper is concluded in Sec. VI. NOMA S YSTEMS
Notation: (·)∗ and |.| denote the complex conjugate opera-
A. PEP Analysis for First User
tion and the absolute value, respectively. Re {.} represents the
real part of a complex number. x̂ represents a detected symbol Without loss of generality, we consider the first user as the
2 2 2
and ∆ denotes (x − x̂). farthest user, therefore, |h1 | < |h2 | < · · · < |hL | . The
received signal at the first user can be represented as follows,
L p
Recalling that the basic idea behind NOMA systems is to p X
utilize the broadcast nature of the wireless channels to allow r1 = h1 α1 P x1 + αl P xl + n1 (3)
multiple users to share the same time, frequency and code
PL √
domains while assigning different power levels for different where l=2 αl P xl represents the interference term from
users, to permit a specific level of interference from the other the other users. PEP is defined as the probability of detecting
users. In this work, downlink transmission NOMA system with the symbol x̂ while symbol x was transmitted [13], which can
L users is considered, where each user is equipped with single be evaluated for the first user as follows,
antenna, as depicted in Fig. 1. Users are classified based on
PEP (x1 , x̂1 ) =
their distance from the base station (BS), where the first user is  2 2 
p p
the farthest user from the BS, consequently, it has the weakest Pr r1 − α1 P h1 x̂1 ≤ r1 − α1 P h1 x1 , x̂1 6= x1 .

channel. On the other hand, the Lth user is the nearest with
the strongest channel. The channels between the BS and the L (4)
users are modeled as independent and identically distributed Using the cumulative distribution function (CDF) of a normal
(i.i.d) Rayleigh flat fading channels. It is worth mentioning that distribution, the conditional PEP for the first user can be
near users are assigned lower power coefficients than far users. represented as given in (5). In (5),
Given the total transmitted signal power is P , the transmitted Z ∞  2
1 u
signal from the BS is given by, Q(x) = √ exp − du (6)
2π x 2
is the Gaussian Q-function [14] and ∆1 = (x1 −x̂1 ). It is worth
noting that the derived PEP expressions are conditioned on
particular interference values, which depend on the transmitted
and detected symbols for each user.
To get the unconditional PEP, we average over the prob-
ability density function (PDF) of |h|. By noting that user
1 has always the weakest channel, and channel gains for
the rest of users are ordered in ascending order, i.e. |h1 | =
min(|h1 |, · · · , |hL |) and |hL | = max(|h1 |, · · · , |hL |), ordered
statistics should be considered when evaluating the PDF of
Fig. 1: Typical NOMA system with L users. |h1 |. Therefore, the PDF of the lth user is given by [15],
L! L−l
L p
X f(l) (x) = fX (x)FX (x)l−1 (1 − FX (x)) .
s= αl P xl (1) (l − 1)!(L − l)!
√ 2
n P √
α1 P |h1 | |∆1 | + 2 |h1 | Re ∆1 l=2 αl P x∗l
PEP (x1 , x̂1 | |h1 |) = Q  √ . (5)
2 |∆1 | σn

Considering that |h| is Rayleigh

 distributed, its PDF and CDF  where
x2 x2
are fX (x) = σx2 exp − 2σ and F (x) = 1 − exp − 2σ ,
" ( L
2 X 2 X
p p

respectively [16]. Therefore, using (7), the PDF of |h1 | , ω1 βl = αl P |∆l | + 2 Re ∆l αn P xn
is given by, ( )# (17)
2ω1 + Re ∆l αq P
fΩ (ω1 ) = 2 exp − 12 (8)
σh 2σh q=1

h i
2 √
where σh2 = E |hl | , l = 1, 2, · · · , L.
υ= 2σn |∆l | . (18)
Hence, the PEP averaged over the PDF of ω1 is
Z ∞

Γ ω1
 To evaluate the unconditional PEP, we average over the PDF
PEP (x1 , x̂1 ) = exp − 2 erfc √ dω1 of |hl | , ωl . Using the PDF of the ordered statistics provided
0 σh2 2σh 2ζ
(9) in (7) and considering that |h| is Rayleigh distributed, the PDF
where of |hl | is,
( L p
) L! ωl
p 2
∗ fΩ (ωl ) = exp − 2
Γ = α1 P |∆1 | + 2Re ∆1 αl P xl (10) (l − 1)!(L − l)! σh2 2σh
l=2 l−1 L−l
ω2 ω2
and 1 − exp − l2 exp − l2 .
√ 2σh 2σh
ζ= 2 |∆1 | σn . (11) (19)
1 x
In (9), we use the identity, Q(x) = 2 erfc( 2 ),
where erfc(x)

To calculate the
is the complementary error function. Solving the integral in n Pnunconditional
n k n−k
 PEP, we use binomial expansion
(a + x) = k=0 x a [18, Eq. 1.111] to represent
(9) gives [17],   k 2 l−1
! the term 1 − exp − 2σ2 . Accordingly, the PEP can be
1 Γσh evaluated using the following integral,
PEP (x1 , x̂1 ) = 1− p . (12)
2 2ζ 2 + Γ2 σh2 l−1  
L! X l−1 2(l−1)−j
which can be averaged over all the possible values of xl , l = PEP (xl , x̂l ) = 2 (−1)
σh (l − 1)!(L − l)! j=0 j
2, · · · , L, to consider all interference scenarios. Z ∞
[L − l + j − 1] ωl2
βl ωl
× ωl exp − Q dωl .
0 2σh2 υ
B. PEP Analysis for the lth User
For the lth user, it first decodes the signals with higher Solving the integral in (20) gives the closed form expression
power, i.e., U1 , · · · , Ul−1 , to perform SIC before detecting for the PEP for the lth user, as shown in (21).
its own signal. The output of the lth SIC receiver can be
represented as,
L l−1 p
C. Asymptotic Analysis
p X p X
r̃l = αl P hl xl + αn P hl xn + αk P hl ∆k +nl PEP represents an upper bound for the BER, and it gives a
n=l+1 k=1
(13) useful insight on the error rate performance when the closed
where ∆k = (xk − x̂k ). The PEP of the lth user can be form expression of the BER can not be found. PEP is used
evaluated as shown in (4), which after simplification can be also to study the achieved diversity, where the diversity gain
represented as shown in (14). Wenwould like to highlight is defined as the magnitude of the slope of the PEP when the
PL √ o signal-to-noise ratio (SNR) value goes to infinity [13],
that for the Lth user, the term Re ∆l n=l+1 αn P x∗n
equals to zero. Hence, the PEP of the Lth user is given in log PEP (xl , x̂l )
ds = lim − (22)
(15). Therefore, using the CDF of a normal Gaussian random γ̄→∞ log γ̄
variable, the conditional PEP of the lth user can be evaluated where γ̄ = E {γ} is the average transmit SNR. Capitalizing
as the following, on the PEP presented in (21), in this section we derive the
asymptotic expression for the PEP of the lth user, which will
|hl | βl
PEP (xl , x̂l | |hl |) = Q (16) be used to evaluate the asymptotic diversity order. In this work
" ( L
) ( l−1
√ √ X √ X √
PEP (xl , x̂l | |hl |) = Pr 2 αl P Re {hl ∆l n∗l } ≤ −h2l 2
αl P |∆l | + 2 αl P Re ∆l αn P x∗n + Re ∆l αk P ∆∗k .
n=l+1 k=1

( L−1
√ √ X√
PEP (xL , x̂L | |hL |) = Pr 2 αL P Re {hL ∆L n∗L } ≤ −h2L 2
αL P |∆L | + 2 αL P Re ∆L ∗
αk P ∆k . (15)

L! X l − 1 (−1) βl σ h
PEP (xl , x̂l ) = 2 1− p 2 2 . (21)
σh (l − 1)!(L − l)! j=0 j [L − l + j + 1] βl σh + [L − l + j + 1] υ 2

we will concentrate on the effective diversity gain, Hence,

log PEP (xl , x̂l ) l−1  
de = − . (23)
X l−1 j 1
log γ̄ PEP (xl , x̂l ) ≤ Al (−1) ×
j γ̄
As it is noticed, when γ̄ → ∞, the effective diversity order Z ∞ j+L−l+1
converges to the asymptotic diversity gain. The conditional γ
1− exp − 2 dγ.
PEP presented in Eqn. (16) can be bounded by the following, 0 γ̄ 4 |∆l |
! (28)
PEP (xl , x̂l | |hl |) ≤ exp − 2 (24) Solving the integral in (28) and after some simplifications, the
4 |∆l |
bounded PEP can be expressed as follows,
where βl is given in (17) and γ = |hl | /σn2 is the instanta- l−1 z   
Al X X l − 1 z
neous SNR, which is modeled as exponential random variable PEP (xl , x̂l ) ≤ (−1)j+z+k (γ̄)−z+k
with PDF, γ̄ j=0 j k
1 γ 2
f (γ) = exp − . (25) 4 |∆l |
γ̄ γ̄ Γ(z − k + 1)
Using (25) and the ordered statistics PDF provided in (7) and
after some manipulations, the ordered PDF of the instanta-
neous SNR at the lth user is given by, where z = j + L − l + 1. At high SNR values and considering
l−1     j+L−l+1 the dominant components from the summations in (29), it is
X l−1 j 1 γ observed that the bounded PEP is proportional to the effective
fl (γ) = Al (−1) exp −
j γ̄ γ̄ diversity order,
L! PEP (xl , x̂l ) ∝ γ̄ −z+k−1 . (30)
where Al = (l−1)!(L−l)! .
The effective diversity order is evaluated from (29) using
numerical methods and results are provided in Sec. V.
Therefore, the asymptotic unconditional PEP can be evalu- It has been demonstrated in literature and using numerical
ated as, and analytical results, that power allocation coefficients play
l−1   an essential rule in determining the overall performance of
X l−1 j 1 the NOMA systems. Proper power allocation among different
PEP (xl , x̂l ) ≤ Al (−1) ×
j γ̄ users can enhance the overall performance remarkably. In this
Z ∞ j+L−l+1 ! section, we will form an optimization problem that aims to

exp − exp − 2 dγ. find the optimum power allocation coefficients that minimizes
0 γ̄ 4 |∆l | the average BER. It is worth mentioning that PEP is used to
(27) calculate a union bound on the BER, as follows [14],
Given that the diversity order is evaluated at high SNR M
 the first exponential in (27) can be approximated as Pe ≤ Pm q(x(m) → x̂(m̂) )PEP(x(m) , x̂(m̂) ) (31)
exp − γ̄γ ≈ (1 − γγ̄ ). m=1 m̃=1
where Pm is the probability that x(m) is transmitted and [13]. It is noted here that diversity gain in NOMA systems
q(x(m) → x̂(m̂) ) is the number of bit errors between x(m) is realized due to the ordered channel gains, which in reality
and x̂(m̂) . Therefore, our aim is to find the optimum power represents how far each user from the BS.
allocation coefficients that minimize the following objective Fig. 4 shows the average and individual error rate perfor-
function, mance of NOMA system with two users scenario over different
M M combinations of power allocation coefficients, where SNR =
Ψ= Pm q(x(m) → x̂(m̂) )PEP(x(m) , x̂(m̂) ) (32) 30 dB. From the figure, it is noticed that the second user can
m=1 m̃=1 achieve the threshold error rate at very low and very high
values of α1 . However, at very low values of α1 , the first
while satisfying a specific error rate performance threshold for user has a very poor performance, and this is justified by
all users to maintain user fairness. Additionally, for normalized the increased interference from the second user. Although the
average power, the some of the power allocation coefficients second user achieves the best performance when α1 = 0.7781,
should equals to 1. Hence, the optimization problem can be at this value of α1 the first user exceeds the threshold value,
represented as, where Pth = 10−3 , hence, user fairness is violated in this
scenario. To achieve users’ fairness, where both users have
Minimize Ψ
 PL error rate performance less than the threshold value while the
j=1 αj = 1, (33)
s.t. average BER is kept to the minimum, α1 should take values
PEP(xl , x̂l ) ≤ Pth . from 0.852 to 0.99. Choosing the optimum power allocation
The above optimization problem is solved using numerical coefficients is a tradeoff problem that is determined based on
methods since closed form expressions for the optimum coef- the targeted average BER and the individual BER of each user.
ficients are hard to derive.
In this section, numerical and simulation results are con- U1
ducted to evaluate the performance of the proposed scheme U2
and to validate the derived analytical results. A conventional 10-2
NOMA system is adopted where a single BS and three users
are considered with power allocation coefficients α1 , α2 and
α3 , for the first, second and third user, respectively. Without
loss of generality, we consider the first user as the farthest user, 10-4
α1 > α2 > α3 . All users are equipped with single antenna

and the link between each user and the BS is considered as

Rayleigh flat fading channel. Transmitted signals are chosen
randomly from quadrature phase shift keying (QPSK) constel- 10-6
lation with average power P = 1. It is worth mentioning that
in the presented results, the transmitted signals of different
users are fixed and imperfect SIC is considered.
Fig. 2 presents the PEP for the three users while considering
imperfect SIC scenarios. Power allocation coefficients are
α1 = 0.7, α2 = 0.2 and α3 = 0.1. This power alloca-
tions coefficients values are chosen based on the evaluated
performance of the system, where it is noted that these 0 5 10 15 20 25 30 35 40
Eb /No (dB)
values give good performance in comparison with other values.
The derived analysis are corroborated with simulation results, Fig. 2: Analytical and simulated PEP for the 3 users with
where it is shown that the derived analysis and simulation imperfect SIC.
results match perfectly for the three users over the entire SNR
range. As expected, the PEP gives an indication about the
performance of the three users in NOMA systems in low and VI. C ONCLUSION
high SNR values, where at high SNR value, the near users In this paper, we investigated the performance of NOMA
show strong performance while the far user has relatively weak systems from error rate standpoint. An exact closed form ex-
performance. pression for the PEP is derived, which represents a tight upper
The effective diversity order of different users is shown in bound for the BER, therefore, it can give useful indication
Fig. 3. From the figure, it is observed that at high SNR values, about the BER performance of each user in NOMA systems.
the diversity order of the lth user converges to l. Which is Using the obtained PEP, asymptotic expression is derived,
expected since the asymptotic diversity gain is achieved when which is then used to evaluate the achieved effective diversity
the PEP of NOMA systems behaves as PEP(x, x̂) ∝ γ̄ −de order. Capitalizing on the importance of the allocated power
Avg. BER
U1 U1
U2 U2
Effective Diversity Order

2.5 10-2





0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
30 40 50 60 70 80 90 100 110 120 α1
Eb /No (dB)

Fig. 3: Effective diversity order for the three users, α1 = 0.7, Fig. 4: Average BER over different power allocations, SNR
α2 = 0.2 and α3 = 0.1. = 30 dB, α2 = 1 − α1 , Pth = 10−3 .

coefficients, constrained optimization problem is introduced to [8] X. Liang, Y. Wu, D. W. K. Ng, Y. Zuo, S. Jin, and H. Zhu, “Outage
evaluate the optimum coefficients that reduce the overall error performance for cooperative NOMA transmission with an AF relay,”
IEEE Commun. Lett., vol. 21, no. 11, pp. 2428–2431, Nov. 2017.
rate. Derived expressions, verified by Monte Carlo simulation [9] T. Qi, W. Feng, Y. Chen, and Y. Wang, “When NOMA meets sparse sig-
results, gave an insightful results about the users’ reliability nal processing: Asymptotic performance analysis and optimal sequence
and error rate performance. design,” IEEE Access, vol. 5, pp. 18 516–18 525, 2017.
[10] R. Jiao, L. Dai, J. Zhang, R. MacKenzie, and M. Hao, “On the
performance of NOMA-based cooperative relaying systems over Rician
VII. ACKNOWLEDGMENT fading channels,” IEEE Trans. Veh. Technol., vol. 66, no. 12, pp. 11 409–
This work was supported by ICT fund grant No. 11 413, Dec. 2017.
[11] D. Zhang, Y. Liu, Z. Ding, Z. Zhou, A. Nallanathan, and T. Sato,
11/15/TRA-ICTFund/KU. “Performance analysis of non-regenerative massive-mimo-noma relay
systems for 5G,” IEEE Trans. Commun., vol. 65, no. 11, pp. 4777–4790,
R EFERENCES Nov. 2017.
[1] Z. Ding, Y. Liu, J. Choi, Q. Sun, M. Elkashlan, C. L. I, and H. V. [12] X. Yue, Y. Liu, S. Kang, and A. Nallanathan, “Performance analysis of
Poor, “Application of non-orthogonal multiple access in LTE and 5G NOMA with fixed gain relaying over Nakagami- m fading channels,”
networks,” IEEE Commun. Mag., vol. 55, no. 2, pp. 185–191, Feb. 2017. IEEE Access, vol. 5, pp. 5445–5454, 2017.
[2] W. Shin, M. Vaezi, B. Lee, D. J. Love, J. Lee, and H. V. Poor, “Non- [13] M. Uysal, “Diversity analysis of space-time coding in cascaded Rayleigh
orthogonal multiple access in multi-cell networks: Theory, performance, fading channels,” IEEE Commun. Lett., vol. 10, no. 3, pp. 165–167, Mar.
and practical challenges,” IEEE Commun. Mag., vol. 55, no. 10, pp. 2006.
176–183, Oct. 2017. [14] J. Proakis and M. Salehi, Digital Communications, 5th ed. UK: Ashford
[3] Z. Ding, Z. Yang, P. Fan, and H. V. Poor, “On the performance of Colour Press, 2014.
non-orthogonal multiple access in 5G systems with randomly deployed [15] P. Dhakal, R. Garello, S. K. Sharma, S. Chatzinotas, and B. Ottersten,
users,” IEEE Signal Process. Lett., vol. 21, no. 12, pp. 1501–1505, Dec. “On the error performance bound of ordered statistics decoding of linear
2014. block codes,” in IEEE International Conference on Communications
[4] Z. Ding, P. Fan, and H. V. Poor, “Impact of user pairing on 5G (ICC), 2016, pp. 1–6.
nonorthogonal multiple-access downlink transmissions,” IEEE Trans. [16] M. Simon, Probability Distributions Involving Gaussian Random Vari-
Veh. Technol., vol. 65, no. 8, pp. 6010–6023, Aug. 2016. ables. Springer US, 2002.
[5] Z. Yang, Z. Ding, P. Fan, and N. Al-Dhahir, “A general power allocation [17] A. P. Prudnikov, Integrals and Series, Volume 2: Special Functions.
scheme to guarantee quality of service in downlink and uplink NOMA CRC Press, 1986.
systems,” IEEE Trans. Wireless Commun., vol. 15, no. 11, pp. 7244– [18] I. Gradshteyn and I. Ryzhik, Table of Integrals, Series, and Products,
7257, Nov. 2016. A. Jeffrey and D. Zwillinger, Eds. Academic Press, 2007.
[6] S. Timotheou and I. Krikidis, “Fairness for non-orthogonal multiple
access in 5G systems,” IEEE Signal Process. Lett., vol. 22, no. 10, pp.
1647–1651, Oct. 2015.
[7] L. Zhang, J. Liu, M. Xiao, G. Wu, Y. C. Liang, and S. Li, “Performance
analysis and optimization in downlink NOMA systems with cooperative
full-duplex relaying,” IEEE J. Sel. Areas Commun., vol. 35, no. 10, pp.
2398–2412, Oct. 2017.