NOMA Receiver Design for Delay-Sensitive Systems

Abstract

Successive interference cancellation (SIC) has been considered widely for the detection of downlink non-orthogonal multiple access (NOMA) signals. However, the sequential detection inherent to SIC process may introduce additional time delay for certain users, making the SIC unsuitable for communication systems with time delay constraints such as wireless networks that utilize unmanned aerial vehicles (UAVs) or low-earth orbit (LEO) satellites. Therefore, this work considers the performance of NOMA systems using a joint multiuser detector (JMuD), which can detect the signals of all users simultaneously, and hence, reduce the detection time requirements. The performance of the JMuD is evaluated in terms of bit error rate (BER), computational complexity and processing time, and compared to the SIC detector (SICD). The exact BER of the JMuD is derived analytically using quadrature phase keying (QPSK) modulation where closed-form expressions are derived for the two and three users scenarios for the air-to-ground channel, which is modeled as a Rician fading channels with order statistics. The obtained analytical results corroborated by Monte Carlo simulation confirm that the BERs of the JMuD and SICD are identical, however, the processing time of the SICD is 51% more than the JMuD for several cases of interest. Index Terms-Non-orthogonal multiple access, NOMA, bit error rate, BER, joint detection, successive interference cancellation , SIC, maximum likelihood detection, MLD.

Full text

IEEE SYSTEMS JOURNAL, OCTOBER. 2020 1
NOMA Receiver Design for Delay-Sensitive
Systems
T. Assaf, Member, IEEE, A. Al-Dweik, Senior Member, IEEE, M. S. El Moursi, Senior Member, IEEE, H.
Zeineldin, Senior Member, IEEE, and M. Al-Jarrah, Member, IEEE
Abstract—Successive interference cancellation (SIC) has been
considered widely for the detection of downlink non-orthogonal
multiple access (NOMA) signals. However, the sequential detec-
tion inherent to SIC process may introduce additional time delay
for certain users, making the SIC unsuitable for communication
systems with time delay constraints such as wireless networks
that utilize unmanned aerial vehicles (UAVs) or low-earth orbit
(LEO) satellites. Therefore, this work considers the performance
of NOMA systems using a joint multi-user detector (JMuD),
which can detect the signals of all users simultaneously, and
hence, reduce the detection time requirements. The performance
of the JMuD is evaluated in terms of bit error rate (BER),
computational complexity and processing time, and compared to
the SIC detector (SICD). The exact BER of the JMuD is derived
analytically using quadrature phase keying (QPSK) modulation
where closed-form expressions are derived for the two and
three users scenarios for the air-to-ground channel, which is
modeled as a Rician fading channels with order statistics.
The obtained analytical results corroborated by Monte Carlo
simulation confirm that the BERs of the JMuD and SICD are
identical, however, the processing time of the SICD is 51% more
than the JMuD for several cases of interest.
Index Terms—Non-orthogonal multiple access, NOMA, bit
error rate, BER, joint detection, successive interference can-
cellation, SIC, maximum likelihood detection, MLD.
I. INTRODUCTION
The utilization of unmanned aerial vehicles (UAVs) and small
low earth orbit satellites (LEOSs), such as CubeSat, for various
applications has witnessed a significant increase in the last few
years. Moreover, UAVs and LEOSs are expected to be part of
future wireless networks [1]–[3], which is part of the emerging
concept of flying networks. In flying networks applications, the
communicating devices are expected to download or relay a
massive amount of information in short time periods that are
T. Assaf, M. El Moursi are with the Department of Electrical Engineering
and Computer Science, Khalifa University, Abu Dhabi 127788, United Arab
Emirates. (E-mail: {tasneem.assaf, mohamed.elmoursi}@ku.ac.ae).
A. Al-Dweik is with the Center for Cyber Physical Systems (C2PS)
and Department of Electrical Engineering and Computer Science, Khalifa
University, Abu Dhabi 127788, UAE. He is also with the Department of
Electrical and Computer Engineering, Western University, London, ON,
Canada. (E-mail: {dweik@fulbrightmail.org).
H. H. Zeineldin is with the Faculty of Engineering, Cairo University, Giza,
Egypt and is currently on leave from Khalifa University, Abu Dhabi, UAE.
(E-mail: hatem.zeineldin@ku.ac.ae).
M. Al-Jarrah is with the School of Electrical and Electronic Engi-
neering, University of Manchester, Manchester M13 9PL, U.K. (e-mail:
mohammad.al-jarrah@manchester.ac.uk)
This work was supported by Khalifa University under the flagship project
entitled “MUSES: Multiuse Space Energy Systems”, Grant no. 8474000026.
The work of A. Al-Dweik was supported by the KU Center for Cyber-
Physical Systems, Grant no. C2PS-T2.
upper bounded by the flyover time, which is typically very
short. Furthermore, the timing window for data transfer might
be even limited to the channel coherence time, which is much
shorter than the flyover time. This is required to avoid extensive
channel state information (CSI) exchange, and hence, reducing
the signaling overhead [4]. Although the downlink data rate is
generally determined by the transmitter, it is not actually the
case when the network protocol supports handshaking to provide
reliable data communications [5]. In such scenarios, the average
data rate is determined by both, the transmitter and receiver, and
if the receiver’s response to the handshaking request is slow, the
average data rate may drop significantly [6], [7]. Consequently,
such systems should be equipped with highly efficient transmitter
and receiver designs to satisfy such stringent timing constraints.
Non-orthogonal multiple access (NOMA) is an efficient mul-
tiple access technique, which is considered as a promising can-
didate for future wireless communication networks [8]–[20].
NOMA can improve the spectral efficiency by allowing multiple
users to share the transmission resources simultaneously, at the
expense of some additional receiver complexity and bit error
rate (BER) degradation [21]. Several NOMA schemes have been
proposed in the literature, but the main categories are the code
[14], [22] and power NOMA [23], [24], which is the focus of this
work.
The basic concept of power NOMA is that multiple users may
share transmission resources simultaneously by assigning each
user a particular power [25]. In the literature, successive inter-
ference cancellation (SIC) has been considered widely [9], [11],
[13], [14], [16]–[18], [26]–[28] as the main detection scheme
for NOMA signals. However, SIC detectors (SICDs) suffer from
long processing times because the nth user has to sequentially
detect, modulate and subtract the signals of users 1,2,: : :,n1
[29]. To reduce the processing time, joint multiuser detection
(JMuD) has been proposed as an alternative for the SICD [29]–
[33].
The BER analysis of NOMA using SICD has received exten-
sive attention in the recent literature [9], [14], [18], [34], [35].
Nevertheless, to the best of the authors’ knowledge, the exact
BER of NOMA using JMuD has never been derived. Moreover,
the computational complexity for the three user NOMA and the
processing time of the JMuD has never been compared to the
SICD. Therefore, unlike the work of [33] which presents an upper
bound on the BER and compares the complexity of the SICD
and JMuD for the two-user scenario, the aim of this work is
to derive the exact BER of NOMA systems using JMuD, and
evaluate its computational time and complexity, and compare
them to the SICD for the two and three user scenarios. Such

IEEE SYSTEMS JOURNAL, OCTOBER. 2020 2
comparison enables the system designer to optimize the system
configuration to satisfy time and complexity constraints. The
obtained analytical and simulation results show that the BERs
of the JMuD and SICD are identical, while the processing time
and complexity are drastically different. More specifically, the
processing time of the JMuD is substantially less than the SICD,
because it can perform the detection process for any user without
waiting for other users’ signal to be detected, modulated and
subtracted. On the contrary, the complexity of the JMuD is
generally higher than the SICD because the transmitted signal
constellation size increases exponentially with the number of
users, while it increases linearly in the case of SICD. Moreover,
the SICD structure depends on the relative power of each user,
hence, receiver reconfiguration is required every time the power
coefficients change their order, which may introduce additional
delay and hardware complexity. Such a problem does not exist
in the JMuD, however, all users should use maximum likelihood
detection (MLD) while considering the superimposed constella-
tion.
The rest of the paper is organized as follows. In Sec. II, the sys-
tem and channel models are presented. The exact BER analysis
for the two and three users NOMA systems are presented in Sec.
III and IV, respectively. The complexity comparison between
SICD and JMuD is represented in Sec. VI, while numerical and
simulation results are shown in Sec. VII. Finally, the work is
concluded in Sec. VIII.
II. SYSTEM AND CHANNEL MODELS
This work considers power-domain downlink NOMA systems
with Nusers, U1,:::,UN, where the users’ equipment (UE)
and the base station (BS) are equipped with a single antenna [9].
Consequently, the multiuser signal transmitted from the BS can
be described as
x=
N
X
n=1 pnPTsn(1)
where snis the data symbol of the nth user that is selected
uniformly from a Gray coded quadrature phase shift keying
(QPSK) symbol constellation, PTis the BS total transmit power,
and nis the allocated power coefficient for the nth user. For
the rest of the paper, the transmit power PTis normalized to
unity, and thus, PN
n=1 n= 1. For notation simplicity, we define
sn=ak!s(k)
n,k2 f0;1;2;3g,a0= 00,a1= 01,a2= 10,
and a3= 11. For the case of QPSK and for N3, the inphase
xI,<(x)and quadrature xQ,=(x)components for each
constellation point of the superimposed symbol xcan be defined
as
Au1u2u3=
3
X
i=1
uipi,ui20;1;
1;2(2)
where 
1,1. For example, given that s(2)
1=s(2)
2, then x=
1010 and xI=p1p2,A
1
10;and xQ=p1+
p2,A110:It should be noted that u3represents the amplitude
of the third user in N= 3 NOMA system.
An example for the case of is shown in Fig. 1, where the two
users use QPSK. The most left and right two bits belong to the
first and second users, respectively. As can be noted from the
Fig. 1. The constellation diagram of the transmitted symbol xfor N= 2.
figure, the constellation of the NOMA symbol xlooks like a 16
quadrature amplitude modulation (QAM) constellation.
In flat fading channels, the received signal at the nth UE can
be expressed as
rn=hnx+wn(3)
where hnrepresents the complex channel frequency response
between the BS and nth user, the channel envelope jhnj,n,
and wnis the additive white Gaussian noise (AWGN), wn
CN 0;22
w. In NOMA systems, it is typically assumed that
the first user has the lowest channel gain, the second user has
the second lowest channel gain, and so forth, i.e., 1< 2<
  < N. Thus, the power coefficients should be allocated in the
opposite order of the channel gains, i.e., 1> 2>  > N.
The channel coefficients considered in this work follow the air-
to-ground (AG) model presented in [36] and [37]. The extensive
measurements in [36], [37] show that the AG link can be modeled
as a frequency-selective Rician fading channel with a Kfactor
that ranges from 2to 20 dB.The channel ordering is assumed to
remain fixed for the coherence time of the channel, which can
be more 5ms [38], which allows a reasonable CSI exchange
frequency.
For coherent detection, the channel phase arg fhng,ncan
be estimated and compensated separately from n. Given that
the channel phase nis estimated and compensated perfectly
at the nth UE receiver, then the received signal after phase
compensation can be written as
rn=rnejn=nx+ wn(4)
where wn=wnejn. Assuming that wnis circularly symmet-
ric, then wnwn CN 0;22
w. To extract the information
symbols sn8n, two possible detectors can be utilized, the SICD,
and JMuD. The JMuD is similar the MLD used with QAM
signals, except that the bits in each symbol belong to multiple
users.

IEEE SYSTEMS JOURNAL, OCTOBER. 2020 3
A. Detection of NOMA Signals using JMuD
Based on the constellation diagram of Fig. 1, the JMuD of the
nth user can be written as,
f^s1;^s2;...;^sNg= arg min
~s1;~s2;...;~sNrn^
hn
N
X
i=1 pi~si
2
(5)
where ~siare the trial values of siand ^
hnis the estimated value
of hn. By noting that the modulation order of the superimposed
symbol is M1M2    MN, and thus, the complexity of
the detector increases exponentially as a function of the number
of users. However, it is interesting to note that to detect the
symbols of the nth user, the JMuD can be designed such that the
constellation size is M1M2Mn, and hence, the detector
will be similar to (5) except that Nis replaced by n. Therefore,
the signals of users fn+ 1,n+ 2,...,Mgcan be considered as
unknown additive noise. For example, the first user detector can
be written as
^s1= arg min
~s1r1^
h1p1~s1
2(6)
which is the typical MLD for QPSK signals. It is straightforward
to show that the BERs of the two detectors in (5) and (6) are
identical, and thus, it is preferable complexity wise to use the
MLD in (6).
B. Detection of NOMA Signals using SICD
The NOMA signals can also be detected using the SICD
approach, where the signal for the nth user is detected after
detecting and subtracting the signals of the first n1users.
Therefore, MLD is applied ntimes, however, the constellation
size in each round is equal to the modulation order of the nth
user signal, and thus,
^sn= arg min
~snrn^
hn
n1
X
i=1 pi^si^
hnpn~sn
2
:(7)
For the first user, it is clear that both the MLD (6) and SICD (7)
have the same structure, and hence, the same BER, which is given
in [9]. Moreover, it is worth noting that the similarity between the
SICD and JMuD in the sense that the signals of U1,:::; Un1are
involved in the detection of the signal of Un:
III. JMUD BER ANALYSIS (U1jN=2 ,U2jN=2)
This section considers a two-user NOMA system using QPSK
modulation. The transmitted superimposed symbol xis the super-
position of two QPSK symbols, and hence, it corresponds to one
out of 16 constellation points each of which has 4bits, as shown
in Fig. 1. The bit representation for each point is expressed as bni,
fn; ig2f1,2g, where nis the user index and iis the bit index.
For the first user, the detection is performed using (6). Based
on the specific value of s1and the inter-user interference (IUI)
caused by U2, symbol xmay become one of the four constellation
points in the neighborhood of s1shown in Fig. 1.
The average BER should consider all possible combinations of
s1and s2,
Pb1i=X
l;k Pb1ijs(l)
1,s(k)
2Ps(l)
1,s(k)
2:(8)
It should be noted that s1and s2are independent, then (8) can be
written as,
Pb1i=1
16
3
X
fl:kg=0 Pb1ijs(l)
1,s(k)
2:(9)
Case 1:s(2)
1,s(0)
2: The error probability of b11,Pb11 js(2)
1,s(0)
2,
depends only on the inphase component of r1, i.e., <(r1),r1
and the specific value of x; x =A1
10 +jA110 . Thus,
Pb11 js(2)
1,s(0)
2= P(r10)
= P(1A1
10 +n10)
= P(n11A1
10)(10)
where r1=1A1
10 +n1,<( w1),n1. Therefore,
Pb11 js(2)
1,s(0)
2=1
p22
n1Z1
1A1
10
ez2
22
n1dn1
=Qp1;1(11)
where 1;1=2
1A2
1
10,=1
2
wand Q(:)denotes the Gaussian
Qfunction.
Case 2:s(2)
1,s(1)
2: By following the same approach of Case
1, the error probability of this case is given by (11) as well.
Case 3:s(2)
1,s(2)
2:
The error probability can be expressed as
Pb11 js(2)
1,s(2)
2= P(r10)
= P(n11A110):(12)
Following the same approach used to derive (11) gives,
Pb11 js(2)
1,s(2)
2=Qp1;2(13)
where 1;2=2
1A2
110.
Case 4:s(2)
1,s(3)
2: The probability of error in this case is
similar to the case of s(2)
1,s(2)
2.
Case 5to C ase 16 are evaluated using the same approach,
except that the value of s1is replaced by a0,a1,a2and a3.
Through the substitution of the 16 cases results in (9), the
conditional BER for the bit b11 can be expressed as
Pb11 =1
2Qp1;1+Qp1;2:(14)
Clearly, Pb12 =Pb11. Therefore, the conditional BER of the first
user is given:
PU1=1
2[Pb11 +Pb12 ]
=1
2Qp1;1+Qp1;2:(15)
As for the second user, the JMuD requires a 16-point MLD as
shown in (5).
The error probability for the second user can be computed by
evaluating the error probability for each individual bit, i.e., b21
and b22. Because the error probability depends on s1and s2, then,
Pb2i=X
l;v Pb2ijs(l)
1,s(v)
2Ps(l)
1,s(v)
2:(16)

IEEE SYSTEMS JOURNAL, OCTOBER. 2020 4
By noting that s1and s2are independent, then (16) for the 16-
point constellation can be written as,
Pb2i=1
16
3
X
fl;vg=0 Pb2ijs(l)
1,s(v)
2:(17)
The following cases cover the error probability of all possible
constellation points of x.
Case 1:s(0)
1,s(0)
2: For this case, the transmitted signal is
x=A110 +jA110 and an error occurs if ^s1=a2(10) or a3
(11). Consequently, the error probability of b21 can be expressed
as,
Pb21 js(0)
1,s(0)
2= P(r22A100;r20[r22A
100)
= P(r22A100;r20) + P(r22A
100)
P (r22A100;r20;r22A
100)
= P(n22A0
10)P(n22A
1
10)
+P(n22A
2
10)(18)
where r2,<(x) = 2A110 +n2, and <( w2),n2. By noting
that
P(n22Au1u2u3) = 1
p22
n2Z1
2Au1u2u3
ez2
22
n2dn2
=Qp2;c(19)
where 2;c =2
2A2
u1u2u3. Then,
Pb21 js(0)
1,s(0)
2=Qp2;1Qp2;2+Qp2;3:(20)
where 2;1=2
2A2
010,2;2=2
2A2
110, and 2;3=
2
2A2
210. Following the same approach for all constellation
points that correspond to the combinations [s(0)
1; s(1)
2],[s(1)
1,
s(0)
2]and [s(1)
1,s(1)
2]gives the same probability of error for
Pb21 js(0)
1,s(0)
2. Therefore, Pb21 js(0)
1,s(0)
2=Pb21 js(0)
1,s(1)
2=
Pb21 js(1)
1,s(0)
2=Pb21 js(1)
1,s(1)
2, which is given by (20).
Case 2:s(0)
1,s(2)
2: For this case x=A1
10 +jA110 . The
error probability of b21 can be obtained as
Pb21 js(0)
1,s(2)
2= P(r22A100 [r22A
100;r20)
= P(r22A100) + P(r22A
100;r20)
P(r22A100;r22A
100;r20)
= P(n22A010)
+P(n22A
210;n22A
110)
= P(n22A010)
+P(2A
210 n22A
110)
=Qp2;1+Qp2;4Qp2;5(21)
where 2;4=2
2A2
1
10 and 2;5=2
2A2
2
10. It can be noted
that the constellation points that correspond to the combinations
[s(0)
1; s(3)
2],[s(1)
1,s(2)
2]and [s(1)
1,s(3)
2]have the same probability,
which is given by (21).
Case 3:s(2)
1,s(0)
2: For this case x=A1
10 +jA110 , then
the error probability can be expressed as
Pb21 js(2)
1,s(0)
2= P(r22A
100 [r22A100;r20)
= P(r22A
100) + P(r22A100 ;r20)
P(r22A
100;r22A100 ;r20)
= P(n22A0
10) + P(n22A2
10)
P(n22A1
10)
=Qp2;1+Qp2;4Qp2;5:(22)
The same error probability in (22) is also applicable for the
combinations [s(2)
1; s(1)
2],[s(3)
1,s(0)
2]and [s(3)
1,s(1)
2].
Case 4:s(2)
1,s(2)
2: For this case x=A110 +jA110 , and
the error probability can be expressed as
Pb21 js(2)
1,s(2)
2= P(r22A
100;r20[r22A100 )
=Qp2;1Qp2;2+Qp2;3:(23)
The result in (23) is also applicable to the combinations
[s(2)
1; s(3)
2],[s(3)
1,s(2)
2]and [s(3)
1,s(3)
2].
Finally, substituting (20), (21), (22), and (23) into (17) gives
the unconditional error probability of b21. Moreover, by follow-
ing the same approach for b22;it can be shown that P22 =P21.
Consequently, the error probability for the second user can be
expressed as
PU2=1
2
5
X
i=1
viQp2;i,v = [2;1;1;1;1] :(24)
Interestingly, PU1and PU2are actually equal to [9, Eq. 12 and
36], respectively, which implies that the BERs of the JMuD and
SICD are identical. Nevertheless, the analysis for the JMuD is
much simpler because it does not require considering the impact
of the SICD process on users 1;2;3; : : : ; n 1:The average
BERs over ordered Rician flat fading channels, 
PU1and 
PU2,
are evaluated by averaging over the probability density function
(PDF) of n;i as shown in Appendix I.
IV. JMUD BER ANALYSIS (U2jN=3,U3jN=3)
In this section, the BER derivation of N= 3 downlink
NOMA system is illustrated. The transmitted symbol xforms
a64-point constellation as shown in [9], and the first, second,
and third users’ signals are given by s1; s2;and s3, respec-
tively. The binary bit representation for the three users is given
by b11 b12 b21 b22 b31 b32, for each bit bni ,n=
f1,2;3g, and i=f1;2g. Similar to N= 2 scenario, the
derivation of the first user BER is the same for both JMuD
and SICD [9] because MLD given in (6) is used for QPSK
constellation.
For the second user, the BER is evaluated as follows. The
probability of error for each bit depends on the values of s1and
s2, and hence, the average BER should consider all possible com-
binations. Nevertheless, due to space limitations, and because the
solution procedure is similar for all cases, we consider only the

IEEE SYSTEMS JOURNAL, OCTOBER. 2020 5
case of [s(0)
1; s(0)
2; s(0)
3]. For this case, x=A111 +jA111 and
b21 can be expressed as,
Pb21 js(0)
1,s(0)
2= P(r22A100;r20[r22A
100)
= P(2A
1
1
1n22A0
1
1) + P(n22A
2
1
1)
=Qp3;5Qp3;4+Qp3;10(25)
where 3;4,3;5;and 3;10 are defined in [9]. By following the
same approach used in Sec. III, the exact BER for both b21 and
b22 can be evaluated for all possible combinations of s1,s2,
which interestingly show that the JMuD and SICD have equal
BERs.
For U3jN=3, the JMuD adopts a 64-point MLD constellation.
Similar to the previous cases, the error probability depends on s1,
s2and s3. Nevertheless, because the solution procedure is similar
for all cases, we only consider the case of [s(0)
1; s(0)
2; s(0)
3]. For
this case, Pb31 can be expressed as shown in (26), where the terms
3;4,3;5,3;10,3;11 ,3;12,3;13 , and 3;18 are defined in [9].
It is worth noting that deriving analytical BER expressions
for NOMA using arbitrary number of users Nand modulation
orders M1,M2,: : :,MNis highly desirable. However, such a
task requires an entirely dedicated work as reported in [39], where
the BER is derived for an arbitrary Mn, but the results are limited
for the two-user scenario, i.e., N= 2.
V. POWER ALLOCATION
The power allocation (PA) problem is formulated to minimize
the overall average BER of the NOMA system while ensuring
fairness for all users. Fairness is defined as maintaining a specific
BER thresholds for each user. Therefore, the problem can be
formulated as
min
•
n
1
N
N
X
n=1

PUn(27a)
subject to:
N
X
n=1
n= 1 (27b)

PUn
Pth
Un,8n2 f1;2; : : : ; N g(27c)
l>k,l6=k,l < k,fl; k g2f1;2; : : : ; N g(27d)
where •
nand 
Pth
Unare the optimal power coefficient and BER
threshold for the nth user, respectively. Constraint (27b) is used to
assure that the total transmit power is limited to unity. Constraint
(27c) is used to assure that each user’s BER is less than its desired
threshold, and (27d) forces the power allocated for each user to be
inversely proportional to its channel gain, i.e., 1> 2>  >
Nare assigned for the users with the channel gains 1< 2<
  < N, respectively. The problem in (27a) is a constrained
non-linear optimization problem which is analytically intractable.
Hence, numerical methods such as brute-force and bisection [7]
can be used to search for the near-optimum power coefficients.
It should be noted that the difference from the optimal solution
depends on the search step size.
TABLE I
COMPUTATIONAL COMPLEXITY OF THE JMUDAND SICD.
Operation SICD JMuD Ratio %
U2jN=2 53 176
70 176
Overall Equivalent Complexity 282 880 32:0
U2jN=3 53 176
70 176
Overall Equivalent Complexity 282 880 32:0
U3jN=3 82 704
108 704
Overall Equivalent Complexity 436 3520 12:3
VI. JMUDAND SICD COMPLEXITY
Although the JMuD and SICD have the same BER perfor-
mance, their receivers’ structure are fundamentally different, and
hence, the complexity for each detector should be evaluated
and compared. The complexity comparison in this work will be
conducted by evaluating the computational time and hardware
complexity. The computational complexity is evaluated in terms
of the number of real arithmetic operations required to evaluate
the detectors described in (5) and (7), where N= 2 and 3,
and all users adopt QPSK modulation. For more informative
comparison, the overall equivalent complexity is also presented
[40] as depicted in Table I. As can be noted from the table, the
computational complexity of the JMuD is considerably higher
than the SICD due to the large number of multiplications as-
sociated with the MLD. Moreover, the JMuD complexity in-
creases exponentially by increasing the user index because the
modulation order of the symbol to be detected increases also
exponentially as a function of its index. Thus, the complexity
depends on the user index rather than the total number of users.
The time complexity corresponds to the latency of receiver,
which is defined as the time required to produce the final hard
bits from the received signal. Generally speaking, the MLD can
perform the detection process for all users simultaneously, and all
the Euclidean distance measurements can be performed in paral-
lel. For the SICD, the detection process per user is performed
sequentially, i.e., first user, then second user, etc. Moreover, the
detected bits of each user has to be re-encoded and re-modulated
to be subtracted from the superimposed signal. As a result, the
time complexity of the SICD is expected to be more than the
JMuD. The results presented in Sec. VII show that the simulation
time required by the SICD is about 86% as compared to the
JMuD in certain scenarios. Therefore, the JMuD is more suitable
for time sensitive applications.
The hardware complexity of the JMuD is similar to a con-
ventional M-ary QAM MLD. For the SICD, the receiver should
also implement part of the transmitter chain for the SIC process.
Therefore, the SICD should consist of an encoder and modulator,
in addition to the MLD. Although most devices are built as
transceivers, using the transmitter subsystem in the detection
process is mostly infeasible as most systems support full-duplex
operations. Therefore, the detector should have dedicated en-
coders and modulators. It is also worth noting that the SICD has
to be reconfigured when the power order of the users is changed,
which may introduce additional delay and hardware complexity.
Such a problem does not exist in the JMuD

IEEE SYSTEMS JOURNAL, OCTOBER. 2020 6
Pb31 js(0)
1,s(0)
2,s(0)
3
= Pr33A110;r33A100 [r33A1
10;r30[r3 3A1
10;r33A
100 [r3 A110
= Pn33A00
1;n33A0
1
1+ P3A111 +n33A1
10,3A111 +n30+ P(3A111 +n3 A110 )
+ P3A111 +n3 3A1
10,3A111 +n33A
100
= Pn33A00
1Pn33A0
1
1+ Pn33A0
2
1Pn33A
1
1
1+ Pn33A
20
1
Pn33A
2
1
1+ Pn33A
2
2
1
=Qp3;4Qp3;5Qp3;10+Qp3;11 +Qp3;12+Qp3;13 +Qp3;18(26)
___________________________________________________________________________________________
010 20 30
10-5
10-4
10-3
10-2
10-1
100
BER
Theoretical
K = - dB
K = 2 dB
K = 6 dB
K = 10 dB
K = 15 dB
K = 20 dB
010 20 30
U1U2
Fig. 2. The average BER for the U1and U2where K=
[1;2;6;10;15;20] dB, 1= 0:7,2= 0:3:
VII. NUMERICAL RESULTS
This section presents the analytical and Monte Carlo simula-
tions results for the JMuD and SICD over Rician channel. Two-
user and three-user NOMA downlink systems are considered,
N= 2 and N= 3. All users are assumed to be equipped with a
single antenna, and the channel between the BS and each user is
modeled as an ordered Rician flat fading channel. The randomly
generated channels are ordered based on their strength, where
the weakest channel is assigned to U1and the strongest channel
is assigned to UN. The transmitted symbols for all users are
selected uniformly from a Gray coded QPSK constellation. The
total transmit power from the BS is unified for all cases, PT= 1.
The simulation results are obtained using a computing machine
that runs Intel Xeon CPU E5-2640 processor, clock frequency of
2:5GHz, 16 GB RAM, and 64 bit operating system.
Fig. 2 presents the analytical and simulation BER performance
for power coefficients 1= 0:7and 2= 0:3and various
values of Kover a range of , where =1
2
w. As can be noted
from the figure, the analytical results obtained using (45) and
(53) perfectly match the simulation results for all the considered
values of Kand . It is worth noting that the Rayleigh fading case
corresponds to K=1 dB. As can be seen from the figure, the
performance of the first user is more sensitive to the variations of
Kas compared to the second user, which is due to the fact that
010 20 30
10-5
10-4
10-3
10-2
10-1
100
BER
Theoretical
K = - dB
K = 2 dB
K = 6 dB
K = 10 dB
K = 15 dB
K = 20 dB
010 20 30
0102030
U2U3U1
Fig. 3. The average BER for the U1,U2and U3;where K=
[1;2;6;10;15;20] dB, 1= 0:8,2= 0:15,3= 0:05:
the fading effect becomes less significant for the near users.
Fig. 3 shows the BER for N= 3 using various Kvalues.
As can be noted from the figure, U3has generally the best per-
formance although it is allocated the smallest power coefficient.
Such performance is obtained because U3has the best channel as
compared to U1and U2. Similar to the two-user NOMA system,
the error performance of the first user is more sensitive to the
value of Kif compared to the second and third users. This is due
to the fact that the fading effect becomes less significant for the
near users.
Although the BER expressions in (46) are represented in terms
of infinite series, the series can be truncated to have fG; Mg
and fG; M; Z gfinite terms, for the first and second users,
respectively. In order to ensure numerical accuracy, the values
of G; M; and Zare selected based on the system parameters,
such as the fading parameter Kand . For simplicity, it is
assumed that G=Mand G=M=Z, for the first and
second users, respectively. Tables II and III show the normalized
truncation error for the average BER expressions of U1and U2.
The normalized truncation error is computed as follows
et=jBERSi mBERTheoj
BERSi m(28)
where the BERTheo and BERSi mare the analytical and simu-
lated BERs, respectively. The depicted results show that when

IEEE SYSTEMS JOURNAL, OCTOBER. 2020 7
TABLE II
THE NORMALIZED TRUNCATION ERROR FOR = 10 AND 20 DBAND
K= 6 DBFOR THE FIRST USER.
fG; M get,= 10 dB fG; M get,= 20 dB
5 0:4636 5 0:2470
10 3:76 10210 2:7103
30 6:3955 10430 1:5103
50 6:3955 10450 1:5103
TABLE III
THE NORMALIZED TRUNCATION ERROR FOR = 10 AND 20 DBAND
K= 6 DBFOR THE SECOND USER.
fG; M; Z get,= 10 dB fG; M; Z g,et,= 20 dB
5 0:4913 5 2:2748
10 5:61 10210 4:11 102
30 1:5777 10430 2:3103
50 1:5777 10450 2:3103
fG; M g= 30, the truncation error is fairly insignificant for
different values and K= 6 dB.
Fig. 4 shows the effect of Kon the BER of each user for
N= 2 with = 16 and 18 dB. The power allocation coefficients
are 1= 0:7and 2= 0:3. As can be noted from the
figure, the BER is highly dependent on K, particularly for U1
who experiences more severe fading conditions. Consequently,
its BER is more sensitive to the values of K. Moreover, the BER
for both users becomes more sensitive to Kas increases, which
is due to the fact that the channel fading effect becomes more
dominant at high SNRs. It is also worth noting from the figure that
the BER of U2does not necessarily decrease when Kincreases.
For example, Fig. 4 where = 18 dB shows that the BER keeps
decreasing until K= 15 dB, and then increases again until it
saturates when the AWGN and IUI have more impact on the BER.
Such behavior can be justified using the envelop of the received
signal PDFs given in Fig. 5, which represents the conditional PDF
for the received signal of the second user at = 10 dB and
K= 8 and 10 dB. Clearly, the two PDFs have the same mean,
but the standard deviation for the case of K= 8 dB is lower.
It should be noted that as K! 1, the BER will be mostly
determined the IUI, the power coefficients, and AWGN. For the
power coefficients considered in this Fig. 4, the BER of both users
converge roughly to the same value.
Table IV presents the optimal power coefficients that minimize
the BER for different values of , and for K= 2;6and 15 dB,
TABLE IV
POWER ALLOCATION TO ACHIEVE MINIMUM AVERAGE BER, K= 2;6
AND 15 dB.
K= 2 dB K= 6 dB K= 15 dB
dB 121212
10 -- ----
20 - - 0:86 0:14 0:83 0:17
30 0:95 0:05 0:94 0:06 0:92 0:08
TABLE V
COMPUTATIONAL TIME IN MILLISECONDS.
TSI CD TJ MuD T(%)
U1jN=2 12 12 0:0
U2jN=2 52 29 79:3
U1jN=3 20 20 0:0
U2jN=3 41 22 86:3
U3jN=3 68 45 51:1
-10 010 20 30 40
10-2
10-1
BER
U1
U2
16 dB
18 dB
Fig. 4. The average BER for the first and second users under Rician channel,
where = 16 and 18 dB.
-1 -0.5 00.5 1
0
0.5
1
1.5
2
K = 8 dB
K = 20 dB
-0.2 00.2
Fig. 5. The conditional PDF for the received signal of the second user at
= 10 dB.
and for BER thresholds of 
Pth
U1= 102and 
Pth
U2= 103. The
power coefficients are obtained using brute-force with a step size
of 0:01. As can be noted from the table, the BER constraint could
not be satisfied for = 10 dB regardless the value of K, and for
K= 2 dB when = 20 dB. The results also show that most
of the power is allocated for U1, particularly at high . The first
user is allocated more than 95% of the total power at = 30 dB
and K= 2 dB. The same trends can be noted for the K= 6 and
15 dB cases. Nevertheless, the power given to the U1generally
decreases by increasing K.
Fig. 6 shows the average BER for the two users and system
average BER over a range of 1values at = 20 dB and K= 2;
and 6dB. At K= 2 dB, minimum average BER cannot be found

IEEE SYSTEMS JOURNAL, OCTOBER. 2020 8
0.6 0.7 0.8 0.9
10-6
10-5
10-4
10-3
10-2
10-1
100
BER
U1
U2
Average BER
15 dB
2 dB
Fig. 6. Average BER for the first and second users over different power
allocation coefficients, K= 2;15 dB and Eb=N0= 20 dB.
for 
Pth
U1= 102and 
Pth
U2= 103. As for the other scenario
where K= 15 dB, the power coefficients are 1= 0:82 and
1= 0:18.
Fig. 7 shows the BER performance for the two users in N= 2
downlink NOMA system using JMuD and SICD [9]. As can be
noted both detectors result in identical BER performance over the
entire range and Kvalues.
To evaluate the impact of the sequential detection on the
delay of the SICD, the simulation time for the SICD, TSI CD , is
measured for a total of 106symbols and compared to the JMuD
simulation time TJM uD . Table V presents TSICD and TJM uD
in milliseconds, and also shows the normalized difference T
between the two detectors, T,jTSI CD TJ MuD j=TJ M uD.
As can be noted from Table V, the SICD requires significant
extra time due to the sequential detection process. It is also worth
noting that the simulation time for the JMuD also increases by
increasing the user order due to the additional computational
complexity caused by the increased constellation order that the
JMuD has to detect. As expected, the simulation time for the first
user is independent of Nand the detector type.
Fig. 8 shows the average BER for N= 2 with imperfect
channel estimates. The channel estimation errors are modeled as
CN 0;  2
^
h[41], [42], where 2
^
h= 1 102and 2103.
As can be noted, both detectors perform equivalently under
channel estimation errors, and both detectors are highly sensitive
to estimation errors.
Fig. 9 presents the BER for the first and second users in N=
2system using JMuD under Rician and Nakagami-mchannels.
Every value of Kcorresponds to an mvalue, where the relation
is as follows [43]
m=
K2
1+2K:(29)
where K+1 ,
K. Clearly as Kincreases the BER results under
the two distributions become closer. For the considered scenario,
the BER of the two channels completely matched when K=
010 20 30
10-5
10-4
10-3
10-2
10-1
100
BER
010 20 30
U1U2
Fig. 7. Average BER comparison between JMuD and SICD under Rician
channels for the U1and U2users, where K= [1;2;10;20] dB.
010 20 30
10-3
10-2
10-1
BER
010 20 30
U1U2
Fig. 8. Average BER for N= 2 downlink NOMA system with JMuD
and SICD with perfect channel estimation, imperfect channel estimation,
1= 0:7, and 2
^
h= 102and 2103.
17 dB and m= 14:1182 dB: It should be noted that for large
values of K, the Rician distribution can be approximated by the
Nakagami-mdistribution.
The diversity gain can be defined as Dn,lim!1 log 
PUn
log ,
which is intractable for the Rician channel due to the hypergeo-
metric and Bessel functions. Therefore, an approximated Dncan
be derived by approximating the Rician PDF by the Nakagami-m
distribution. Consequently, using the BER expression for NOMA
in Nakagami-mchannels [9, Eq. (101)], and considering only the

IEEE SYSTEMS JOURNAL, OCTOBER. 2020 9
010 20 30
10-5
10-4
10-3
10-2
10-1
100
BER
Rician
Nakagami-m
K = 2 dB
K = 10 dB
K = 14 dB
K = 17 dB
010 20 30
U1U2
Fig. 9. The average BER for the first and second users under Rician and
the equivalent Nakagami-mfading channels, where K= 2;10;14;and 17
dB.
TABLE VI
THE THEORITICAL AND NUMERICAL DIVERSITY FOR N= 2 NOMA
SYSTEM,K= [1, 2, 6] dB.
KdB D1(30) ~
D11D2(30) ~
D12
1 1 0:998 0:002 2:0 1:99 0:005
2 1:6024 1:490 0:069 3:204 3:11 0:0274
6 2:7684 2:653 0:041 5:880 5:676 0:0244
dominant components, the diversity gain can be given as
Dn  lim
!1
log 1
mn
log 
mn: (30)
Table VI shows the diversity gain for N= 2 and K=
[1;2;6] dB obtained from (30) compared to another approx-
imation obtained from the BER curves denoted as ~
Dn, which is
the BER slope at high values. As can be noted from the table,
the normalized difference n,Dn~
Dn=Dn. As can be
indicated from (30), the diversity gain of the nth user converges
to mn. Moreover, the normalized difference nis fairly small
for all the considered cases.
VIII. CONCLUSION AND FUTURE WORK
This work presented the BER analysis of the JMuD for down-
link NOMA systems where a closed-form analytical expressions
are derived for several cases of interest. Moreover, the work
presented a general comparison for the JMuD and the widely
adopted SICD. The comparison is performed in terms of BER,
computational complexity, processing time, and sensitivity to
channel estimation errors. Although the two detectors are funda-
mentally different, the obtained analytical and simulation results
show that both detectors have identical BERs in the case of
perfect and imperfect channel estimates, yet the derivation is
significantly different. The BER analysis for the JMuD requires
considering much smaller number of cases, and hence, can be
considered much simpler than the SICD BER analysis. The
BER sensitivity to channel estimation errors was evaluated using
Monte Carlo simulation for both systems, and the obtained results
show that both systems have equivalent sensitivity. The compu-
tational complexity results show that the SICD has much lower
complexity as compared to the JMuD, but at the expense of a
longer processing time. Therefore, the JMuD would be preferable
for cases where the processing delay is paramount while the
SICD would be preferable in all other cases.
As can be noted from this work and the listed references,
the exact BER derivation for an arbitrary number of users and
modulations schemes remains and open problem that will be
considered in our future work. Moreover, the analysis will be
extended to the case where the phase for each user can be
modified to improve the system performance.
IX. APPENDIX I: AVERAGE BER OVER RICIAN FADING
CHANNEL
As described in Section II, the channel between the UAV and
ground follows the Rician distribution, thus, the envelops of the
channels n8n2Nare ordered set of Rician random variables.
Using the theory of order statistics, the PDF for each channel can
be expressed as [44, pp. 225],
fn() = Knf() [F()]n1[1 F()]Nn(31)
where Kn=N!
(n1)!(Nn)! ,is the channel envelop which
follows Rician distribution, f()and F()are the PDF and
CDF of , respectively, which are given by
f() = 2
K
e
K2
+KI0 r4K
K2
!(32)
and
F()=1Q1 p2K; r2K2
!(33)
where Kand are the Rician distribution parameters, K+ 1 ,

K,I0()is the Modified Bessel function of the first kind, and
Q1(;)is the Marcum-Q function. The infinite series represen-
tation of Q1can be represented as [45]
Q1(x; y)=ex2+y2
2
1
X
m=0 x
ym
Im(xy):(34)
Therefore, the general ordered PDF of the nth channel envelope
over Rician channel is
fn() = 2
KKn
 e
K2
+K"1Q1 p2K; r2
K
2!#n1
"Q1 p2K; r2
K2
!#Nn
I0 r4K
K2
!:(35)
In order to avoid confusion, let’s denote fn()!fn(n).
Now, , follows the noncentral chi square 2distribution with
the following PDF and CDF [46]
f() = 
Ke(K+
K$)
I0p4K
K$(36)

IEEE SYSTEMS JOURNAL, OCTOBER. 2020 10
and
F() = 1 Q1p2K; p2
K$(37)
where $=
:By using (31), the general ordered PDF for , can
be expressed as follows
fn() = 
KKne(K+
K$)
I0p4K
K$
"1e(K+
K$)1
X
m=0 rK

K$ !m
Imp4K
K$#n1
"e(K+
K$)1
X
m=0 rK

K$ !m
Imp4K
K$#Nn
:
To avoid confusion, let fn()!fnn;i, where iis a para-
meter index. For simplicity, let n;i =
KeK
n;i ,n;i =4K
K
n;i ,
'= 2K, and n;i =2
K
n;i .
Then, for the two-user NOMA system, the ordered PDFs for
the first and second users n;i,n2 f1;2g,N= 2 can be,
respectively, represented as
f11;i=K11;i e1;i
4I0p1;i 1;ie1;i 1;i

1
X
m=0 '
1;i 1;i !m
2
Imp1;i 1;i(38)
and
f22;i=K22;i e1;i 2;i
2I0p2;i2;i 
K22;ie'
2I0p2;i 2;ie2;i 2;i

1
X
m=0 '
2;i 2;i !m
2
Imp2;i 2;i:(39)
In order to evaluate the average BER of the two users in N= 2
system over Rician channel, (38), (39) are utilized as follows
Z1
0
Qpn;ifnn;i dn;i :(40)
By using (15) and (38), the average BER for n= 1 can be
computed as
PU1=1
2
2
X
i=1 Z1
0
Qp1;if11;i d1;i 
=K1
2
2
X
i=1
1;ie'
2Z1
0
Qp1;i
I0p1;i1;i e1;i1;i

1
X
m=0 '
1;i1;i m
2
Imp1;i1;i d1;i (41)
By replacing the Qfunction by its alternative representation
defined by [47], Q(x) = 1
R
2
0ex2
2 sin2()d, (41) becomes
as in (42), where n;i (%) = 2 sin2(n;i )
2%sin2(n;i)+1 .
Evaluating the inner integral in (42) gives the
single integral expression given in (43) [45], where
2F1([t1; t2; :::tp] ; [A1; A2; :::; gq] ; J)is the generalized
hypergeometric function which can be represented by the
following infinite series [48]
2F1([t1; t2; :::tp] ; [A1; A2; :::; Aq] ; J) =
1
X
g=0
Jgp
Q
l=1
(tl+g)
(tl)
q
Q
L=1
(AL+g)
(AL)
:
(44)
Then, by replacing the generalized hypergeometric function by
its series expansion, the average BER for the first user can be
expressed as in (45) in the next page.
In order to evaluate the average BER for the second user, both
(24) and (38) are used as given in (46). Now, the term P1
U2can be
computed as
P1
U2=K22;i Z1
0
Qp2;ie2;i 2;i
2I0p2;i 2;id2;i:
(47)
By using the integral representation of the Qfunction,
P1
U2=K22;i
Z
2
0Z1
0
e
22;i
2;i(2;i )
I0p2;i 2;id2;id2;i (48)
The evaluation of the inner integral in (48) is given as [45]
P1
U2=K22;i
Z
2
0
2;i 2;i e
2;i 2;i
2!2;i
4d2;i:(49)
The integral in (49) is analytically intractable, hence, the follow-
ing series representation for exis utilized
ex=
1
X
z=0
xz
z!= 1 + x+x2
2+x3
6+x4
24 :(50)
Then,
P1
U2=K22;i

1
X
z=0
1
z!2;i
4zZ
2
02;i 2;i
2z+1
d2;i
(51)
Evaluating the integral gives
P1
U2=K22;i
p
1
X
z=0
1
z! 2;i
22;i !z
0
@22F11
2;3
2+z;3
2;1
2
3
2
2;i  (z+ 1) 13
2+z+s
2
2;i 1
A(52)
The second term P2
U2in (46) can be computed by following
the approach of PU1. As a result, the total average BER for the
second user can be represented as in (53) in the next page.
REFERENCES
[1] S. Sekander, H. Tabassum, and E. Hossain, “Multi-tier drone architec-
ture for 5G/B5G cellular networks: Challenges, trends, and prospects,”
IEEE Commun. Mag., vol. 56, no. 3, pp. 96–103, Mar. 2018.
[2] M. Jia, X. Gu, Q. Guo, W. Xiang, and N. Zhang, “Broadband hybrid
satellite-terrestrial communication systems based on cognitive radio
toward 5G,” IEEE Wireless Commun., vol. 23, no. 6, pp. 96–106, Dec.
2016.

IEEE SYSTEMS JOURNAL, OCTOBER. 2020 11
PU1=K1e'
2
2
2
X
i=1
2
6
4Z1
0Z
2
0
1;ie
 1;i
2 sin2(1;i)!d1;i I0p1;i1;i e1;i1;i
1
X
m=0 s'
1;i1;i !m
Imp1;i1;i d1;i 3
7
5
=K1e'
2
2
2
X
i=1 2
4
1
X
m=0
1;i '
1;i !m
2Z
2
0Z1
0
1
1;im
2
e
1;i
1;i(1;i )I0p1;i1;i Imp1;i1;i d1;i d1;i3
5:(42)
_____________________________________________________________________________________________________________________________
PU1=K1e'
2
2
2
X
i=1
2
6
4
1
X
m=0
1;i1;i 1;i '1;i
41;i m
2
 (m+ 1) 2F11 + m
2;m
2+1
2;m+ 1; m + 1; 1;i1;i 1;i d1;i3
7
5(43)
_____________________________________________________________________________________________________________________________
PU1=K1e'
2
2
2
X
i=1
2
6
4
1
X
m=0
1
X
g=0
1;i (m+ 1) '1;i
41;i m
2g!
 (m+1+g)2"1 + m
2+g
1 + m
2m
2+1
2+g
m
2+1
2#g
1;i
Z
2
01;i 1;ig+1 d1;i #
=K1e'
2
p
2
X
i=1
2
6
42
6
4
1
X
m=0
1
X
g=0
1;i 1;i'
41;i m
21;i
2gm
2+g (m+ 2g)
( (g+ 1)  (m+g+ 1))23
7
5
0
B
@22F11
2;3
2+g;3
2;1
21;i 3
2+g
21;ig+3
2
+p (g+ 1)
21;ig+1 1
C
A3
7
5:(45)
_____________________________________________________________________________________________________________________________
PU2=1
2
5
X
i=1
viZ1
0
Qp2;if22;i d2;i ,v = [2;1;1;1;1]
=K2
2
5
X
i=1
viZ1
0
Qp2;i2;i e
2;i2;i
2I0p2;i 2;id2;i e'
2Z1
0
Qp2;i2;i I0p2;i 2;i e2;i2;i

1
X
m=0 '
2;i 2;i !m
2
Imp2;i 2;id2;i3
5
=1
2
5
X
i=1
vihP1
U2P2
U2i(46)
_____________________________________________________________________________________________________________________________
PU2=K2
2p
5
X
i=1
2
6
6
42;i
1
X
z=0
2;i
22;i z
z!0
@22F11
2;3
2+z;3
2;1
23
2+z

3
2
2;i  (z+ 1)
+s
2
2;i 1
A2K22;i e'
2
p
1
X
m=0
1
X
g=0
2;i '
42;i m
22;i
2gm
2+g (m+ 2g)
( (g+ 1)  (m+g+ 1))20
B
B
@
22F11
2;3
2+g;3
2;1
22;i 3
2+g
22;i g+3
2
+p (g+ 1)
22;i g+1 1
C
C
A;
v = [2;1;1;1;1] :(53)
_____________________________________________________________________________________________________________________________

IEEE SYSTEMS JOURNAL, OCTOBER. 2020 12
[3] M. A. Al-Jarrah, M. A. Yaseen, A. Al-Dweik, O. A. Dobre, and
E. Alsusa, “Decision fusion for IoT-based wireless sensor networks,”
IEEE Internet Things J., vol. 7, no. 2, pp. 1313–1326, Feb. 2020.
[4] Y. Iraqi and A. Al-Dweik, “Adaptive bit loading with reduced computa-
tional time and complexity for multicarrier wireless communications,”
IEEE Trans. Aerosp. Electron. Syst., vol. 56, no. 3, pp. 2497–2506,
June 2020.
[5] J. Tooker and M. C. Vuran, “Mobile data harvesting in wireless
underground sensor networks,” in SECON, June 2012, pp. 560–568.
[6] H. Mukhtar et al., “Adaptive hybrid ARQ system using turbo product
codes with hard/soft decoding,” IEEE Commun. Lett., vol. 17, no. 11,
pp. 2132–2135, Nov. 2013.
[7] H. Mukhtar, A. Al-Dweik, and M. Al-Mualla, “Low complexity power
optimization algorithm for multimedia transmission over wireless net-
works,” IEEE J. Sel. Topics Signal Process., vol. 9, no. 1, pp. 113–124,
Feb. 2015.
[8] C. Deng, M. Liu, X. Li, and Y. Liu, “Hardware impairments aware
full-duplex NOMA networks over rician fading channels,” IEEE Syst.
J., pp. 1–4, 2020.
[9] T. Assaf, A. Al-Dweik, M. E. Moursi, and H. Zeineldin, “Exact BER
performance analysis for downlink NOMA systems over Nakagami- m
fading channels,” IEEE Access, vol. 7, pp. 134 539–134 555, 2019.
[10] D. Hu, Q. Zhang, Q. Li, and J. Qin, “Joint position, decoding order,
and power allocation optimization in UAV-based NOMA downlink
communications,” IEEE Syst. J., vol. 14, no. 2, pp. 2949–2960, June
2020.
[11] Q. He, Y. Hu, and A. Schmeink, “Closed-form symbol error rate
expressions for non-orthogonal multiple access systems,” IEEE Trans.
Veh. Technol., vol. 68, no. 7, pp. 6775–6789, July 2019.
[12] X. Yan, H. Xiao, K. An, G. Zheng, and S. Chatzinotas, “Ergodic
capacity of NOMA-based uplink satellite networks with randomly
deployed users,” IEEE Syst. J., vol. 14, no. 3, pp. 3343–3350, Sep.
2020.
[13] M. Zeng, A. Yadav, O. A. Dobre, G. I. Tsiropoulos, and H. V. Poor,
“Capacity comparison between MIMO-NOMA and MIMO-OMA with
multiple users in a cluster,” IEEE J. Sel. Areas Commun., vol. 35,
no. 10, pp. 2413–2424, Oct. 2017.
[14] F. Kara and H. Kaya, “On the error performance of cooperative-NOMA
with statistical CSIT,” IEEE Commun. Lett., vol. 23, no. 1, pp. 128–
131, Jan. 2019.
[15] M. Tian, S. Zhao, Q. Li, and J. Qin, “Secrecy sum rate optimization
in nonorthogonal multiple access AF relay networks,” IEEE Syst. J.,
vol. 13, no. 3, pp. 2712–2715, 2019.
[16] F. Kara and H. Kaya, “BER performances of downlink and uplink
NOMA in the presence of SIC errors over fading channels,” IET
Commun., vol. 12, no. 15, pp. 1834–1844, Sep. 2018.
[17] S. K. Zaidi, S. F. Hasan, and X. Gui, “Evaluating the ergodic rate in
SWIPT-aided hybrid NOMA,” IEEE Commun. Lett., vol. 22, no. 9, pp.
1870–1873, Sep. 2018.
[18] L. Bariah, S. Muhaidat, and A. Al-Dweik, “Error probability analysis
of non-orthogonal multiple access over Nakagami-mfading channels,”
IEEE Trans. Commun., vol. 67, no. 2, pp. 1586–1599, Feb. 2019.
[19] C. Xu, M. Wu, Y. Xu, and Y. Fang, “Uplink low-power scheduling for
delay-bounded industrial wireless networks based on imperfect power-
domain NOMA,” IEEE Syst. J., vol. 14, no. 2, pp. 2443–2454, June
2020.
[20] X. Wang, M. Jia, Q. Guo, I. W. Ho, and F. C. Lau, “Full-duplex relay-
ing cognitive radio network with cooperative nonorthogonal multiple
access,” IEEE Syst. J., vol. 13, no. 4, pp. 3897–3908, Dec. 2019.
[21] L. Dai, B. Wang, Z. Ding, Z. Wang, S. Chen, and L. Hanzo, “A survey
of non-orthogonal multiple access for 5G,” IEEE Commun. Surv. Tut.,
vol. 20, no. 3, pp. 2294–2323, May 2018.
[22] M. Al-Imari, M. A. Imran, R. Tafazolli, and D. Chen, “Subcarrier and
power allocation for LDS-OFDM system,” in IEEE VTC-Spring, May
2011, pp. 1–5.
[23] S. M. R. Islam, N. Avazov, O. A. Dobre, and K. Kwak, “Power-domain
non-orthogonal multiple access (NOMA) in 5G systems: Potentials and
challenges,” IEEE Commun. Surveys Tuts, vol. 19, no. 2, pp. 721–742,
Oct. 2017.
[24] J. Ding, J. Cai, and C. Yi, “An improved coalition game approach for
MIMO-NOMA clustering integrating beamforming and power alloca-
tion,” IEEE Trans. Veh. Technol., vol. 68, no. 2, pp. 1672–1687, Feb.
2019.
[25] M. S. Ali, E. Hossain, A. Al-Dweik, and D. I. Kim, “Downlink power
allocation for CoMP-NOMA in multi-cell networks,” IEEE Trans.
Commun., vol. 66, no. 9, pp. 3982–3998, Sep. 2018.
[26] J. Garnier, A. Fabre, H. Fares and R. Bonnefoi, “On the performance
of QPSK modulation over downlink NOMA: from error probability
derivation to SDR-based validation,” IEEE Access, vol. 8, pp. 66495–
66507, 2020.
[27] F. Kara and H. Kaya, “Error probability analysis of NOMA-based
diamond relaying network,” IEEE Trans. Veh. Technol., vol. 69, no. 2,
pp. 2280–2285, Feb. 2020.
[28] Z. Gao, A. Liu and X. Liang, “The performance analysis of downlink
NOMA in LEO satellite communication system,” IEEE Access, vol. 8,
pp. 93723–93 732, 2020.
[29] B. K. Ng and C. Lam, “Joint power and modulation optimization in
two-user non-orthogonal multiple access channels: A minimum error
probability approach,” IEEE Trans. Veh. Technol., vol. 67, no. 11, pp.
10693–10 703, Nov. 2018.
[30] C. Yan, A. Harada, A. Benjebbour, Y. Lan, A. Li, and H. Jiang, “Re-
ceiver design for downlink non-orthogonal multiple access (NOMA),”
in IEEE VTC-Spring, May 2015, pp. 1–6.
[31] M. Qiu, Y. Huang, S. Shieh, and J. Yuan, “A lattice-partition framework
of downlink non-orthogonal multiple access without SIC,” IEEE Trans.
Commun., vol. 66, no. 6, pp. 2532–2546, June 2018.
[32] S. Shieh, C. Lin, Y. Huang, and C. Wang, “On gray labeling for
downlink non-orthogonal multiple access without SIC,” IEEE Commun.
Lett., vol. 20, no. 9, pp. 1721–1724, Sep. 2016.
[33] J. S. Yeom, H. S. Jang, K. S. Ko, and B. C. Jung, “BER performance of
uplink NOMA with joint maximum-likelihood detector,” IEEE Trans.
Veh. Technol., vol. 68, no. 10, pp. 10295–10 300, Aug. 2019.
[34] S. Baig, U. Ali, H. M. Asif, A. A. Khan, and S. Mumtaz, “Closed-form
BER expression for Fourier and wavelet transform-based pulse-shaped
data in downlink NOMA,” IEEE Commun. Lett., vol. 23, no. 4, pp.
592–595, Apr. 2019.
[35] F. Kara and H. Kaya, “Performance analysis of SSK-NOMA,” IEEE
Trans. Veh. Technol., vol. 68, no. 7, pp. 6231–6242, July 2019.
[36] R. Sun and D. W. Matolak, “Air-ground channel characterization for
unmanned aircraft systems part II: Hilly and mountainous settings,”
IEEE Trans. Veh. Technol., vol. 66, no. 3, pp. 1913–1925, Mar. 2017.
[37] D. W. Matolak and R. Sun, “Air-ground channel characterization
for unmanned aircraft systems-part III: The suburban and near-urban
environments,” IEEE Trans. Veh. Technol., vol. 66, no. 8, pp. 6607–
6618, Aug. 2017.
[38] Morteza Banagar, Harpreet S. Dhillon, and Andreas F. Molisch, “Impact
of UAV wobbling on the air-to-ground wireless channel,” arXiv, Apr.
2020, DOI: arXiv:2004.02771.
[39] T. Assaf, A. Al-Dweik, M. S. El Moursi, H. Zeineldin, and M. Al-
Jarrah, “Exact bit error-rate analysis of two-user NOMA using QAM
with arbitrary modulation orders,” IEEE Commun. Lett., 2020, early
Access.
[40] M. Tariq, A. Al-Dweik, B. Mohammad, H. Saleh, and T. Stouraitis,
“Computational power evaluation for energy-constrained wireless com-
munications systems,” IEEE Open J. Commun. Soc., vol. 1, pp. 308–
319, Mar. 2020.
[41] F. Kara and H. Kaya, “Error probability analysis of non-orthogonal
multiple access with channel estimation errors,” arXiv, Apr. 2020, DOI:
arXiv:2004.12402.
[42] ——, “Improved user fairness in decode-forward relaying non-
orthogonal multiple access schemes with imperfect SIC and CSI,” IEEE
Access, vol. 8, pp. 97540–97 556, 2020.
[43] M. Mirahmadi, A. Al-Dweik, and A. Shami, “Interference modeling
and performance evaluation of heterogeneous cellular networks,” IEEE
Trans. Commun., vol. 62, no. 6, pp. 2132–2144, June 2014.
[44] D. Anirban, “Finite sample theory of order statistics and extremes.”
Probability for Statistics and Machine Learning : Fundamentals and
Advanced Topics, p. 221, 2011.
[45] I. Gradshteyn and I. Ryzhik, Table of Integrals, Series, and Products,
7th ed. Academic Press, 2007.
[46] M. Simon and M. Alouini, Digital Communication over Fading Chan-
nels. John Wiley and Sons, Jan. 2000.
[47] J. Craig, “A new, simple and exact result for calculating the prob-
ability of error for two-dimensional signal constellations,” in Proc.
MILCOM91, Nov. 1991.
[48] G. Gasper and M. Rahman, Basic Hypergeometric Series (Encyclopedia
of Mathematics and its Applications), 2nd ed. Cambridge University
Press, 2004.

IEEE SYSTEMS JOURNAL, OCTOBER. 2020 13
PLACE
PHOTO
HERE
Tasneem Assaf (S’14) received the B.Sc. de-
gree in commu-nications engineering from Khal-
ifa University (KU), Abu Dhabi, UAE, in 2014,
and the master’s degree in electrical engineering
from the American University of Sharjah (AUS),
Sharjah, UAE, in 2016. She is currently pursuing
the Ph.D. degree from KU. Her research interests
include smart grids, optimization, wireless commu-
nications, and machine learn-ing.
PLACE
PHOTO
HERE
Arafat Al-Dweik (S’97-M’01-SM’04) received the
M.S. (Summa Cum Laude) and Ph.D. (Magna
Cum Laude) degrees in electrical engineering from
Cleveland State University, Cleveland, OH, USA in
1998 and 2001, respectively. He was with Efficient
Channel Coding, Inc., Cleveland-Ohio, from 1999
to 2001. From 2001 to 2003, he was the Head
of Department of Information Technology at the
Arab American University in Palestine. Since 2003,
he is with the Department of Electrical Engineer-
ing, Khalifa University, United Arab Emirates. He
joined University of Guelph, ON, Canada, as an Associate Professor from
2013-2014. Dr. Al-Dweik is a Visiting Research Fellow at the School
of Electrical, Electronic and Computer Engineering, Newcastle University,
Newcastle upon Tyne, UK and he is Research Professor Western University,
London, {ON}, Canada and University of Guelph. Dr Al-Dweik has extensive
editorial experience where he served as an Associate Editor at the IEEE
Transactions on Vehicular Technology and the IET Communications. Dr
Al-Dweik has extensive research experience in various areas of wireless
communications that include modulation techniques, channel modeling and
characterization, synchronization and channel estimation techniques, OFDM
technology, error detection and correction techniques, MIMO and resource
allocation for wireless networks. Dr. Al-Dweik is a member of Tau Beta Pi
and Eta Kappa Nu. He was awarded the Fulbright scholarship from 1997
to 1999. He received the Hijjawi Award for Applied Sciences in 2003 and
the Fulbright Alumni Development Grant in 2003 and 2005, and the Dubai
Award for Sustainable Transportation in 2016. He is a Senior Member of the
IEEE, and a Registered Professional Engineer in the Province of Ontario,
Canada.
PLACE
PHOTO
HERE
Mohamed El Moursi (M’12, SM15) received the
B.Sc. and M.Sc. degrees from Mansoura University,
Mansoura, Egypt, in 1997 and 2002, respectively,
and the Ph.D. degree from the University of New
Brunswick (UNB), Fredericton, NB, Canada, in
2005, all in electrical engineering. He was a Re-
search and Teaching Assistant in the Department of
Electrical and Computer Engineering, UNB, from
2002 to 2005. He joined McGill University as
a Postdoctoral Fellow with the Power Electronics
Group. He joined Vestas Wind Systems, Arhus,
Denmark, in the Technology Rn&D with the Wind Power Plant Group.
He was with TRANSCO, UAE, as a Senior Study and Planning Engineer.
He is currently a Professor in the Electrical and Computer Engineering
Department at Khalifa University of Science and Technology- Masdar
Campus and seconded to a Professor Position in the Faculty of Engineering,
Mansoura University, Mansoura, Egypt and currently on leave. He was
a Visiting Professor at Massachusetts Institute of Technology, Cambridge,
Massachusetts, USA. Dr. Shawky is currently an Editor of IEEE Transactions
on Power Delivery, IEEE Transactions on Power Systems, Associate Editor of
IEEE Transactions on Power Electronics, Guest Editor of IEEE Transactions
on Energy Conversion, Guest Editor-in-Chief for special section between
TPWRD and RPWRS, Editor for IEEE Power Engineering Letters, Regional
Editor for IET Renewable Power Generation and Associate Editor for IET
Power Electronics Journals. His research interests include power system,
power electronics, FACTS technologies, VSC-HVDC systems, Microgrid
operation and control, Renewable energy systems (Wind and PV) integration
and interconnections.
PLACE
PHOTO
HERE
Hatem Zeineldin (M’06–SM’13) received the
B.Sc. and M.Sc. degrees in electrical engineering
from Cairo University, Giza, Egypt, in 1999 and
2002, respectively, and the Ph.D. degree in electri-
cal and computer engineering from the Uni-versity
of Waterloo, Waterloo, ON, Canada, in 2006. He
was with Smith and Andersen Electrical Engineer-
ing, Inc., North York, ON, USA, where he was
involved in projects involving distribution system
designs, protection, and distributed gen-eration. He
was a Visiting Professor with the Massachusetts
Institute of Technology, Cambridge, MA, USA. He is cur-rently a Professor
with the Khalifa University of Science and Technology, Abu Dhabi, UAE,
and is currently on leave from the Faculty of Engineering, Cairo University,
Giza. His current research interests include distribution system protec-tion,
distributed generation, and micro grids. He is currently an Editor for the IEEE
Transactions on Energy Conversion and the IEEE Transactions on Smart
Grid.
PLACE
PHOTO
HERE
Mohammad A. Al-Jarrah received his B.Sc. and
M.S. de-grees in Electrical Engineering/Wireless
Communications from Jordan University of Science
and Technology(JUST) in 2011. From 2017 to
2019, he had been working as a Lab instructor
at Khalifa University (KU), United Arab Emirates.
Currently, he is a Marie Curie early stage researcher
(ESR) and Ph.D. candidate at the Dept. of Elec-
trical and Electronic Engineering, The University
of Manchester, U.K. His re-search interests include
distributed decision fusion systems, statistical sig-
nal processing, target tracking in wireless sensor networks, RFID commu-
nications, cooperative spectrum sensing, free-space optical communications
(FSO), coopera-tive communications, and backhauling and cellular planning
for 5G cellular networks.