1188 IEEE COMMUNICATIONS LETTERS, VOL. 24, NO.

6, JUNE 2020

Bit Error Rate for NOMA Network

Mahmoud Aldababsa , Caner Göztepe , Güneş Karabulut Kurt , and Oğuz Kucur

Abstract— This letter examines the bit error rate (BER) per- Nakagami-m fading channels, their exact BER expressions
formance of downlink non-orthogonal multiple access networks are obtained in closed-form only for the special case m=1,
for binary phase-shift keying modulation. Exact BER expression i.e., Rayleigh fading channels. The BER performance of the
is derived for each user in closed-form under additive white
Gaussian noise and Rayleigh fading channels in perfect and NOMA network is analyzed in [4] when the space shift keying
imperfect successive interference cancellation (SIC) cases. Next, is used. However, the closed-form exact BER expression
in perfect SIC case, the asymptotic BER expression in a high of the considered NOMA network is only provided for the
signal-to-noise ratio (SNR) region is obtained to express the three-user scenario. Meanwhile, the exact average symbol
behavior of the network with diversity and array gains. On the error rate (SER) expressions are also obtained under Rayleigh
other hand, in imperfect SIC case, the upper bound for BER
is attained, and at high SNR values, the BER reveals an error fading channels for the downlink NOMA network when the
floor, and hence a zero diversity gain is achieved. Then, a feasible quadrature amplitude modulation (QAM) is used in [5] as
range of power allocation coefficients is found such that a well as both QAM and pulse amplitude modulation in [6].
good BER performance can be provided for each user. Finally, Nevertheless, in both works, the two-user NOMA scenario is
through simulations and software-defined radio-based real-time considered. It is noteworthy to mention that the BER of uplink
tests, analytical results are validated.
NOMA is also considered in [7] and [8]. However, through
Index Terms— Bit error rate, non-orthogonal multiple access,
binary phase shift keying, successive interference cancellation, the BER analysis in [6] and [8], by ignoring their statistics,
power allocation. the fading coefficients are assumed to be constant. Hence,
I. I NTRODUCTION the channel affects as additive white Gaussian noise (AWGN)
channel. The BER is investigated for both downlink and uplink
N ON-ORTHOGONAL multiple access (NOMA) is a
promising radio access scheme for the future wire-
less communications [1]. Different from orthogonal multiple
NOMA networks under Rayleigh fading channels in [9], where
exact and approximate closed-form expressions are derived for
access schemes, NOMA can serve multiple users by utilizing downlink and uplink, respectively, however, based on the only
the same time/code/spectrum resources. Particularly, in the two-user scenario.
power-domain NOMA, the transmitter superposes the signals The most significant constraint for the aforementioned BER
of multiple users, to which assigned different power factors and SER works is that the error rate performance is analyzed
into one signal, and the receiver applies the successive interfer- for the two-user and/or three-user NOMA networks with
ence cancellation (SIC) to detect the desired signals. Recently, specific assumptions. It is worth mentioning that NOMA
a great deal of attention has been devoted to study outage basically aims to serve as many users as possible with good
and capacity performance of the NOMA networks. However, performance, i.e., not suggested to serve only two or three
the error rate performance of NOMA networks has been users. With this motivation, in this letter, we examine the
investigated for special cases in a few works. In particular, error rate performance of the downlink NOMA network for
by claiming that exact bit error rate (BER) analysis of the any number of users in perfect and imperfect SIC cases. For
NOMA systems is intractable [2], the authors derive the both cases, the exact BER expression is derived in closed-
expression of the pairwise error probability for the downlink form for each user under AWGN and Rayleigh fading channels
NOMA network and use the union bound for the BER, which when the binary phase shift keying (BPSK) modulation is
considerably deviates from the simulation results for specific used. Furthermore, in the perfect SIC case, at the high signal-
scenarios. However, in [3], the BER performance of the to-noise ratio (SNR) values, the asymptotic BER is obtained
downlink NOMA network is studied for the quadrature phase to explain the BER performance of the NOMA network in
shift keying. Nevertheless, the exact BER expression is merely terms of its diversity and array gains. On the other hand,
derived for particular NOMA systems; two-user and three-user in the imperfect SIC case, the upper bound BER and error
scenarios. Although the authors investigate both systems over floor (EF) expressions are attained. Moreover, in order to
achieve good performance, a power allocation criterion of the
Manuscript received February 21, 2020; accepted March 10, 2020. Date NOMA network is provided according to the constellation of
of publication March 16, 2020; date of current version June 10, 2020. This
work was supported by the Scientific and Technological Research Council of
the BPSK modulation. Finally, the Monte Carlo simulation
Turkey (TÜBİTAK) with the project number EEEAG-118E274. The associate results and software-defined radio (SDR) based test-bed imple-
editor coordinating the review of this letter and approving it for publication mentation verify the derived closed-form BER expressions.
was X. Lei. (Corresponding author: Mahmoud Aldababsa.)
Mahmoud Aldababsa and Oğuz Kucur are with the Department of Electron-
The obtained asymptotic and upper bound BER expressions
ics Engineering, Gebze Technical University, 41400 Gebze, Turkey (e-mail: are also compatible with the exact ones in practically high
[email protected]; [email protected]). SNR regions.
Caner Göztepe and Güneş Karabulut Kurt are with the Department of We use the following notations throughout the letter:
Electronics and Communication Engineering, Istanbul Technical University,
34469 Istanbul, Turkey (e-mail: [email protected]; [email protected]). Ckn = nk . |.| and E[.] are absolute and expectation values,
2
Digital Object Identifier 10.1109/LCOMM.2020.2981024 respectively. nm ∼ CN (0, σm ) is zero mean complex
1558-2558 © 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: ULAKBIM UASL - Karaelmas Universitesi. Downloaded on April 06,2021 at 18:50:28 UTC from IEEE Xplore. Restrictions apply.
ALDABABSA et al.: BIT ERROR RATE FOR NOMA NETWORK 1189

2 2
additive Gaussian noise with variance σm at mth user. FX (x) Proof: Assume that σm = σ 2 , then the PDF, fnm (x) can
x2
and fX (x) are cumulative distribution function (CDF) and be written as fnm (x) = √2πσ1
e− 2σ2 . Now, given that λ+
2 m
probability density function (PDF) of random variable X, occurs. This means a combination from λ+ is transmitted.
m
respectively. Then, using fnm (x), the conditional PDF of ym given λ+ m,j
+
(ym −λ )2
m,j
II. S YSTEM M ODEL can be expressed as f (ym /λ+ m,j ) =
√ 1
2πσ2
e− 2σ2 .
In a power-domain downlink NOMA network, the base Since the error occurs when ym < 0, the error probability
station (BS) assigns distinct power levels to M users labeled when λ+ m,j is transmitted can be expressed as Pr (e/λm,j ) =
+
+
as U1 , . . . , Um , . . . , UM . The users’ power levels are opposite 0 (ym −λ
m,j
)2
ym −λ+
√ 1
2πσ2 −∞
e− 2σ2 dym . Substituting z = m,j
,
to their channel conditions. In other words, the users with good σ
+
channel conditions are allocated with low power coefficients, then dz = σ . Hence, Pr (e/λm,j ) can be rewritten
dym
 ∞ z2
while the users with bad channel conditions are allocated in terms of Q(x) = √12π x e− 2 dz as Pr (e/λ+ m,j ) =
with high power coefficients. Specifically, without loss of  λ
+

− σ m,j 2
∞ 2 λ+
generality, we assume that a1 ≥ . . . ≥ am ≥ . . . ≥ aM √ 1 − z2
= √1 − z2
= σ ).
m,j
2πσ 2 −∞
e σdz 2π λ
+
m,j
e dz Q(
opposite to |h1 |2 ≤ . . . ≤ |hm |2 ≤ . . . ≤|hM |2 , where λ+
σ
M
am is the mth user’s power coefficient with m=1 am = 1
m,j
However, the expression inside the Q(.) function can
σ
and hm is the fading channel gain between the BS and mth + λ+ + √
be rewritten in terms of and γ as m,j
βm,j = βm,j γ.
+ √
σ
user. Under NOMA concept, the  received signal at the mth +
Accordingly, Pr (e/λm,j ) = Pr (ym < 0) = Q(βm,j γ).
user is expressed as rm = hm m−1 i=1 √αi si + hm αm sm + Similarly, given that λ− −
M m is transmitted. Then, Pr (e/λm,j ) =
hm i=m+1 αi si + nm , where αi = P ai , P is transmit + √ + −
Q(βm,j γ). Next, using Pr (λm,j ) = Pr (λm,j ) = 2M1−m ,
power at the BS, si is the signal of the ith user with unit energy the conditional probability for ym given λ+
2 m and ym
and nm ∼ CN (0, σm ). The SIC is applied at the receiver given λ− can be expressed, respectively as Pr (e/λ+ m) =
to lessen the interference and yield reliable detection for the 2M −m m
+ + 1
2M −m + √
Pr (e/λ )P (λ
r m,j ) = M −m Q(β m,j γ)
2M −m 2 + √
signals of the users. In particular, the receiver of the mth user j=1 m,j j=1
− 1
cancels the interfered signals from U1 , . . . , Um−1 and treats and Pr (e/λm ) = 2M −m j=1 Q(βm,j γ). Finally, using
1
Um+1 , . . . , UM as interference. Consequently, by eliminating these expressions and Pr (λ+ −
m ) = Pr (λm ) = 2 in Pr (e) =
m−1
the term hm i=1 αi si in rm , the received signal at the mth Pr (e/λ+ + − −
m )Pr (λm )+Pr (e/λm )Pr (λm ), then, the BER for mth
user after applying the SIC process can be stated as user in AWGN is found as in (3). Now, we assume identically
M and independently distributed (i.i.d.) Rayleigh fading channel
ym = hm αm sm + hm αi si + nm . (1) coefficients. The CDF and PDF of the square of Rayleigh
i=m+1 random variable, can be defined respectively as FX (x) =
1 − e− Ω , and fX (x) = Ω1 e− Ω , where Ω = E[|X|2 ].
x x

III. P ERFORMANCE A NALYSIS

Proposition 2: The BER for the mth user in Rayleigh fading
A. Exact BER Calculation With Perfect SIC
channels can be expressed as
In the case of only AWGN, i.e., hm = 1, ym in (1) is written  ηm (ω−1)(−1)t+k CtM −m Ckm+t−1
as ym = λm + nm , where λm = αm sm + M i=m+1 αi si . P̄b,m = , (4)
i,t,k 2M −m+1 (k+1)ω
When the BPSK modulation is used for transmission, sm
 2M −m M−m m+t−1
and si in λ m are assigned as +1 or −1.  Let’s define where = and ω =
λ+ M − M  i,t,k i=1 t=0 k=0
m = αm + i=m+1 αi si and λm = −αm + i=m+1 αi si 2(k+1)
as the summation of users’ signals when sm = +1 and 1+ +
(βm,i )2 γΩ
. Note that the result obtained in (4) will be
sm = −1 are transmitted, respectively. If λ+ m,j and λm,j
−
realized on the real-time test system, which will be detailed
denote the jth combination of λ+ −
m and λm , respectively, where in Section IV.
j = 1, . . . , 2 M−m
, then there are 2M−m different Proof: The conditional error probability for the mth
+
combinations for λm expressed as user given that ϕm = |hm |2 can be expressed as
λ+ 2M −m + √
m,1 = αm + αm+1 + . . . + αM (2a) Pr (e/ϕm ) = 2M1−m i=1 Q(βm,i γϕm ). Meanwhile,
λ+ = α − α + . . . + α (2b) the PDF of ordered random variable ϕ is stated as fϕm (x) =
m,2 m m+1 M
M−m   m
... ηm t=0 (−1) t M−m
t fX (x)FX (x) m+t−1
, where ηm =
λ+m,2M −m
= αm − αm+1 − . . . − αM (2c) M !/(M − m)!(m − 1)!. Therefore, the error probability for
the mth user can be stated as
As λ− +
m,j = −λm,2M −m −j+1 , there are also 2
M−m
different  ∞
−
combinations for λm . P̄b,m = Pr (e/x)fϕm (x)dx
Proposition 1: The BER for the mth user in AWGN can be 0
−m
expressed as 2M
 M−m
ηm
2M

−m = (−1)t
1 + √ 2M−m Ω
Pb,m = M−m Q(βm,j γ). (3)  i=1 t=0
2 ∞
+ √
γx)e− Ω (1 − e− Ω )m+t−1 dx.
x x
j=1 M−m
Ct Q(βm,i
λ+ 0
+
In (3), βm,j = √m,j
P
and γ = P
σ2 is the SNR. (5)

Authorized licensed use limited to: ULAKBIM UASL - Karaelmas Universitesi. Downloaded on April 06,2021 at 18:50:28 UTC from IEEE Xplore. Restrictions apply.
1190 IEEE COMMUNICATIONS LETTERS, VOL. 24, NO. 6, JUNE 2020

With the benefit of binomial expansion and integration by

parts, the integral in (5) can be easily calculated. Finally,
substituting the obtained term into (5), the BER for the mth
user in Rayleigh fading channels can be found as in (4).
B. Exact BER Calculation With Imperfect SIC
When SIC process m−1is not being carried out perfectly,
the term em = hm i=1 αi si in rm results in a propaga-
tion error modelled as a complex Gaussian random variable,
i.e., em ∼ CN (0, σe2m ), where σe2m = ζΩ m−1 2
i=1 αi and
0 < ζ < 1 denotes error propagation factor. Now, assume
2
σm = σ 2 + σe2m and follow the same steps in poofs of Fig. 1. A picture of the designed test bed using four USRP-2943Rs. The
NOMA users are indicated as U1 , U2 , U3 and U4 , respectively.
propositions 1 and 2, then the BER for the mth user under
AWGN and Rayleigh fading channels with imperfect SIC can LabVIEW, which is one of the graphical programming lan-
+
be obtained in (3)
and (4), respectively, besides replacing βm,j guages, is used as the programming software. The same
+,sic +
by βm,j = βm,j P
P +γσe2m . computer is used for aspects of the control, configuration, and
data collection aspects of the transmitter and the receivers. The
C. Asymptotic BER, Upper Bound BER and Error Floor carrier frequency is selected as 2.40 GHz with a bandwidth
Performance results obtained for BER expression in (4), of 1 MHz. Root-raised cosine filter with filter length L = 6
does not reveal any information about the diversity order and roll-off factor α = 0.5 is used to shape the symbols.
and array gain of the NOMA network. So, the asymptotic A tight synchronization needs to be maintained between the
behavior of the BER needs to be analyzed. Using a high transmitter and the real-time receiver communications through
SNR approximation technique [10] and if γ goes to ∞ a two-step approach: coarse synchronization and fine syn-
∞ ∞
in FX (x) and fX (x), then FX (x) = Ω x
and fX (x) = chronization. For a coarse synchronization, the NI CDA-2990
1 ∞
Ω , respectively. Hence, f ϕm (x) = m
Ωm mC M m−1
x . Using 8-Channel clock distribution source is used. To maintain
fϕ∞m (x) and the approach in [10], the asymptotic BER is fine synchronization, the sample timing offset (STO) and
asym 2M −m √2πΩm Cm M +
Γ(m+ 12 )(βm,j )−2m −m the carrier frequency offset (CFO) between the transmitter
P̄b,m = i=1 2M −m γ . As obvi-
and the receivers need to be estimated and then digitally
ously, since P̄b,m can be expressed as (Ga,m γ)−Gd,m , then
asym
compensated. Pilot symbols are used for the estimation of
the diversity order Gd,m = m and the array gain Ga,m =
2M −m √2πΩm Cm M +
Γ(m+ 12 )(βm,j )−2m − 1 STO and CFO, using the approaches given in [11] at each
( i=1 2M −m
) m. Now, in the imper- of the three receivers. BPSK modulation is used for the pilot
+,sic +
fect SIC case, by replacing βm,j = βm,j P
P +γσ2 instead symbols. A fixed period of 1/5 pilot symbol/data symbol is
em
+
of βm,j asym
in P̄b,m , then the upper bound of the mth user is used during the measurements. The estimated STO and CFO
+,sic,∞ values are then digitally compensated. The pilot symbols are
obtained. In addition, when γ goes to ∞, then βm,j =
also used for SNR estimation and channel gain estimation for 3
+ P +
βm,j γσe2m and hence replacing it instead of β m,j in (4), (4) users. The channel gains are estimated by means of a least-
then the EF expression of mth user is attained. square estimator, which is often preferred in communication
systems due to its simplicity. Using the estimated channel
D. Feasible Range of Power Allocation Coefficients
coefficients, zero-forcing equalizers are used in the receivers
For practicable BPSK constellation of NOMA, the probable and the channel effect on the received signal is removed. Each
power allocation coefficient for mth user, α∗m can be obtained user has only its corresponding channel information with the
when λ+ −
m is maximized or λm is minimized. Nevertheless, transmit power coefficients. Moreover, all users except the
λ+
m is maximized if α∗
m − α −
m+1 − . . . − αM > 0 and λm is user with the worst channel apply SIC to the received signal.
minimized if −α∗m + αm+1 − . . . + αM < 0. After tedious The theoretical analysis presented in Section III is based on
calculations, α∗m can be expressed as the Rayleigh fading assumption. In the test-bed, in order to

avoid a line-of-sight link between the transmitter and the

∗ 1/2, m=1
αm > (6) receiver, an obstacle blocking the transmission is formed with
(1 − . . . − αm−1 )/2, m > 1. silicon and wood. A reflective surface is also placed in an
opposite location of the receivers. This physical configuration
IV. T EST-B ED
resulted in the statistical characteristics of Rayleigh fading
In the designed NOMA real-time test system, five channels. The average channel gains in the test-bed are mea-
USRP-2943Rs are used as the SDR nodes. One SDR node sured and scaling the transmit gain to obtain the relationship
is designated as the transmitter. The three (four) remaining between the gains of the channels used in (4) and thus,
SDR nodes serve as the receivers of 3 (4) different users. to make the correct comparison with the test-bed results,
In the three-user scenario, there is one meter distance between the relations are estimated as E[|h3 |] = 2.5317 × E[|h1 |]
all SDR nodes. Moreover, there is a distance of 0.80 meters and E[|h2 |] = 1.6582 × E[|h1 |] for three-user scenario; and
between all SDR nodes in the four-user scenario. Physical E[|h4 |] = 3.1275 × E[|h1 |], E[|h3 |] = 2.2432 × E[|h1 |] and
separation of users is enabled by the use of 5-meter RF cables. E[|h2 |] = 1.6167 × E[|h1 |] for four-user scenario.

Authorized licensed use limited to: ULAKBIM UASL - Karaelmas Universitesi. Downloaded on April 06,2021 at 18:50:28 UTC from IEEE Xplore. Restrictions apply.
ALDABABSA et al.: BIT ERROR RATE FOR NOMA NETWORK 1191

the BER of the three-user scenario considering different ζ.

According to (6), the power allocation coefficients are selected,
respectively as {a1 , a2 , a3 } = {0.7, 0.25, 0.05}. We notice that
when ζ increases from 0 to 0.001, then the BER performance
for 2nd and 3rd users decreases since SIC is not perfectly
conducted at the receiver of the 2nd and 3rd users. On the other
hand, the performance of the 1st user is still the same since
it does not use SIC to decode its signal. Besides, the upper
bounds for BER of the 2nd and 3rd users are compatible with
the exact ones in Fig. 3.

Fig. 2. The BER of four-user NOMA network versus the SNR for perfect VI. C ONCLUSION
SIC.
In this letter, the bit error behavior of the downlink NOMA
network has been analyzed over AWGN and Rayleigh fading
channels. The exact BER expression is derived in closed-
form for the BPSK modulation in perfect and imperfect SIC
cases. Then, in the high SNR regime, the behavior of the
network can be easily expressed in terms of diversity and
array gains in perfect SIC. On the other hand, in imperfect
SIC, the corresponding upper bound expression of the BER is
attained, and the BER performance reaches the EF values.
Finally, the theoretical analysis results are verified by the
Monte Carlo simulation and SDR implementation results. The
results show that the diversity gain is dominated by the user
Fig. 3. The BER of three-user NOMA network versus the SNR for imperfect order, so the users with higher-order experience notable BER
SIC. advantage over the users with the lower-order.

V. N UMERICAL R ESULTS R EFERENCES

In this section, the analytical BER expression of the NOMA [1] M. Aldababsa, M. Toka, S. Gokceli, G. K. Kurt, and O. Kucur, “A
network is verified via Monte Carlo simulation and SDR test- tutorial on nonorthogonal multiple access for 5G and beyond,” Wireless
bed implementation in Rayleigh fading channels and Ω = 1. Commun. Mobile Comput., vol. 2018, Jun. 2018, Art. no. 9713450.
[2] L. Bariah, S. Muhaidat, and A. Al-Dweik, “Error probability analysis
The BER of the four-user scenario is depicted in Fig. 2 versus of non-orthogonal multiple access over Nakagami-m fading channels,”
SNR for perfect SIC. According to (6), the power allocation IEEE Trans. Commun., vol. 67, no. 2, pp. 1586–1599, Feb. 2019.
coefficients are selected, respectively as {a1 , a2 , a3 , a4 } = [3] T. Assaf, A. Al-Dweik, M. E. Moursi, and H. Zeineldin, “Exact BER
performance analysis for downlink NOMA systems over Nakagami-m
{0.75, 0.18, 0.05, 0.02}. We observe that the simulation results fading channels,” IEEE Access, vol. 7, pp. 134539–134555, 2019.
exactly match with the corresponding ones in (4). Moreover, [4] F. Kara and H. Kaya, “Performance analysis of SSK-NOMA,” IEEE
the BER performance of the users are different since each Trans. Veh. Technol., vol. 68, no. 7, pp. 6231–6242, Jul. 2019.
[5] I.-H. Lee and J.-B. Kim, “Average symbol error rate analysis for non-
user experiences different interference and ordered as P̄b,1 > orthogonal multiple access with M -Ary QAM signals in Rayleigh
P̄b,2 > P̄b,3 > P̄b,4 since U1 suffers from interference of three fading channels,” IEEE Commun. Lett., vol. 23, no. 8, pp. 1328–1331,
users (U2 ,U3 and U4 ), U2 from that of two users (U3 and Aug. 2019.
[6] Q. He, Y. Hu, and A. Schmeink, “Closed-form symbol error rate
U4 ) and U3 from that of one user (U4 ) and U4 from no user. expressions for non-orthogonal multiple access systems,” IEEE Trans.
However, in order to make this more clear, the asymptotic BER Veh. Technol., vol. 68, no. 7, pp. 6775–6789, Jul. 2019.
asym [7] H. Haci, H. Zhu, and J. Wang, “Performance of non-orthogonal multiple
curves are plotted by using P̄b,m to depict the diversity order
access with a novel asynchronous interference cancellation technique,”
clearly. It is obviously shown that the diversity order depends IEEE Trans. Commun., vol. 65, no. 3, pp. 1319–1335, Mar. 2017.
on the order index of the user, m such that the diversity orders [8] X. Wang, F. Labeau, and L. Mei, “Closed-form BER expressions of
for the four users are Gd,1 = 1, Gd,2 = 2, Gd,3 = 3 and QPSK constellation for uplink non-orthogonal multiple access,” IEEE
Commun. Lett., vol. 21, no. 10, pp. 2242–2245, Oct. 2017.
Gd,4 = 4. The results obtained in the real-time test system are [9] F. Kara and H. Kaya, “BER performances of downlink and uplink
also shown in Fig. 2. The BER values at the six different SNR NOMA in the presence of SIC errors over fading channels,” IET
levels have a linear relationship with the theoretical curve for Commun., vol. 12, no. 15, pp. 1834–1844, Sep. 2018.
[10] Z. Wang and G. B. Giannakis, “A simple and general parameterization
the 10 - 35 dB SNR range. However, there is an increasing quantifying performance in fading channels,” IEEE Trans. Commun.,
difference between the theoretical curve and the obtained BER vol. 51, no. 8, pp. 1389–1398, Aug. 2003.
values after the 20-25 dB SNR level. Consequently, the EF, [11] D. Oh, S. Kim, S.-H. Yoon, and J.-W. Chong, “Two-dimensional
ESPRIT-like shift-invariant TOA estimation algorithm using multi-band
which is frequently encountered in real-time test systems, chirp signals robust to carrier frequency offset,” IEEE Trans. Wireless
appears as a result of the estimation processes. Fig. 3 plots Commun., vol. 12, no. 7, pp. 3130–3139, Jul. 2013.

Authorized licensed use limited to: ULAKBIM UASL - Karaelmas Universitesi. Downloaded on April 06,2021 at 18:50:28 UTC from IEEE Xplore. Restrictions apply.