Received: 8 October 2017 Revised: 14 May 2018 Accepted: 1 July 2018

DOI: 10.1002/dac.3789

RESEARCH ARTICLE

Power allocation schemes for wireless powered NOMA

systems with imperfect CSI: An application in multiple
antenna–based relay

Thanh-Luan Nguyen1 Dinh-Thuan Do2

1
Faculty of Electrical and Electronics
Engineering, Bach Khoa University, Ho Summary
Chi Minh City, Vietnam The simultaneous wireless information and power transfer or energy harvest-
2
Wireless Communications Research
ing (EH) can be combined in nonorthogonal multiple access (NOMA) as green
Group, Faculty of Electrical and
Electronics Engineering, Ton Duc Thang applications towards 5G. This paper investigates a new cooperative EH-NOMA
University, Ho Chi Minh City, Vietnam protocol, where the intermediate relay has not equipped the fixed power source
and acts as a wireless powered relay to help signal transmission to representative
Correspondence
Dinh-Thuan Do, Wireless weak user and strong user in NOMA. However, impacts of imperfect channel
Communications Research Group, Faculty estimation contribute to outage system performance evaluations. We formulate
of Electrical and Electronics Engineering,
Ton Duc Thang University, Ho Chi Minh
the power resource assignment paradigms as two schemes, namely, fixed power
City, Vietnam. allocation and dynamic power allocation, by considering imperfect channel state
Email: dodinhthuan@tdt.edu.vn information (CSI). To solve this problem, we derive the closed-form expressions
of outage probability under imperfect CSI and the power allocation constraints.
The expected numerical results related to the derived expressions for the outage
probability are examined that numerical and the Monte Carlo simulations are
strictly matching lines.

K E Y WO R D S
energy harvesting, imperfect channel state information, NOMA, outage probability

1 I N T RO DU CT ION

Radio access technologies development play important role to adapt with requirements in the next generation 5G
wireless communications which require the high data rate and connectivity. As RATs, nonorthogonal multiple access
(NOMA) novel is considered as performance enhancement application for cellular communications. The attracted exten-
sive research topic–related NOMA is considered as promising candidate of RAT applied in 5G cellular networks. In
existing wireless communication systems, traditional orthogonal multiple access (OMA) has been widely deployed. In
contrast with OMA, NOMA holds the prospective characteristic to additional expand both connectivity capability and
system spectral efficiency. Such new RAT called as NOMA is proposed to encounter the heterogeneous requirements
including enormous connections, enhanced fairness, little latency, high reliability, and high throughput. The key idea in
NOMA showed that multiple users are served in the same time slot/subcarrier/spreading code.1 Key techniques are inte-
grated with NOMA, including relaying networks, multiple-input multiple-output, space-time coding, beamforming, and
network.2 Several parameters such as user pairing and power allocation algorithms are presented in downlink of NOMA.3
It is confirmed that the crucial feature such as power allocation, decoding order selection, and user grouping in NOMA
system can be solved in reasonable schemes.4-6 In principle, water-filling power allocation is deployed in conventional

Int J Commun Syst. 2018;e3789. wileyonlinelibrary.com/journal/dac © 2018 John Wiley & Sons, Ltd. 1 of 17
https://doi.org/10.1002/dac.3789
2 of 17 NGUYEN AND DO

system while NOMA assigns more power to users for worse channel conditions and vice versa. Such principle of power
allocation is designed to identify a better trade-off in two terms including user fairness and system throughput. Fang et al7
suggested an innovative power allocation in terms of subchannels to obtain maximal energy efficiency. In addition, with
regard to multiple-input single-output applied in NOMA system, downlink sum rate maximization is calculated using
nonconvex and intractable optimization problem.8
To 5G communications come true, energy harvesting (EH) is proposed to transfer energy to wireless devices.9 It has
the ability to lengthen the energy-constrained nodes' lifetime and evade frequent recharging and battery replacement at
the same time due to several drawbacks brought (eg, super-mini medical devices inserted confidential in a human body).
The radio frequency (RF) signals in the ambient wireless have emerge as a reliable energy source apart from natural ones
like vibration, wave, and solar. It is shown that both energy and information can be transmitted by RF signals together,
from the same received RF signals; Nguyen and Do10 have put forward the simultaneous wireless information and power
transfer (SWIPT). A wireless power transfer is proposed as a promising EH technology in which users harvest energy from
electromagnetic radiation.11 The ambient electromagnetic sources transfer electromagnetic energy opportunistically, and
such scheme is deployed in SWIPT system. In SWIPT, energy can be extracted from information carrying signals and
effective SWIPT needs some fundamental alterations in wireless communication networks. For example, information
transfer rates and the reception reliability are the conventional criteria to evaluate the system performance.12 Because of
difficulties of a receiver in harvesting energy and processing information from the same signals, there were two policies
regarding the wireless EH and information processing given in Nasir et al.13 The first one is the power splitting (PS) policy,
where the received power is spitted by the receiver between EH and information processing. Time switching is the sec-
ond policy, where the time is divided between EH and information processing. Such simultaneous EH and information
processing can be applicable to cooperative relaying.14-16 Using the broadcast nature of the wireless medium, the source
transmits the information to the relay node, where the information is transferred to the destination. In this model, energy
is harvesting energy from the received RF information signal, using it afterwards to transmit the information to the des-
tination. Applied in wireless transmission, the lifetime of a relay can be lengthened by EH as well as the information
transfer, which promote the information cooperation successively.
Motivated by the necessities of 5G and the benefits of NOMA and SWIPT, the new paradigm of the SWIPT-NOMA
scheme is introduced to avoid lifetime limitation of energy-constrained relay in role of the NOMA user using wireless
power harvested from the received signals. The main characteristic is that such relay procedures harvested energy to
help information transmission to far NOMA user. Several NOMA user selection models are investigated, including ran-
dom near NOMA user and random far NOMA user selection, nearest near NOMA user and nearest far NOMA user
selection, and nearest near NOMA user and farthest far NOMA user selection, to determine system performance. Sub-
sequently, to enhance the outage performance, Do et al17 presented a best-near best-far NOMA user selection scheme.
Furthermore, the application of SWIPT to NOMA was explored in a single-input single-output scenario,18 in which three
user-paring schemes are examined via the outage performance considering on the cell-center users. Zhang et al19 con-
sidered the outage performance for EH enabled NOMA relaying networks, in which the base station (BS) determines
the transmitting antenna and the multiple users conduct maximal ratio combining to finalize received signal. In addi-
tion, the outage probability and their high signal-to-noise ratio (SNR) approximations are expressed in the closed-form to
describe the performance of EH-assisted fixed NOMA and EH-assisted cognitive radio NOMA. The two power allocation
schemes are developed as analytical results to establish different trade-offs between the system complications, reception
reliability, and user fairness.20 Regarding NOMA in relay, three schemes including amplify-forward, decode-forward, and
quantize-map-forward at the multiantenna relay user are investigated in the closed-form formula for evaluation of the
outage probability performance.21 The recent work shown several results related to a new relay selection scheme sug-
gested in condition of known channel state information (CSI) and battery status of relays.22 Two cases of single relay and
multiple relays with partial relay selection strategy are investigated how important parameters including PS ratio, EH
efficiency, and the location of relay nodes affect on outage performances.23
Most of the prior literature focusing on NOMA are investigated over perfect CSI consideration on wireless Rayleigh fad-
ing assumptions. Unfortunately, the existence of the channel estimation errors, this statement is too idealistic in NOMA
scheme. Considering on imperfect CSI, Cai et al27 investigated the power allocation scheme together with a user choice
step to minimize the outage probability constraints of all the users. However, it cannot be seen impact of imperfect CSI in
EH-NOMA as in this work. Motivated by these, this investigation shows EH-NOMA in relationship with problem of imper-
fect CSI under Rayleigh and Nakagami-m fading channels. Such system model is considered for a downlink EH-NOMA
network. The main contributions of this paper are summarized as follows:
NGUYEN AND DO 3 of 17

• This paper combines traditional NOMA with wireless powered relaying networks in imperfect CSI scenario to
improve spectral efficiency of the network, and the outage performance for the network is studied.
• To provide more selection to harvest energy, multiple antenna is equipped at relay. In addition, two wireless power
transfer mode, namely, linear EH and nonlinear EH, are carefully studied.
• As important achievement, we derive exact expressions for the outage probability in the closed-form, which is the
theoretical fundamental to optimize and introduce the real communication system design.
• Finally, the accuracy of our analytical results can be evaluated by simulation results, and the advantage of NOMA
is explored. Moreover, an error floor appears in the outage probability due to the presence of imperfect CSI even
though high SNR regime.
The rest of the paper is structured as follows. Section 2 defines the system model. Section 3 provides preliminary results
to support analysis in other sections. Sections 4 and 5 study for power allocation schemes of EH-NOMA with imperfect
CSI, where Section 4 adopts fixed power allocation (F-PA) while Section 5 uses dynamic power allocation (D-PA). In
addition, outage probability in OMA system is also derived for comparison purpose in Section 5. In Section 6, numerical
results are introduced and Monte Carlo simulations are applied to validate the exactness of the proposed expressions.
Finally, Section 7 concludes the paper.
Throughout this paper, Pr{.} denotes the probability, fZ (𝛾) and FZ (𝛾) denote the probability density function (PDF)
and the cumulative distribution function (CDF) of random variable Z, respectively, and E[.] represents the expectation
operator.

2 S Y ST E M MO D E L

Consider a downlink cooperative NOMA system, where the BS communicates with two users, the weak user (WU) and
the strong user (SU), with the aid of an EH-assisted relay (R). The two users and the BS are equipped with a single antenna
while the relay is equipped with NR receive antennas and a single transmit antenna.
In Figure 1, (hRW , dRW ) and (hRS , dRS ) are the channel gains and the distances from relay-to-WU and relay-to-SU, respec-
tively. Since the Line-of-Sight (LoS) component of the BS-to-relay links may exist, the channels from the BS to the receive
antennas of the relay are modeled as independent and not-necessarily identically distributed Nakagami-m random vari-
ables, each m(n)
BR
is the fading factor and Ω(n) BR
is the variance. Assume that the links between the relay and users follow
independent and identical flat fading Rayleigh. Further, it is worth noting that user's channel condition is determined
by both user's channel power gain and the distance from the transmitter. Hence, the term SU means the user has better
channel condition and vice versa.
The transmission from the BS to the users is divided into two time slots. In the first time slot, the BS transmits to the
relay a superimposed signal intended for both users. In NOMA, we denote the signal transmitted from the BS as
√ √
xB ≜ 𝛾PB xW + (1 − 𝛾) PB xS , (1)

FIGURE 1 System model of downlink cooperative energy harvesting–non-orthogonal multiple access. BS, base station; SU, strong user;
WU, weak user
4 of 17 NGUYEN AND DO

where PB is the transmit power of the BS, xW and xS are defined as the information signals of the WU and the SU, respec-
tively, and 𝛾 is the power allocation at the BS. Accordingly, the received signal at the n-th antenna in the relay is acquired

√ h(n)
𝑦(n)
R
≜ 1 − 𝜃xB √BR + nR , (2)
d𝛼BR

where 𝜃 ∈ [0, 1] is the PS ratio in PS protocol, nR denotes the additive white Gaussian noise with zero mean and variance
𝜎02 , h(n)
BR
is the channel fading gain between the BS the nth receive antenna in the relay, dBR denotes the distance between
BS and relay, and 𝛼 is the path-loss exponent. Note that we consider maximal ratio combining at the relay. Subsequently,
to decode the information signal of WU, the relay considers the signal of SU as interference. Let 𝜌B ≜ PB ∕𝜎02 ; thus, the
instantaneous signal-to-interference-plus-noise ratio (SINR) at the relay to detect xW is given by
𝛾 (1 − 𝜃) 𝜌B 𝛾BR
𝛾W = , (3)
(1 − 𝛾) (1 − 𝜃) 𝜌B 𝛾BR + 1
∑NR | (n) |2 𝛼 1
where 𝛾BR = n=1 |hBR | ∕dBR . If 2 log2 (1 + 𝛾W ) > RW , where RW is the target data rate for the WU, the relay can success-
| |
fully decode the signal xS . Hence, the relay can carry out successive interference cancellation (SIC) to remove xW from
the mixed signal, and therefore, the SNR at the relay to decode the signal xS is given by

𝛾S = (1 − 𝛾) (1 − 𝜃) 𝜌B 𝛾BR . (4)

Similarly, when 12 log2 (1 + 𝛾S ) > RS , where RS is the target data rate for the SU, the relay is able to detect xS successfully.
Note that the relay in this system uses decode-and-forward protocol to transmit the information signal to the users, PS
scheme in EH-NOMA is also adopted here as a well-known protocol to harvest energy from the BS.11 Consequently, to
ensure the detection of the two signals, ie, xW and xS , and harvest maximum energy from the received signal, the PS ratio,
𝜃, and the power allocation factor, 𝛾, must satisfy the under constraint
{( ) ( )}
C= 𝛾W = 22RW − 1 and 𝛾S = 22RS − 1 . (5)

Solving C with respect that 𝛾, 𝜃 ∈ [0, 1], we then simply obtain the following results:
{ }
𝜀S
𝛾 = min 1 − , 1 (6)
𝜀B
and { }
𝜀B
𝜃 = max 0, 1 − , (7)
𝜌B 𝛾BR
where 𝜀S = 22RS − 1 and 𝜀B = 22RW +2RS − 1. Therefore, the transmit power of the relay is given by

PR = 𝜂𝜃PB 𝛾BR , (8)

where 𝜂 ∈ (0, 1) denotes the EH efficiency and PB 𝛾 BR is the received signal power at the relay. Assume that all har-
vested energy is consumed by the relay and the power consumption for transmit/received circuitry is negligible. Hence,
substituting (7) into (8), the maximized transmit power at the relay is given as
{ }
PR = max 0, 𝜂 (𝜌B 𝛾BR − 𝜀B ) 𝜎02 . (9)

Remark 1. Linear energy harvesting (LEH) model is widely used in literature related to SWIPT-NOMA because of its
tractability in analysis. However, this model is not practical since the maximal harvested energy is limited by energy
havester circuitry.24-26 In Dong et al,25 a piecewise linear EH (PWLEH) model is proposed to characterize practical EH
circuit, where the a constant power is achieved if the received power at the relay is larger than a harvested power
threshold, denoted as Pth . Further, it is worth noticing that PWLEH model can be used as a tight approximation for
nonlinear EH model as in Shi et al.26 In fact, the harvested energy for PWLEH model is PRPWLEH = 𝜂𝜃 min {PB 𝛾BR , Pth },
which is difficult to approach closed-form for outage probabilities in this paper. Therefore, in Section 6, comparison
of the two models, ie, PWLEH and LEH, is given to provide some practical insights, where the results for PWLEH
NGUYEN AND DO 5 of 17

are obtained through Monte Carlo simulation. Further, the results in Sections 4 and 5 are achieved by assuming the
energy harvester operates in linear region, ie, PB 𝛾 BR < Pth or Pth → ∞.
Assume that perfect CSI of the BS-to-relay link is obtained while imperfect CSIs of the relay-to-User links are con-
sidered. This setting can be applied for relay selection systems, in which the BS selects a relay with the highest channel
estimation accuracy to support its transmission to the users, where the accuracy of channel estimation processes at
the users is unknown at the BS and needed to be analyzed. Further, it is assumed that the feedback links to the BS and
the relay and the users are instantaneous and error free, which means that this CSI is also achievable at transmitter
whatever CSI the receiver has.27 In particular, the minimum mean square error (MMSE) can be applied in channel
estimation error model. Therefore, by assuming MMSE estimation error, the channels between the relay and the users
are estimated as
hX
√ =̂ hX + e, (10)
d𝛼X

( )
where e is the channel estimation error and is modeled as e ∼  0, 𝜎e2 and ̂ hX is the estimated channel using
( 2)
MMSE and is modeled as ̂ hX ∼  0, 𝜎X , in which 𝜎X2 = d−𝛼 X
− 𝜎e
2
. Note that 𝜎e2 is defined as the correctness of
| |2 | |2
channel estimation and ̂ hX is independent with e. We also define 𝛾 RW and 𝛾 RS as 𝛾RW ≜ |̂ h | and 𝛾RS ≜ |̂ h |,
| RW | | RS |
respectively, for brevity. Further, the channel condition is then ordered as 𝛾 RW < 𝛾 RS .
After the relay successfully decodes xW and xS , the relay then adopts NOMA scheme to forward a superimposed
signal of both users as
√ √
xR ≜ 𝛽PR xW + (1 − 𝛽) PR xS , (11)

where 𝛽 is the power allocation coefficient. Hence, the received signal at the WU and SU is obtained as
( )
𝑦 W = xR ̂ hRW + e + nW , (12)

( )
𝑦 S = xR ̂hRS + e + nS , (13)

respectively, where nW and nS denote the additive white Gaussian noise at the WU and the SU, respectively.
Accordingly, since 𝛾 RW < 𝛾 RS , the instantaneous SINR of the relay-to-WU link is

𝛽𝜌R 𝛾RW
𝛾W
′
=( ) , (14)
(1 − 𝛽) 𝛾RW + 𝜎e2 𝜌R + 1

where 𝜌R ≜ PR ∕𝜎02 , the information signal xS is considered as cochannel interference at WU. At the SU, the achievable
SINR for SU to detect xW is given by
𝛽𝜌R 𝛾RS
𝛾S∕W = ( ) . (15)
(1 − 𝛽) 𝛾RS + 𝜎e2 𝜌R + 1
( )
If 12 log2 1 + 𝛾S∕W > RW , the SU is able to detect the signal xW and can successfully remove the detected information
signal from its received signal due to NOMA scheme. Therefore, the instantaneous SNR of the relay-to-SU link is
given by
(1 − 𝛽) 𝜌R 𝛾RS
𝛾S′ = . (16)
𝜌R 𝜎e2 + 1

3 PRELIMINARY MAT HEMATICS

In this section, we tend to derive the exact PDF of 𝛾 BR under a specific constrain. Recall that h(n)
BR
's are mutually indepen-
(n) | (n) |2 𝛼
dent and have nonnecessarily identical Nakagami-m distribution. Let 𝛾BR = |hBR | ∕dBR ; the exact PDF of 𝛾 BR is given in
| |
Theorem 1.
6 of 17 NGUYEN AND DO

Theorem 1. Without loss of generality, we assume that 𝜇1 ≥ 𝜇2 ≥ · · · ≥ 𝜇NR , where 𝜇n = Ω(n)

BR
∕m(n) d𝛼 n =
BR BR
1, 2, … , NR . Therefore, the exact PDF of 𝛾 BR is given by
() i ( )
∑ ∑ 𝜙i,𝑗 () z ( )
𝑗−1
𝑓𝛾BR (z) = 𝑗
z exp − , i () ∈ Z+ (17)
i=1 𝑗=1 (𝑗 − 1)!𝜇{i} 𝜇{i}

N
in which  = {𝜇n }n=1R
; ie, a vector contains all 𝜇n "s in decreasing order,  () denotes the number of distinct elements
of , and 𝜇{1} > 𝜇{2} > · · · > 𝜇{()} are the distinct elements of  in decreasing order. Let i () is the number of
∑i
elements in  identical to 𝜇{i} and i () = s (), 0 () = 0; thus, the value of i () is determined by
s=1

i ()
∑
i () = m(n)
BR
. (18)
r=i−1 ()+1

Further, 𝜙i,𝑗 () is obtained by

[ ()
]|
1 d𝑗−1 ∏ ( )−k () |
|
𝜙i,i ()−𝑗+1 () = 𝑗−1
1 + 𝜇{k} 𝜐 | . (19)
(𝑗 − 1)!𝜇{i} d𝜐𝑗−1 k=1,k≠i |
|𝜐=−1∕𝜇{i}

(n)
Proof. The PDF of 𝛾BR is given by
( )
1 (n) x
𝑓𝛾 (n) (x) = ( ) x mBR −1
exp − . (20)
BR m(n) 𝜇n
Γ m(n)
BR
𝜇 n
BR

Subsequently, to determine the PDF of 𝛾 BR , we first derive the characteristic function of 𝛾 BR , which is given by

[ ] ∏R [ (n)
] N
Φ𝛾BR (𝑗𝜔) = E e𝑗𝜔𝛾BR = E𝛾 (n) e𝑗𝜔𝛾BR
BR
n=1
(21)
∏
NR
1
= (n)
,
n=1 (1 − 𝑗𝜔𝜇n )mBR
√
where 𝑗 = −1. The last equation is achieved by adopting (20) and Jeffrey and Zwillinger.28, Equation 3.382.6 Without loss
of generality, we can rewrite Φ𝛾BR (𝑗𝜔) as
()
∏ 1
Φ𝛾BR (𝑗𝜔) = ( )i ()
i=1 1 − 𝑗𝜔𝜇{i}
(22)
()i ()
∑ ∑ 𝜙i,𝑗 ()
= ( )𝑗 ,
i=1 𝑗=1 1 − 𝑗𝜔𝜇{i}

where the second equality is obtained using partial fraction decomposition technique.28 Therefore, using (22), the
PDF of 𝛾 BR is obtained as
∞
1
𝑓𝛾BR (𝛾) = e−𝑗𝜔x Φ𝛾BR (𝑗𝜔) d𝜔
2𝜋 ∫−∞
()i () (23)
1 ∑ ∑
∞
e𝑗𝜔x
𝜙i,𝑗 ()
∫−∞ 1 + 𝑗𝜔𝜇 )𝑗
(
= d𝜔
2𝜋 i=1 𝑗=1 {i}

by adopting Jeffrey and Zwillinger.28, Equation 3.382.6

NGUYEN AND DO 7 of 17

In addition, when m(n)

BR
s are positive integers, ie, m(n)
BR
∈ Z+ (∀n), (42) becomes

()i ()mBR
(r)
( )
∑ ∑ ∑ 𝜙i,𝑗 () z
𝑓𝛾BR (z) = 𝑗
z𝑗−1 exp − . (24)
i=1 r=1 𝑗=1 (𝑗 − 1)!𝜇{i} 𝜇{i}

Remark 2. Theorem 1 can be applied to characterize the PDF of 𝛾 BR as long as i () ∈ Z+ , i = 1, · · ·,  (), which
can be easily achieved by assuming m(n)
BR
∈ Z+ (∀n). Further, Theorem 1 can also be adopted for some m(n) BR
∈ R(∀n),
(n)
(mBR ≥ 0.5) as depicted in the simulation results in Section 6.

4 F-PA S C HE ME

In this section, the relay allocates the transmit power, PR , to both users by a power allocation that is carefully chosen to
meet user requirements and is fixed as long as the required data rates remain constant. Specifically, the power allocation
factor for the F-PA scheme is independent with the user channel gains.
Further, the outage performance is examined with imperfect CSI. By fundamental performance evaluation, the outage
probability is equivalent with the event that the data rate caused by instantaneous channel realizations is less than a
threshold rate. As a result, the outage probability is a significant performance metric to establish how the predefined
quality of service (QoS) is satisfied in delay-sensitive transmission systems, where the information sent by the BS often
operates at a fixed rate, and the network throughput is defined as the coverage probability times the fixed rate.
In particular, the outage event in EH-NOMA of the SU happens when (i) the relay cannot decode the received signals
correctly or (ii) the SIC is carried out incorrectly and therefore this user cannot detect the WU's signal or (iii) the SU
cannot decode xS after successful SIC. Let 𝜀W = 22RW − 1; thus, the outage probability at the SU, SUF−PA
, is mathematically
expressed as
{ }
SUF−PA
= 1 − Pr PR > 0, 𝛾S∕W > 𝜀W , 𝛾S′ > 𝜀S . (25)

Theorem 2. For the F-PA case, the outage probability of the SU is obtained as
3 ()i ()
∑ ∑ ∑ 𝜙i,𝑗 ()
SU
F-PA
=1− (−1)k−1
k=1 i=1 𝑗=1
(𝑗 − 1)!
( ) 𝑗−1 ( ) ( )𝑗−𝓁−1 (26)
𝜀 ∑ 𝑗−1 𝜀B Δ𝓁,k (𝛽)
× exp − B
λ{i} 𝓁=0 𝓁 λ{i} 2𝓁

in which
⎧0 𝛽 ⩽ 1 − 𝜀 1+1
⎪ 1
W
𝜀S
Δ𝓁,k (𝛽) = ⎨ exp (−𝜗k 𝛿11 ) 𝓁+1 (𝜗k 𝛿12 ) 1 − 𝜀 +1 < 𝛽 < 1 − 𝜀B
, (27)
⎪ exp (−𝜗k 𝛿21 ) 𝓁+1 (𝜗k 𝛿22 ) 𝛽 ⩾ 1 − 𝜀S
W

⎩ 𝜀 B

𝜀 1 𝜀W 𝜀S 1 𝜀S
where 𝛿11 = 𝜎e2 𝛽 𝜀 +1W −𝜀 , 𝛿12 = 𝜂 𝛽 (𝜀W +1)−𝜀W
, 𝛿21 = 𝜎e2 1−𝛽 , 𝛿22 = 𝜂 1−𝛽
, 𝜗1 = 1∕𝜎RW
2
, 𝜗3 = 1∕𝜎RS
2
, 𝜗2 = 𝜗1 + 𝜗3 ,
(W ) W
( √ )𝓁 ( √ )
𝓁 (x) = 2 x∕λ{i} K𝓁 2 x∕λ{i} , in which Kv (𝛾) is the vth order modified Bessel function of the second kind.28

Proof. See Appendix A.

The outage event of the WU occurs when either (i) the relay cannot decode the received signals or (ii) the WU
cannot decode its own information signal. Therefore, the outage probability at the WU, WU
F−PA
, can be expressed as
{ }
WU
F−PA
= 1 − Pr PR > 0, 𝛾W′
> 𝜀W . (28)
8 of 17 NGUYEN AND DO

For the F-PA case, the outage probability of the WU expressed in closed-form is given by
()i () ()
∑ ∑ 𝜀B
WU
F−PA
=1− exp −𝜗2 𝛿11 −
i=1 𝑗=1
λ{i}
( )𝑗−𝓁−1 (29)
𝑗−1 ( ) 𝜙i,𝑗 ()
∑ 𝑗−1 𝜀B
× i+1 (𝜗2 𝛿12 ) .
𝓁=0
𝓁 2𝓁 (𝑗 − 1)! λ{i}

Proof. Substituting (9) and (14) into (28), we obtain

{ }
𝛽𝜌R 𝛾RW
WU
F-PA
= 1 − Pr PR > 0, ( ) > 𝜀W
(1 − 𝛽) 𝛾RW + 𝜎e2 𝜌R + 1
{ } (30)
𝜀 1
= 1 − Pr 𝛾BR > B , 𝛾RW > 𝛿11 + 𝛿12 .
𝜌B 𝜌B 𝛾BR − 𝜀B

Note that when 𝛾 RW < 𝛾 RS , the CDF of the ordered 𝛾 RW is F𝛾RW (𝛾) = 1 − exp (−𝜗2 𝛾). Hence, WU
F−PA
can be rewritten
as
()i ()
∑ ∑ 𝜙i,𝑗 ()
WU
F-PA
=1− 𝑗
i=1 𝑗=1 (𝑗 − 1)!𝜇{i}
∞ [ ( )] ( ) (31)
𝑗−1 1 𝛾
× z exp −𝜗2 𝛿11 + 𝛿12 exp − d𝛾.
∫ 𝜌B 𝛾 − 𝜀B 𝜇{i}
𝜀B ∕𝜌B

Applying the steps used in (A6) and (A7) into (30), we can easily obtain (29).

5 D-PA S C HE ME

As discussed in Section 4, the F-PA case, one has to choose the power allocation factor carefully to achieve high perfor-
mance at both users. Another approach is considered in this section, where the relay can adjust the power allocation, ie,
𝛽, regarding the channel conditions to ensure the transmission for the WU. Note that, if the relay decides to adjust 𝛽 to
ensure the performance of SU, the outage probability of the WU will further increase, and therefore, this user may not be
able to decode its own signal correctly. Subsequently, the power allocation factor has to meet the following condition:
𝛽𝜌R 𝛾RW
( ) > 𝜀W . (32)
(1 − 𝛽) 𝛾RW + 𝜎e2 𝜌R + 1

Solving the above inequality, we get

{ ( )}
𝜌R 𝛾RW − 𝜀W 𝜎e2 𝜌R + 1
𝛽 = 1 − max 0, . (33)
(1 + 𝜀W ) 𝜌R 𝛾RW

Based on (25), the outage probability at the SU for the D-PA case is defined as
{ ( ) }
1
SU
D-PA
= 1 − Pr 𝜌R > 0, 𝛾RW > 𝜀W 𝜎e2 + , 𝛾S′ > 𝜀 S . (34)
𝜌R

Theorem 3. For the D-PA case, the outage probability of the SU expressed in integral form is given by

SU
D−PA
= 1 − I − II , (35)
NGUYEN AND DO 9 of 17

where
( )𝑗−𝓁−1 ( )
𝜙i,𝑗 () ∑ ( 𝑗−1 ) 𝜀B λ{i}
() i () 𝑗−1
∑ ∑ 𝜗2 𝜀B
I = {i} 𝓁+1 (36)
i=1 𝑗=1
(𝑗 − 1)! 𝓁=0 𝓁 2𝓁 𝜂

and
() i () ( )
∑ ∑ ∑ 𝜙i,𝑗 () 𝜀
II = exp − B
X∈{RW,RS} i=1 𝑗=1
(𝑗 − 1)! λ{i}
( ) 𝜀 +𝜀 𝜀 (37)
𝑗−1 (
∑ ) 𝜀B ∕λ{i} 𝑗−𝓁−1 S W S
𝑗−1
× 𝛶X (𝜅) d𝜅,
𝓁 2𝓁 𝜎 2 ∫
𝓁=0 X 0

{ ( ) ( )} [ ]
 (𝜅) 2 X (𝜅) [ ] 𝜀S (1+𝜀W ) 1 1
where we denote ΥX (𝜅) = 𝜎e2 𝓁+1 X𝜂 + 
𝜂λ{i} 𝓁 𝜂
exp −X (𝜅) 𝜎e2 and X (𝜅) = 𝜅 𝜎X2̄
+ 𝜎X2
(𝜅 + 𝜀W ).

Proof. See Appendix A.

Note that II can further be approximated by adopting the Gaussian-Chebyshev quadrature as

() () i ()
∑ ∑ 𝜀B ∑ 𝜙i,𝑗 ()
II ≈ exp −
X∈{RW,RS} i=1
λ{i} 𝑗=1
(𝑗 − 1)!
( )𝑗−𝓁−1
(38)
𝑗−1 ( ) 𝜀B ∕λ{i}
∑ 𝜀S + 𝜀W 𝜀S ∑ 𝜋
N
| ( 2n − 1 )|
× 𝑗−1 |sin 𝜋 || 𝛶X (𝜅n ) ,
𝓁 |
𝓁=0 2𝓁 𝜎X2 2 n=1
N | 2N |
( ( ))
𝜀 +𝜀 𝜀
where 𝜅n = S 2W S 1 + cos 2n−1 2N
𝜋 .
Based on (30) and the newly defined power allocation, the outage probability at the WU for the D-PA case is defined
as
{ ( )}
1
WU
D-PA
= 1 − Pr 𝜌R > 0, 𝛾RW > 𝜀W 𝜎e2 + . (39)
𝜌R

Similarly, the outage probability at the WU expressed in closed-form can be easily obtained by

()i ()
∞ ( )
∑ ∑ 𝜙i,𝑗 () −𝜗2 𝜀W 𝜂 1
z
−𝜗2 𝜀W 𝜎e2 (𝜌B z−𝜀B ) z𝑗−1
WU
D−PA
=1−e 𝑗 ∫
e exp − dz
i=1 𝑗=1 (𝑗 − 1)!𝜇{i} 𝜇{i}
0
() i () ( )
∑ ∑ 𝜙i,𝑗 () 𝜀
=1− exp −𝜗2 𝜀W 𝜎e2 − B (40)
i=1 𝑗=1
(𝑗 − 1)! λ{i}
( ) ( )
𝑗−1 (
∑ ) 𝜀B ∕λ{i} 𝑗−𝓁−1 𝜗2 𝜀W
𝑗−1
× 𝓁+1 .
𝓁=0
𝓁 2𝓁 𝜂

Before moving to the next section, we derive the outage probability of OMA to compare with F-PA-NOMA and
D-PA-NOMA. In traditional OMA systems, the BS spends up to four time slots to transmit both xS and xW to its user.
Similar to (9), the transmission power for each user is given by
{ }
PX = max 0, 𝜂 (𝜌B 𝛾BR − 𝜀X ) 𝜎02 , (41)
10 of 17 NGUYEN AND DO

FIGURE 2 Outage probability for the F-PA case versus PB , where dBR = 5 m, dRW = 10 m, and dRS = 5 m. F-PA, fixed power allocation;
NOMA, non-orthogonal multiple access; OMA, orthogonal multiple access; SU, strong user; WU, weak user

where X ∈ {RW, RS}, in which 𝜀RW = 24RW − 1 and 𝜀RS = 24RS − 1. The achievable rate for each user in OMA is
obtained as
1
RX = (1 + 𝜑X ) , (42)
4
𝜌 𝛾X
where 𝜑X = 𝜌 𝜎X 2 +1 . Hence, the outage probability of the WU in OMA system is given by
X e

WU
OMA
= 1 − Pr {𝜑RW > 𝜀RW , PRW > 0}
{ ( ) }
1 𝜀
= 1 − Pr 𝛾RW > 𝜀RW 𝜎e2 + , 𝛾BR > RW . (43)
𝜌RW 𝜌B

The above equation holds the same form as (39); thus, after some manipulations, the closed-form of WU
OMA
is
obtained as
() i () ( )
∑ ∑ 𝜀
WU
OMA
=1− exp −𝜗2 𝜀RW 𝜎e2 − RW
i=1 𝑗=1
λ{i}
𝑗−1 ( ) 𝜙i,𝑗 () ( ) 𝑗−𝓁−1 ( )
∑ 𝑗−1 𝜀RW 𝜀RW 𝜗2
× 𝓁+1 . (44)
𝓁=0
𝓁 2𝓁 (𝑗 − 1)! λ{i} 𝜂

Similarly, the outage probability of SU in OMA system is defined as

{ ( ) }
1 𝜀RS
SU = 1 − Pr 𝛾RS > 𝜀RS 𝜎e +
OMA 2
, 𝛾BR > (45)
𝜌RS 𝜌B
and can be easily obtained in closed-form as follows:
() i () ( )
∑ 3
∑ ∑ 𝜀RS
SU
OMA
=1− (−1) k−1
exp −𝜗k 𝜀RS 𝜎e −
2
(46)
k=1 i=1 𝑗=1
λ{i}
∑𝑗−1 ( ) 𝜙i,𝑗 () ( 𝜀 )𝑗−𝓁−1 (
𝜀RS 𝜗k
)
𝑗−1 RS
×  𝓁+1 . (47)
𝓁=0
𝓁 2𝓁 (𝑗 − 1)! λ{i} 𝜂

6 NUMERICAL RESULTS

In this section, numerical examples are presented to validate the outage performance of the downlink EH-NOMA net-
work under Rayleigh fading channels with imperfect CSI. Moreover, EH-NOMA is compared with two proposed power
NGUYEN AND DO 11 of 17

FIGURE 3 Outage probability for the D-PA case versus PB , where dBR = 5 m, dRW = 10 m, and dRS = 5 m. D-PA, dynamic power allocation;
NOMA, non-orthogonal multiple access; OMA, orthogonal multiple access; SU, strong user; WU, weak user

FIGURE 4 Outage probability of F-PA-NOMA/D-PA-NOMA for imperfect CSI and perfect CSI case versus different channel estimation
error, where PB = 15 dBm. CSI, channel state information; D-PA, dynamic power allocation; F-PA, fixed power allocation; NOMA,
non-orthogonal multiple access; SU, strong user; WU, weak user

allocation schemes, where power assigned to users to satisfy user fairness. In Figures 4 to 7, the distances between the
relay to each node, the BS, the SU, and the WU, are dBR = dRS = 5 m, dRW = 15 m. The path-loss exponent 𝛼 = 2.4 and
the noise power is −95 dBm, 𝜎e2 = 0.0001 and 𝜂 = 0.8. Notice that we adopt the carrier frequency, transmit, and receive
antenna gains in Yang et al.20 The unit name for the target data rates, ie, RS and RW , is bits/s/Hz. Further, from Figures 2
to 6, the number of receive antenna in the relay is 1, in which m(1)
BR
= Ω(1)
BR
= 1.0.
In Figure 2, the outage probability versus transmit power at BS is presented in different threshold bit rate, and the
analytical curves for SU and WU are obtained from (26) and (29), respectively. First of all, the exact analytical results
and simulation results are in excellent agreement. Moreover, as the transmit power of BS in system increases, the outage
performance will be better and then reaches a floor level at high transmit power, which in this case, higher than 20 dBm.
It can be seen clearly that the outage probabilities of the SU and WU decrease as RW and RS increase implying that higher
target rate results more outage for each user. Note that, for F-PA case, the outage probability curves of WU and SU are
greatly separated, where WU F−PA
> SU
F−PA
, which causes an unbalance in the system QoS. Further, it is seen that NOMA
outperforms OMA when the power allocation is carefully chosen to meet users target data rates, ie, 𝛽 = 0.8 for RS = 1.5
(bits/s/Hz) and RW = 0.5 (bits/s/Hz). However, when 𝛽 is chosen incorrectly, NOMA results in a worse performance
compared with traditional OMA.
12 of 17 NGUYEN AND DO

FIGURE 5 Outage probability of F-PA-NOMA/D-PA-NOMA for imperfect CSI and perfect CSI case versus different PB (dBm). CSI,
channel state information; D-PA, dynamic power allocation; F-PA, fixed power allocation; NOMA, non-orthogonal multiple access; SU,
strong user; WU, weak user

FIGURE 6 Outage probability of fixed power allocation–non-orthogonal multiple access with PWLEH and LEH model. LEH, linear
energy harvesting; PWLEH, piecewise linear energy harvesting; SU, strong user; WU, weak user

Figure 3 illustrates the impact of the transmit power at BS on the outage performance where the analytical curves are
obtained from (35) for SU and (40) for WU. The same trend as F-PA case is observed in this figure, where the outage
performance decreases with higher transmit power. Note that the D-PA case focuses to improve the performance of the WU,
hence the outage probability of WU decreased but increase that of SU (WU F−PA
< SUF−PA
). As a result, the system QoS of the
D-PA case is more balanced than that of F-PA case, where the outage curves of SU and WU are closer to each other. Further,
it is seen that the D-PA-NOMA provides better performance for the WU than OMA since it is crucial for D-PA-NOMA
to ensure successful decoding for this user. However, for the SU, OMA may result a better outage performance than
D-PA-NOMA especially when the required target data rates are relatively low, eg, RW = 0.5 (bits/s/Hz) and RS = 1.0
(bits/s/Hz).
Figures 4 and 5 demonstrate the outage probability in EH-NOMA systems for imperfect CSI and perfect CSI case.
Figure 4 is plotted as a function of the channel estimation error while Figure 5 is drawn in terms of transmit power PB . It
can be observed from Figure 4, with imperfect CSI at high error level, the outage probabilities reach floor values, where
𝜎e2 = 0 (perfect CSI), and EH-NOMA achieves no diversity gain. The main reason is that the channel estimation error
contributesas a source of interference. The outage probability of the WU for the F-PA case is higher than that for the D-PA
case (𝛽 = 0.8), where the SU in D-PA case is more likely to experience outage than in F-PA case.
NGUYEN AND DO 13 of 17

FIGURE 7 Outage probability for the F-PA and D-PA as a function of base station transmit power, PB (dBm), with different number of
receive antennas. D-PA, dynamic power allocation; F-PA, fixed power allocation; NOMA, non-orthogonal multiple access; SU, strong user;
WU, weak user

Figure 6 gives comparison of PWLEH and LEH models where the results of PWLEH are obtained through Monte
Carlo simulation. F-PA-NOMA scheme is adopted here; however, it should be noticed that D-PA NOMA also pro-
vides similar results. It can be seen that PWLEH and LEH models achieve the same results when the saturation
threshold is relatively large, eg, 10 dBm, and/or the BS transmits with low signal power since the energy harvester
operates in linear region with these settings. However, when the threshold is low, eg, −20 dBm, the relay can oper-
ate in nonlinear region; thus, the actual outage probability obtained from PWLEH model can be higher than that of
LEH model.
Figure 7 shows outage probability of each user in F-PA-NOMA (the left figure) and D-PA NOMA (the right figure).
{ }NR { }NR { }NR { }NR
In this figure, m(n)
BR n=1
≡ Ω(n) BR n=1
= (1.0) for NR = 1; m(n) BR n=1
= (1.0, 1.0) and Ω(n)
BR n=1
= (1.0, 1.5) for
{ }NR { }NR
NR = 2; and m(n) BR
= (1.0, 1.0, 0.6, 1.4) and Ω (n)
BR
= (1.0, 1.0, 1.2, 2.8) for NR = 4. In low transmit power
n=1 n=1
region, as we increase the number of receive antenna in the relay, the outage performance is significantly improved.
However, when the BS transmits with high power, eg, 20 dBm, further increase NR does not affect the outage probabil-
ities of both users in both cases, ie, F-PA-NOMA and D-PA NOMA, which can be observed by the floor values in the
two figures.

7 CO N C LU S I O N

In this paper, the outage performance of EH-NOMA networks with imperfect CSI is examined in downlink. To offer
valuable strategies in the 5G communication system design, we derived several expressions for the exact of the out-
age probability, which are obtained in closed-form. Simulation results demonstrate outage performance in EH-NOMA
through considering impacts of transmit power of BS, channel estimation errors, noise. It is observed that an outage floor
appears in the outage probability due to the channel estimation errors. In addition, it is proved that F-PA-NOMA achieves
better performance for the SU than D-PA-NOMA but poor balance in QoS for both users. In addition, D-PA-NOMA
provides higher success probability for the WU than F-PA-NOMA and better for both users. It is also confirmed that
14 of 17 NGUYEN AND DO

F-PA-NOMA outperforms traditional OMA. Further, the outage performance of both users can be improved by increasing
the number of receive antenna in the relay.

ORCID

Dinh-Thuan Do http://orcid.org/0000-0003-2072-069X

REFERENCES
1. Ding Z, Lei X, Karagiannidis GK, Schober R, Yuan J, Bhargava VK. A survey on non-orthogonal multiple access for 5G networks: research
challenges and future trends. IEEE J Sel Areas Commun. 2017;35(10):2181-2195.
2. Islam SMR, Avazov N, Dobre OA, Kwak K-S. Power-domain non-orthogonal multiple access (NOMA) in 5G systems potentials and
challenges. IEEE Commun Surv Tutorials. 2017;19(2):721-742.
3. Islam SMR, Zeng M, Dobre OA, Kwak K-S. Resource allocation for downlink NOMA systems: key techniques and open issues. arxiv.org/
abs/1801.00121 [Online].
4. Shi S, Yang L, Zhu H. Outage balancing in downlink nonorthogonal multiple access with statistical channel state information. IEEE Trans
Wirel Commun. 2016;15(7):4718-4731.
5. Di B, Song L, Li Y. Sub channel assignment, power allocation, and user scheduling for non-orthogonal multiple access networks. IEEE
Trans Wirel Commun. 2016;15(11):7686-7698.
6. Lei L, Yuan D, Ho CK, Sun S. Power and channel allocation for non orthogonal multiple access in 5G systems: tractability and computation.
IEEE Trans Wirel Commun. 2016;15(12):8580-8594.
7. Fang F, Zhang H, Cheng J, Leung VCM. Energy efficient resource allocation for downlink non orthogonal multiple access network. IEEE
Trans Commun. 2016;64(9):3722-3732.
8. Hanif MF, Ding Z, Ratnarajah T, Karagiannidis GK. A minorization maximization method for optimizing sum rate in the downlink of
non orthogonal multiple access systems. IEEE Trans Signal Process. 2016;64(1):3722-3732.
9. Do D-T, Nguyen H-S. A tractable approach to analyze the energy-aware two-way relaying networks in presence of co-channel interference.
EURASIP J Wirel Commun Netw. 2016;271.
10. Nguyen X-X, Do D-T. Maximum harvested energy policy in full-duplex relaying networks with SWIPT. Int J Commun Syst (Wiley).
2017;30(17):e3359.
11. Krikidis I, Timotheou S, Nikolaou S, Zheng G, Derrick WKN, Robert S. Simultaneous wireless information and power transfer in modern
communication systems. IEEE Commun Mag. 2014;52(11):104-110.
12. Ding Z, Zhong C,Wing KND, et al. Application of smart antenna technologies in simultaneous wireless information and power transfer.
IEEE Commun Mag. April 2015;53(4):86-93.
13. Nasir AA, Zhou X, Durrani S, Kennedy RA. Wireless-powered relays in cooperative communications: time-switching relaying protocols
and throughput analysis. IEEE Trans Commun. 2015;63(5):1607-1622.
14. Nguyen T-L, Do D-T. A new look at AF two-way relaying networks: energy harvesting architecture and impact of co-channel interference.
Annals of Telecommunications. 2017;72:669-678.
15. Yang Z, Ding Z, Fan P, Karagiannidis GK. Outage performance of cognitive relay networks with wireless information and power transfer.
IEEE Trans Veh Technol. May 2016;65(5):3828-3833.
16. Nguyen NT, Do D-T, Tran TP, Voznak M. Time switching for wireless communications with full-duplex relaying in imperfect CSI
condition. KSII Trans Internet Inf Syst. 2016;10(9):4223-4239.
17. Do NT, da Costa DB, Duong TQ, An B. A BNBF user selection scheme for noma-based cooperative relaying systems with SWIPT. IEEE
Commun Lett. 2016;11(99):1-5.
18. Liu Y, Ding Z, Elkashlan M, Poor HV. Cooperative non-orthogonal multiple access with simultaneous wireless information and power
transfer. IEEE J Sel Areas Commun. 2016;34(4):938-953.
19. Zhang Y, Ge J. Performance analysis for non-orthogonal multiple access in energy harvesting relaying networks. IET Commun;
11(11):1768-177.
20. Yang Z, Ding Z, Fan P, Al-Dhahir N. The impact of power allocation on cooperative non-orthogonal multiple access networks with SWIPT.
IEEE Trans Wirel Commun. 2017;16(7):4332-4343.
21. Ashraf M, Shahid A, Jang J-W, Lee K-G. Energy harvesting non-orthogonal multiple access system with multi-antenna relay and base
station. Access IEEE. 2017;5:17660-17670.
22. Ha D-B, Nguyen QS. Outage performance of energy harvesting DF relaying NOMA networks. Mobile Networks and Applications. 2017:1-14.
(Online First).
23. Guo W, Wang Y. Cooperative non-orthogonal multiple access with energy harvesting, information (MDPI).
24. Boshkovska E, Ng D, Zlatanov N, Schober R. Practical non-linear energy harvesting model and resource allocation for SWIPT systems.
IEEE Commun Lett. Dec. 2015;19:2082-2085.
25. Dong Y, Hossain M, Cheng J. Performance of wireless powered amplify and forward relaying over Nakagami-m fading channels with
nonlinear energy harvester. IEEE Commun Lett. Apr. 2016;20(4):672-675.
NGUYEN AND DO 15 of 17

26. Shi L, Zhao L, Liang K, Chu X, Wu G, Chen H-H. Profit maximization in wireless powered communications with improved non-linear
energy conversion and storage efficiencies. In: 2017 IEEE International Conference on Communications; 2017:1-6.
27. Cai W, Chen C, Bai L, Jin Y, Choi J. User selection and power allocation schemes for downlink NOMA systems with imperfect CSI. In: in
Proc. of IEEE 84Th Vehicular Technology Conference (VTC-Fall); 2016; Montreal, QC, Canada.
28. Jeffrey A, Zwillinger D, eds. Table of Integrals, Series, and Products. New York: Academic Press; 2007.

How to cite this article: Nguyen TL, Do DT. Power allocation schemes for wireless powered NOMA sys-
tems with imperfect CSI: An application in multiple antenna–based relay. Int J Commun Syst. 2018;e3789.
https://doi.org/10.1002/dac.3789

APPENDIX A: PROOF OF THEOREM 2

From (25), the outage probability of the SU can be expressed as

{
(1 − 𝛽) 𝜌R 𝛾RS
SU
F-PA
= 1 − Pr PR > 0, > 𝜀S ,
𝜌R 𝜎e2 + 1
}
𝛽𝜌R 𝛾RS
( ) > 𝜀W (A1)
(1 − 𝛽) 𝛾RS + 𝜎e2 𝜌R + 1
{ [ ] }
𝛿12 𝛿22 𝜀
= 1 − Pr 𝛾RS > max 𝛿11 + , 𝛿21 + , 𝛾BR > B .
𝜌B 𝛾BR − 𝜀B 𝜌B 𝛾BR − 𝜀B 𝜌B

Since 𝛾 RS < 𝛾 RW , the CDF function of ordered 𝛾 RS can be easily achieved as

∑
3
F𝛾RS (𝛾) = 1 − (−1)k−1 exp (−𝜗k 𝛾) . (A2)
k=1

1
Substituting (A2) into (A1), the outage probability SU
F−PA
when 𝛽 > 1 − 𝜀W +1
can be expressed as

3 ()i ()
∑ ∑ ∑ 𝜙i,𝑗 ()
SU
F-PA
=1− (−1)k−1 𝑗
k=1 i=1 𝑗=1 (𝑗 − 1)!𝜇{i}
( [ ] ) (A3)
∞
𝑗−1 𝛿12 𝛿22 z
× z exp −𝜗k max 𝛿11 + , 𝛿21 + − d𝛾.
∫𝜀B ∕𝜌B 𝜌B 𝛾 − 𝜀B 𝜌B 𝛾 − 𝜀B 𝜇{i}

Using the change of variable 𝜅 → 𝜌B 𝛾 − 𝜀B , SU

F−PA
can be rewritten as
3 () i ()
∑ ∑ ∑ 𝜙i,𝑗 ()
SU
F-PA
=1− (−1)k−1
k=1 i=1 𝑗=1 (𝑗 − 1)!λ𝑗{i}
( ) ∞ ( )
𝜀 𝜅 (A4)
× exp − B (𝜅 + 𝜀 B )𝑗−1 exp −𝜗k Υmax (𝜅) − d𝜅 ,
λ{i} ∫ λ{i}
0
⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟
𝑗
[ ]
𝛿12 𝛿22
where Υmax (𝜅) = max 𝛿11 + 𝜅
, 𝛿21 + 𝜅
and λ{i} = 𝜇{i} 𝜌B . To evaluate (A4), we consider two cases for Υmax (𝜅). The
𝛿12 𝛿22 𝜀S 𝜀S
first case occurs when 𝛿11 + 𝜅
> 𝛿21 + 𝜅 , ie, 𝛽 < 1− 𝜀B
, where the second case happens when 𝛽 ≥ 1 − 𝜀B
. Note that
𝜀S 1
1− 𝜀B
>1− 𝜀W +1
as
( ) ( )
𝜀 1 𝜀W
1− S − 1− = > 0. (A5)
𝜀B 𝜀W + 1 𝜀B (𝜀W + 1)
16 of 17 NGUYEN AND DO

1 𝜀S
Therefore, we then derive 𝑗 for the first case, ie, 1 − 𝜀W +1
<𝛽 <1− 𝜀B
, as follows:
∞ ( )
1 𝜅
𝑗 = exp (−𝜗k 𝛿11 ) (𝜅 + 𝜀B )𝑗−1 exp −𝜗k 𝛿12 − d𝜅
∫0 𝜅 λ{i}
𝑗−1 ( ) ( )
(a) ∑ 𝑗−1
∞
𝑗−𝓁−1 𝓁 1 𝜅
= 𝜀 exp (−𝜗 k 𝛿11 ) 𝜅 exp −𝜗 k 𝛿12 − d𝜅 (A6)
𝓁 B ∫0 𝜅 λ{i}
𝓁=0
𝑗−1 (
( √ )
(b) ∑ 𝑗−1
) ( ) 𝓁+1
𝜗k 𝛿12
𝑗−𝓁−1
= 𝓁
𝜀B exp (−𝜗k 𝛿11 ) 2 𝜗k 𝛿12 λ{i} 2
K𝓁+1 2 ,
𝓁=0
λ{i}
( )
n
where (a) is due to binomial expansion, k
denote the binomial coefficients, and (b) is achieved with the help of Guo
𝜀S
and Wang. 23, Equation 3.471.9
Similarly, the second case for 𝑗 , ie, 𝛽 ≥ 1 − 𝜀B
, can be obtained as

𝑗−1 (
( √ )
∑ ) ( ) 𝓁+1 𝜗k 𝛿22
𝑗 = 𝑗−1
𝓁
𝜀𝑗−𝓁−1
B
exp (−𝜗k 𝛿21 ) 2 𝜗k 𝛿22 λ {i}
2
K𝓁+1 2 . (A7)
𝓁=0
λ{i}

Substituting (A7) and (A6) into (A4) with respect to 𝛽, after some mathematical manipulations, the theorem is then
proved.

APPENDIX B: PROOF OF THEOREM 3

From (33), the outage probability of the SU in (34) can be further obtained as
{ ( )}
𝜀 1
SU
D-PA
= 1 − Pr 𝛾RS > 𝛾RW , 𝛾BR > B , 𝛾RW > 𝜀B + 𝜎e2
𝜌B 𝜌R
⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟
≜I
{ ( )
𝜀S 𝜌R 𝜎e2 + 1 (1 + 𝜀W ) 𝛾RW
− Pr 𝛾RS > ( ) ,
𝜌R 𝛾RW − 𝜌R 𝜎e2 + 1 𝜀W
( ) ( )}
𝜀 1 1
𝛾BR > B , 𝜀W + 𝜎e2 < 𝛾 RW < 𝜀B + 𝜎e2 . (B1)
𝜌B 𝜌R 𝜌R
⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟
≜II

Note that the above equality is achieved using the fact that 𝜀S + 𝜀W + 𝜀S 𝜀W = 𝜀B . Since 𝛾 RS > 𝛾 RW , the joint PDF of
𝛾 RS and 𝛾 RW is given by20
( ) ( )
∑ 1 x 1 𝑦
𝑓𝛾RW , 𝛾RS (x, 𝑦) = exp − 2 exp − 2 , (0 < x < 𝑦) . (B2)
X∈{RW,RS} 𝜎X 𝜎X 𝜎X2̄ 𝜎X̄
2

Substituting (B2) into (B1), the first probability in (B1) is calculated as follows:
() i ()
∑ ∑ ∑ 𝜙i,𝑗 () 1
∞ ∞
I = z𝑗−1 e−𝜗2 x e z∕𝜇{i} dxdz
X∈{RW,RS} i=1 𝑗=1
𝑗
( 𝑗 − 1)!𝜇{i} 𝜎X2 ∫𝜀B ∕𝜌B ∫ 𝜀B
𝜂(𝜌 z−𝜀 )
+𝜎e2 𝜀B
B B

() i () ( )
∑ 1 ∑ ∑ 𝜙i,𝑗 () ∞
𝜗𝜀 𝜅
= {i} (𝜅 + 𝜀B )𝑗−1 exp − 2 B − d𝜅
X∈{RW,RS} X 𝜗2
𝜎 2
i=1 𝑗=1 ( 𝑗 − 1)!λ𝑗{i} ∫0 𝜂 λ{i}
() i ()
∑ 1 ∑ ∑ 𝜙i,𝑗 ()
= {i}
X∈{RW,RS} 𝜎X 𝜗2 i=1 𝑗=1
2 ( 𝑗 − 1)!
( ) ( √ )𝓁+1 ( √ )
𝑗−1 (
∑ ) 𝜀B ∕λ{i} 𝑗−𝓁−1 𝜗2 𝜀B 𝜗2 𝜀B
𝑗−1
× 2 K𝓁+1 2 , (B3)
𝓁=0
𝓁 2𝓁 𝜂λ{i} 𝜂λ{i}
NGUYEN AND DO 17 of 17

[ )]
𝜀 (
where {i} = exp − λ B 𝜗2 𝜎e2 λ{i} + 1 . Accordingly, the second probability in (B1) is obtained as follows:
{i}

( )
1
∞ 𝜀B 𝜎e2 + 𝜂 ∞
(𝜌B z−𝜀B )
II = ( ) 𝑓𝛾RW ,𝛾RS (x, 𝑦) dxd𝑦𝑓𝛾BR (z) dz
∫ 𝜀B ∫𝜀W 𝜎e2 + 𝜂 1 ∫Θ(x,z)
𝜌B (𝜌B z−𝜀B )
( )
𝜀 ∞ 𝜀B 𝜎e2 + 𝜗1
−λB 𝑗−1 (
∑ () i ()
∑ ∑ 𝜙i,𝑗 ()e ) 𝑗−𝓁−1
1 ∑ 𝑗−1 𝜀B
x
{i} 𝜗 − Δ(𝜗,x)
2 −
(a) 𝓁 − 𝜂𝜇{i} 𝜎2
= 𝜗 e e 𝜎
X̄ X dxd𝜗
X∈{RW,RS} i=1 𝑗=1 ( 𝑗 − 1)!λ𝑗{i} 𝜎X2 𝓁=0 𝓁 𝜂 𝓁+1 ∫ (
∫ )
0
𝜀W 𝜎e2 + 𝜗1
𝜀
() i () −λB
(b) ∑ ∑ ∑ 𝜙i,𝑗 ()e {i}
1
=
X∈{RW,RS} i=1 𝑗=1 (𝑗 − 1)!λ𝑗{i} 𝜎X2
[
∑ ( 𝑗−1 ) 𝜀𝑗−𝓁−1
𝑗−1 ∞
− 𝜂λ𝜗
𝜀S +𝜀W 𝜀S
𝜗𝜎e2 + 1
× B
𝜗𝓁 e {i}
𝓁 𝜂 𝓁+1 ∫0 ∫0 𝜗
𝓁=0
( )]
𝜀S (1 + 𝜀W ) 𝜗𝜎e2 + 1 1 𝜗𝜎e2 + 1
× exp − (𝜅 + 𝜀W ) − (𝜅 + 𝜀W ) d𝜅d𝜗, (B4)
𝜅𝜎X2̄ 𝜗 𝜎X2 𝜗
𝜀S (1+𝜀W )(𝜂(𝜌B z−𝜀B )𝜎e2 +1)x 𝜀S (1+𝜀W )x
where we define Θ(x, z) = 𝜂(𝜌B z−𝜀B )x−𝜀W (𝜂(𝜌B z−𝜀B )𝜎e2 +1)
and Δ(𝜗, x) = 𝜗 for brevity. Note that (a) is obtained by letting
x−𝜀W
𝜗𝜎e2 +1
𝜗
𝜗 = 𝜂 (𝜌B 𝜑BR − 𝜀B ) 𝜎02 = 𝜌R while as in (b), 𝜅 = 𝜗𝜎e2 +1
x − 𝜀W . The above equations can further be expressed as follows:

𝜀
−λB 𝑗−1 (
∑ ()i ()
∑ ∑ 𝜙i,𝑗 () e ) 𝑗−𝓁−1
{i}
1 ∑ 𝑗−1 𝜀B
II =
X∈{RW,RS} i=1 𝑗=1 (𝑗 − 1)!λ𝑗{i} 𝜎X2 𝓁=0 𝓁 𝜂 𝓁+1
𝜀S +𝜀W 𝜀S ∞ ( )⎤
[ ] ( )
1 1 𝜗 ⎥ d𝜗d𝜅,
× exp −X (𝜅) 𝜎e2 𝜗𝓁 𝜎e2 + exp −X (𝜅) − (B5)
∫ ∫ 𝜗 𝜗 𝜂λ{i} ⎥
0 0 ⎦
where (c) is due to Jeffrey and Zwillinger.28, Equation 3.471.9 Substituting (B5) and (B3) into (B1), the proof is done.