See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/318812721

Energy-efficient resource scheduling for NOMA

systems with imperfect channel state
information

Conference Paper · May 2017

DOI: 10.1109/ICC.2017.7996360

CITATIONS READS

0 53

4 authors, including:

Fang Fang Julian Cheng

University of British Columbia - Okanagan University of British Columbia - Okanagan
4 PUBLICATIONS 32 CITATIONS 220 PUBLICATIONS 2,179 CITATIONS

SEE PROFILE SEE PROFILE

Victor C. M. Leung
University of British Columbia - Vancouver
1,149 PUBLICATIONS 15,203 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Cognitive Platform for Ubiquitous Cloud-based Gaming View project

Coordination models for building Fog Computing applications. View project

All content following this page was uploaded by Fang Fang on 25 August 2017.

The user has requested enhancement of the downloaded file.

 Energy-Efficient Resource Scheduling for NOMA
Systems With Imperfect Channel State Information
Fang Fang† , Haijun Zhang‡ , Julian Cheng† , Victor C.M. Leung§
† School of Engineering, The University of British Columbia, Kelowna, BC, Canada
‡ Beijing
Engineering and Technology Research Center for Convergence Networks and Ubiquitous Services,
University of Science and Technology Beijing, Beijing, China
§ Department of Electrical and Computer Engineering, The University of British Columbia, Vancouver, BC, Canada

Email: † [email protected], ‡ [email protected], † [email protected], § [email protected]

Abstract—Non-orthogonal multiple access (NOMA) is consid- first derived the outage probability for downlink single-cell
ered as a promising technology for the fifth generation mobile NOMA systems with perfect channel state information (CSI).
communications. Energy-efficient resource allocation scheme is In [7], the outage performance was further studied with the
studied for a downlink NOMA wireless network, where multiple
users can be multiplexed on the same subchannel by applying assumption of two types of partial CSI in the NOMA system.
successive interference cancellation technique at the receivers. Recently, the energy consumption of wireless networks is
Most previous works focus on resource allocation for sum rate rapidly increasing. Therefore, the improvement of energy effi-
maximization with perfect channel state information (CSI) in ciency is an important and practical consideration in wireless
NOMA systems. We formulate the energy-efficient resource allo- communication systems.
cation as a probabilistic mixed non-convex optimization problem
by considering imperfect CSI. To solve this problem, we decouple In [8], [9], an energy-efficient subchannel allocation al-
it into user scheduling and power allocation sub-problems. We gorithm was proposed for a simple downlink NOMA net-
propose a low-complexity suboptimal user scheduling algorithm work with perfect CSI. However, perfect CSI is normally
and a power allocation scheme to maximize the system energy difficult to achieve in practice due to channel estimation
efficiency under the maximum transmitted power limit, imperfect
errors, limited feedback and quantization errors [10]. In this
CSI and the outage probability constraints. Simulation results
are provided to show that the proposed algorithms yield much paper, we focus on the energy-efficient resource allocation
improved energy efficiency performance over the conventional in a downlink NOMA network and use bits per Joule to
orthogonal frequency division multiple access scheme. measure the energy efficiency performance of the system. The
resource allocation scheme is designed based on imperfect
I. I NTRODUCTION
CSI. In this situation, the user data rate may not meet the
Orthogonal frequency division multiple access (OFDMA) minimum data rate requirement, determined by quality of
has been widely adopted in the fourth generation (4G) mobile service (QoS). Therefore, an outage probability requirement is
communication systems. However, in the next generation mo- considered for the resource allocation to maximize the system
bile communication systems, the demand for mobile traffic da- energy efficiency. By formulating user scheduling and power
ta volume is expected to be 1,000 times larger. As a result, non- allocation as an energy efficiency optimization problem, a low-
orthogonal multiple access (NOMA) was proposed to meet complexity suboptimal user scheduling scheme is proposed
overwhelming requirement of data rates [1]. By employing to achieve the maximum of the system energy efficiency. A
successive interference cancellation (SIC) technique, NOMA gradient assisted binary search algorithm is utilized for the
can achieve higher data rates than the traditional OFDMA power allocation accross the subchannels to further maximize
system [2]. Therefore, NOMA is considered as a promising the energy efficiency of the system.
candidate for the fifth generation (5G) mobile communication
systems [3]. II. S YSTEM M ODEL AND P ROBLEM F ORMULATION
Since the introduction of NOMA that shows user through-
put improvement [1], NOMA has attracted much research We consider a downlink single-cell NOMA network with
attention. The resource allocation for NOMA systems was one basestation (BS) and a number of paring users. The BS
investigated in [4], [5]. In [4], a greedy subchannel and transmits its signals to M user equipments (UEs) through N
power allocation algorithm was proposed to maximize the subchannels (SCs). We denote m as the index for the mth
data rate for OFDMA based NOMA systems. A cooperative mobile user where m ∈ {1, 2, · · · , M }, and denote n as the
NOMA transmission scheme with fixed choices of power index for the nth subchannel where n ∈ {1, 2, · · · , N }. In
allocation coefficients was presented in [5]. Moreover, the this system model, the BS is located in the cell center and
outage performance of NOMA systems was studied in [6], M users are uniformly distributed in a circular region. The
[7] for a cellular downlink scenario with randomly deployed total bandwidth of the system, BW , is equally divided into
mobile users. By assuming that randomly deployed users N subchannels where the bandwidth of each subchannel is
were uniformly distributed within a cell, the authors in [6] Bsc = BW/N .
A. Signal and Channel Model C. Optimization Problem Formulation
Assume that the feedback to the transmitter is instantaneous In this subsection, we formulate the energy-efficient user
and error free in this system. That means the received CSI at scheduling policy, U = {U Em , ∀m ∈ {1, 2, · · · , M }}, and
the receiver can be known by the transmitter [11], [12]. By power allocation scheme, P = {pn > 0, ∀n ∈ {1, 2, · · · , N }},
using the minimum mean square error channel estimation error as an optimization problem in the NOMA system. Due to the
model, we mathematically describe the small scale fading of decoding complexity of the receivers, we consider only two
the link between the BS and U Em on the nth subchannel users multiplexed on one subchannel, with |ĥ1,n |2 ≥ |ĥ2,n |2
(SCn ) as corresponding to U E1 and U E2 on SCn . The sum rate of the
hm,n = ĥm,n + em,n (1) NOMA system can be described as

where hm,n is the realistic Rayleigh fading channel gain on ∑

N
R(U, P) = Rn (U, P) (4)
SCn from the BS to U Em . Here hm,n ∼ CN (0, σh2 ) is a
n=1
complex Gaussian distribution with mean zero and variance
σh2 where σh2 captures both path-loss and shadowing effects. In where
( )
(1), we denote ĥm,n ∼ CN (0, σĥ2 ) as the estimate of hm,n and |ĥ1,n |2 β1,n pn
em,n ∼ CN (0, σe2 ) as the estimation error with σĥ2 = σh2 − σe2 . Rn (U, P) =Bsc log2 1 +
σe2 pn + σn2
We assume that ĥm,n and em,n are uncorrelated to each other. ( )
|ĥ2,n |2 β2,n pn
Considering Mn users are multiplexed on SCn , without loss + Bsc log2 1 + .
of generality, we assume that the estimated channel gains in the |ĥ2,n |2 β1,n pn + σe2 pn + σn2
cell are sorted as |ĥ1,n | > |ĥ2,n | > · · · > |ĥMn ,n |. Therefore, (5)
the received signal at U Em on SCn is For energy-efficient communications, it is desirable to maxi-
mize the amount of transmitted data bits with a unit energy.
∑
Mn
√ ∑
Mn
√ The energy efficiency of the system can be expressed as
ym,n = ĥm,n βi,n pn si + em,n βi,n pn si + zm,n
R(U, P)
i=1 i=1
ηEE (U, P) = (6)
( )(2) PT (U, P) + Pc
where si is the modulated symbol and zm,n ∼ CN 0, σn2 is
the additive white Gaussian noise (AWGN) with mean zero ∑
N
where PT (U, P) = pn is the transmitted power by the
and variance σn2 . Let pn be the power assigned to SCn and n=1
denote βm,n as the power allocation factor for U Em , with BS and Pc is the circuit power consumption of BS. The
∑n
M optimization problem can be formulated as
β1,n < β2,n < · · · < βMn ,n and βi,n = 1.
i=1
max ηEE (U, P) (7)
U ,P
B. Outage Performance for the NOMA System
In the NOMA system, having SIC technology at the receiv- m,n ≤ εout
subject to C1 : Pout
er, U Em is required to detect message from User l (m < l) and C2 : 0 ≤ βm,n ≤ 1
remove the U El ’s message from its observation in a successive
C3 : pm,n ≥ 0 (8)
manner. Therefore, the outage event only occurs at U Em when
it is unable to decode the message from users l ≥ m. Inspired ∑
N
C4 : pn ≤ Pmax
by [6], the outage of U Em on SCn can be written by
n=1

Mn ! ∑ ( m−1 )
m−1
r where εout is the outage probability requirement, and Pmax
out
Pm,n = (−1)
(m − 1)!(Mn − m)! r=0 r is the maximum power constraint for the BS. As observed
( ( ∗ 2 ))Mn −m+r+1 (3) in (7), user scheduling and power allocation are coupled
ξ (ρσ +1)
1 − exp − m σ2e with each other. Thus, the optimization problem of (7) under
× h
the constraints in (8) is a constrained combinatorial non-
Mn − m + r + 1
convex optimization problem, and it is challenging to find the
∗
where ρ = pn
2 ,
σn ξm = max {ξm+1 , ξm+2 , · · · , ξMn } and global optimal solution within polynomial time. To solve this
∑
l−1 problem efficiently, we divide it into two sub-problems: the
ξl , γl
∑
l−1
conditioned on βl > γl βi,n . Define user scheduling sub-problem and the power allocation sub-
ρ(βl −γl βi ) i=1
i=1 problem.
R̃m as the minimum data rate of User m and γm = 2R̃m − 1.
In a NOMA system with imperfect CSI, the communication III. E NERGY-E FFICIENT R ESOURCE A LLOCATION
fails when the user data rate is less than the minimum data A LGORITHMS IN THE NOMA S YSTEM
rate QoS requirement. Therefore, the outage performance is a In this section, we first transform the probabilistic mixed
useful metric to evaluate whether the user can meet the QoS non-convex optimization problem into a non-probabilistic op-
requirement. timization problem. Then we present a low-complexity subop-
timal user scheduling algorithm and a power allocation scheme Algorithm 1 Low-Complexity User Scheduling Algorithm
in the NOMA system. 1:
∆
Construct the estimate channel gain H = [|ĥm,n |]M ×N .
A. Optimization Problem Transformation 2: Initialize the sets UU nAll and SCU nAll to record the unal-
located users and unallocated subchannels in the system.
Based on (3), the outage probability of U E1 and U E2
3: Initialize the lists for all the subchannels SAllocated (n) to
sharing SCn can be respectively expressed as
record the allocated users on SCn , ∀n ∈ {1, 2, · · · , N }.
∗
Pout 2
1,n =F|ĥ1,n |2 (ξ1 (ρσe + 1)) 4: Initialize the lists for all the subchannels EE(n) to record
( )2 the energy efficiency of SCn .
ξ1∗ (ρσe2 + 1) (9) 5: while SCU nAll ̸= ∅ do
= 1 − exp(− )
σĥ2 6: Find the maximum value |ĥm,n | in H, and assign the
7: user m onto the subchannel n,
and
Pout
2,n =F|ĥ |2 (ξ2∗ (ρσe2 + 1)) |ĥm,n | = arg max (H)
( )
2,n
m∈U nAl ,n∈HU nAl
ξ2∗ (ρσe2 + 1) UU nAll =UU nAll \U Em .
=2 1 − exp(− ) 8:
σĥ2 (10) 9: EEn,possible =∅.
( )2 10: EEn,i =∅.
ξ ∗ (ρσe2 + 1)
− 1 − exp(− 2 ) 11: for each user in SCU nAll do
σĥ2 12: a) Find power allocation factor βm,n for U Em and
where F|ĥ |2 (·) is the cumulative density functions (CDF) 13: U Ei by exhaustive searching in Section III.
m,n
14: b) Calculate the energy efficiency EEn of these
of the unordered channel coefficient |ĥm,n |2 . The probabilistic 15: two users on SCn .
constraint C1 in (7) can be written by 16: c) EEn possible ← EEn,i .
{
F|ĥ |2 (ξ1∗ (ρσe2 + 1)) ≤ εout 17: end for
C1 : Pm,n ≤ εout ⇒
out 1,n
18: EE(n) = max (EEn possible ).
F|ĥ |2 (ξ2∗ (ρσe2 + 1)) ≤ εout U Ei ,i∈UU nAl
SCU nAll =SCU nAll \SCn .
2,n
(11) 19:
where ξ1∗ = max{ξ1 } = ξ1 = ρ(βγ1,n l
, ξ ∗
= max{ξ , ξ }= 20: UU nAll =UU nAll \U Ei .
) 2 1 2
−1 21: Let the mth and ith row’s elements in H be zeros.
ξ2 = ρ(β2,n −γl β1,n ) , l ≥ 1. Define F
γl
2 (·) as the inverse
|ĥm,n | 22: Let the nth column’s elements in H be zeros.
CDF of |ĥm,n | . Therefore, the constraint C1 can be converted
2 23: end while
to 

2
 γ1 (ρσe +1)
≤ F −1 2 (εout )
ρβ1,n σ2
ĥ |ĥ1,n | (12)

 ρ(β2,n −γ
γ2 (ρσe2 +1)
≤ F −1 2 (εout ) allocated. First, UU nAll is initialized to record the users who
2 β1,n ) σ2
ĥ |ĥ2,n |
have not been allocated on any subchannel, and SCU nAll is
where √ initialized to record the subchannels where no user has been
F −1 2 (εout ) = −σĥ2 log(1 − εout )
|ĥ1,n | allocated. In the allocating procedure, we first find the user
√ (13)
F −1 2 (εout ) = −σĥ2 log( 1 − εout ). who has the maximum channel gain and allocate it to the
|ĥ2,n | corresponding subchannel. On this subchannel, the second
Therefore, the power allocation factor range of the two users user should be selected from the remaining unallocated users
can be determined by (12) and (13). in UU nAll . The users who are chosen to multiplex on this
subchannel must have the maximum energy efficiency. This
B. User Scheduling Algorithm process terminates if there is no user left to be allocated.
The user scheduling algorithm starts with assigning equal
1) Complexity Analysis: For a given power allocation
power allocation across subchannels pn = Pmax /N . The
scheme, the optimal user scheduling algorithm can only be
optimal solution of user scheduling is difficult to obtain in
obtained through exhaustive search. The time complexity of
practice as it requires to search all the possible combinations
exhaustive searching is O( (2N
2N
)!
). The complexity of the pro-
of every two users in the system. Thus, we propose a low- 2
posed algorithm is O(2N ). Taking natural logarithm of the
complexity suboptimal user scheduling algorithm. In this
complexity, O(ln N ) < O(N ln N ). Therefore, the complexity
user scheduling algorithm, the power allocation factor βm,n
of the proposed algorithm is much less than the optimal user
among the multiplexed users sharing the same subchannel is
scheduling scheme.
determined to maximize the system energy efficiency.
Algorithm 1 describes the proposed suboptimal user 2) Power Allocation Factors Determination: In the step of
scheduling process. We denote UU nAll as a set of users determining power allocation factor βm,n , the optimal value
who have not been allocated on any subchannel, and denote can be found through an exhaustive search for all the values
SCU nAll as a set of subchannels where no users have been in [0, 1].
 IV. E NERGY-E FFICIENT P OWER ALLOCATION FOR and g1 > 0, g2 > 0. g1 (pn ) and g2 (pn ) are concave functions.
SUBCHANNELS As a result, R(P) is concave.
In Algorithm 1, it is required to determine the power After we prove the quasi-concavity of problem (14), the
proportional factor βm,n for subchannel multiplexing users. In gradient assisted binary search algorithm (GABS) can be used
this section, given the user scheduling scheme, a novel power for power allocation as shown in Algorithm 2. The GABS
allocation for subchannels algorithm is proposed to replace the algorithm is specifically described in [13]. [P]+ sets the
equal power allocation scheme. negative elements in vector P to be zero.

A. Energy-Efficient Power Allocation Scheme Algorithm 2 Power Allocation Based on Gradient Assisted
Given the user scheduling and power allocation factors by Binary Search Algorithm
Algorithm 1, the optimization problem can be rewritten by 1: Initialization P(0) and tolerance ε.
∑
N 2: Optimization power allocation vector P. Let P = Po .
Rn 3: Repeat
n=1
max
P≻0
ηEE (P) =
∑
N
(14) 4: use GABS [ to find the optimal] step size t∗
+
pn + Pc 5: P = P + t∗ ∇ηEE (P0 )
1 0
n=1
6: if ηEE (P ) − ηEE (P ) < ε
1 0

∑
N 7: Return P1 .
subject to : pn ≤ Pmax (15) 8: else
n=1 9: P0 = P1 .
T
where P = [p1 , p2 , · · · , pn , · · · , pN ] represents the assigned 10: end
T
powers on the subchannels, (·) is denoted as the transpose,
and P ≻ 0 means all elements of P are positive. However,
the utility function in (14) is a nonlinear function respect V. S IMULATION R ESULTS
to the transmission power P. It is difficult to solve (14) In this section, simulation results are presented to evaluate
with the constraint (15). Therefore, to reduce the computation the performance of the proposed resource allocation scheme
complexity, we utilize the gradient assisted binary search for the NOMA system. In the simulations, we consider one
algorithm to achieve the energy-efficient power allocation [13], basestation located in the cell center and M UEs are uniformly
[14]. Before describing the algorithm, we first prove that distributed in the circular range. Denote the bandwidth of the
(14) is a quasi-concave function by using Proposition 1 and system as 5 MHz and the small-scale fading gain as Rayleigh
Proposition 2 as follows. distributed with σn2 = 1. In NOMA systems, to reduce
Proposition 1: If R(P) is strictly concave in P, ηEE (P) is demodulation complexity of the SIC receiver, each subchannel
strictly quasi-concave. is shared by two users. In OFDMA systems, each user can only
Inspired by [15], we can prove Proposition 1 as follows. be assigned to one subchannel. In the simulations, we compare
Proof : Denote the upper contour sets of ηEE (P) as our proposed resource allocation scheme for NOMA systems
Sα = {P ≻ 0|ηEE (P) ≥ α} . (16) with a conventional OFDMA system. Since the algorithm has
been designed based on imperfect CSI, the energy efficiency
According to a proposition in [16], ηEE (P) is strictly quasi- performance of the system with different channel estimation
concave if and only if Sα is strictly convex for any real number error variance σe2 is shown in Fig. 2. For our simulations,
α. When α < 0, there are no points satisfying ηEE (P) = α. we set BS peak power Pmax to 1 W and circuit power
When α = 0, only 0 is on the contour ηEE (0) = α. When consumption Pc = 0.1 W. The maximum number of users
α > 0, Sα = {P ≻ 0|αPc + αPT (P) − R (P) ≤ 0}. Since is 60 and σn2 = N B
N0 , where N0 = −174 dBm/Hz is the
PT (P) is strictly convex in P and R(P) is concave, which is AWGN power spectral density. The minimum data rate QoS
to be proved by Proposition 2. Therefore, we prove the strict requirements of U E1 and U E2 are R̃1 = R̃2 = 0.5 bits/s/Hz.
quasi-concavity of ηEE (P). In Fig. 1, performance of energy efficiency is evaluated
Proposition 2: Rn (pn ) = log2 (g1 (pn )) + log2 (g2 (pn )) versus the number of users M . In the simulation environment,
is concave if g1 and g2 are concave and positive, where we set the channel estimation error variance as σe2 = 0.005
|ĥ1,n | β1,n pn |ĥ2,n | β2,n pn
2 2

g1 (pn ) = 1 + σ2 pn +σ2 and g2 (pn ) = 1 + Ωm,n and the outage probability εout as 0.01. It is shown that the
e n
2 system energy efficiency increases when the number of users
with Ωm,n = |ĥ2,n | β1,n pn + σe2 pn + σn2 . Proof : Since
grows. As the number of users grows, the energy efficiency
∂ 2 g1 −|ĥ1,n |2 β1,n σn2 continues to increase, but the rate of growth becomes slower.
= 4 < 0,
∂p2n (σe2 pn + σn2 ) The performance of our proposed resource scheduling scheme
( ) for the NOMA system achieves higher energy efficiency than
∂ 2 g2 −|ĥ2,n |2 β2,n σn2 2Ωm,n |ĥ2,n |2 β1,n + σe2 the OFDMA scheme as well as the NOMA system with the
= <0
∂p2n Ω4m,n equal power allocation (NOMA-EQ) scheme. For example,
(17) when the number of users is 20, the energy efficiency of
 8

5
x 10
energy efficiency of the NOMA system was higher than the
4.5 OFDMA scheme, and the energy efficiency of the NOMA
system deteriorates when the channel estimation error is
Energy efficiency of system (bits/Joule)

3.5 increased.
3
VII. ACKNOWLEDGMENT
2.5
This work was supported by the National Natural Science
2
Foundation of China (61471025), the Open Research Fund of
1.5
The State Key Laboratory of Integrated Services Networks,
1
NOMA Pc=0.1w Xidian University (ISN17-02), and the Fundamental Research
0.5 NOMA−EQ Pc=0.1w
OFDMA Funds for the Central Universities.
0
0 10 20 30 40 50 60
Number of users per BS R EFERENCES
[1] Y. Saito, Y. Kishiyama, A. Benjebbour, T. Nakamura, A. Li, and K.
Fig. 1. Energy efficiency performance comparison with existing schemes. Higuchi, “Non-orthogonal multiple access (NOMA) for cellular future
radio access,” Proc. IEEE VTC Spring, pp. 1–5, June 2013.
x 10
8 [2] K. Higuchi and Y. Kishiyama, “Non-orthogonal access with random
7
beamforming and intra-beam SIC for cellular MIMO downlink,” Proc.
IEEE VTC Fall, pp. 1–5, Sept. 2013.
6
[3] L. Dai, B. Wang, Y. Yuan, S. Han, C.-L. I, and Z. Wang, “Nonorthogonal
Energy efficiency of system (bits/Joule)

multiple access for 5G: Solutions, challenges, opportunities, and future

5
research trends,” IEEE Commun. Mag., vol. 53, no. 9, pp. 74-81, Sept.
2015.
4
[4] P. Parida and S. Das, “Power allocation in OFDM based NOMA system:
A DC programming approach,” IEEE Globecom Workshops, pp. 1026–
3
1031, Dec. 2014.
2
[5] Z. Ding, M. Peng, and H. V. Poor, “Cooperative non-orthogonal multiple
access in 5G systems,” IEEE Commun. Lett., vol. 19, no. 8, pp. 1462–
1 NOMA error variance=0.0005 1465, June 2015.
NOMA error variance=0.005 [6] Z. Ding, Z. Yang, P. Fan, and H. V. Poor, “On the performance of non-
NOMA error variance=0.05
0
orthogonal multiple access in 5G systems with randomly deployed users,”
0 10 20 30 40 50 60 IEEE Sig. Process. Lett., vol. 21, no. 12, pp. 1501–1505, Dec. 2014.
Number of users per BS
[7] Z. Yang, Z. Ding, P. Fan, and G. K. Karagiannidis, “On the performance
of non-orthogonal multiple access systems with partial channel informa-
Fig. 2. Energy efficiency of proposed resource allocation scheme with tion,” IEEE Trans. Commun., vol. 64, no. 2, pp. 654–667, Feb. 2016.
different σe2 values. [8] F. Fang, H. Zhang, J. Cheng, and V. C. M. Leung, “Energy efficiency of
resource scheduling for non-orthogonal multiple access (NOMA) wireless
network,” Proc. IEEE ICC 2016, Kuala Lampur, Malaysia, May, 2016.
[9] F. Fang, H. Zhang, J. Cheng, and V. C. M. Leung, “Energy-efficient
proposed resource allocation scheme for the NOMA system resource allocation for downlink non-orthogonal multiple access (NOMA)
is 21% more than that of the OFDMA scheme, and is 12% network,” IEEE Trans. Commun., Sept. 2016.
[10] H. Zhang, Y. Nie, J. Cheng, V. C. M. Leung, and A. Nallanathans,
more than that of the NOMA-EQ scheme. “Sensing Time Optimization and Power Control for Energy Efficient
In Fig. 2, the energy efficiency of the NOMA system Cognitive Small Cell With Imperfect Hybrid Spectrum Sensing,” IEEE
with different estimation error variances is presented versus Trans. Wireless Commun., vol. 16, no. 2, pp. 1536–1276, Feb. 2017.
[11] H. Zhang, C. Jiang, X. Mao and H. Chen, “Interference-Limited Re-
the number of users M . As observed in Fig. 3, the energy source Optimization in Cognitive Femtocells with Fairness and Imperfect
efficiency of the system deteriorates when the error vari- Spectrum Sensing,” IEEE Trans, Veh. Technol., vol. 65, no. 3, pp. 1761–
ances increases. When the number of users is 30, the energy 1771, Mar. 2016.
[12] H. Zhang, C. Jiang, N. Beaulieu, X. Chu, X. Wang and T. Quek,
efficiency of the proposed resource allocation scheme with “Resource Allocation for Cognitive Small Cell Networks: A Cooperative
σe2 = 0.0005 is 5% more than that with σe2 = 0.005 and Bargaining Game Theoretic Approach,” IEEE Trans. Wireless Commun.,
is 44% more than that with σe2 = 0.05. Thus, as expected, vol. 14, no. 6, pp. 3481–3493, June 2015.
[13] G. Ye, H. Zhang, H. Liu, J. Cheng, and V. C. M. Leung, “Energy
the channel estimation error can degrade the energy efficiency Efficient Joint User Association and Power Allocation in a Two-Tier
performance. Heterogeneous Network,” Proc. IEEEE Globecom 2016, Washington, DC,
Dec. 4-8, 2016.
VI. C ONCLUSION [14] H. Liu, H. Zhang, J. Cheng, and V. C. M. Leung, “Energy Efficient
Power Allocation and Backhaul Design in Heterogeneous Small Cell
We studied the resource allocation in the NOMA system Networks,” Proc. IEEE ICC 2016, Kuala Lampur, Malaysia, May 23-
by considering imperfect CSI. By formulating the resource 27, 2016.
[15] G. Miao, N. Himayat, and G. Y. Li, “Energy-efficient link adaptation in
allocation as a probabilistic mixed non-convex optimization frequency-selective channels,” IEEE Trans. Commun., vol. 58, no. 2, pp.
problem, a low-complexity suboptimal user scheduling algo- 545–554, Feb. 2010.
rithm was proposed to maximize the system energy efficiency. [16] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge Uni-
versity Press, New York, NY, USA, 2004.
Given the user scheduling scheme, we presented an energy-
efficient power allocation scheme. The performance of the
proposed resource allocation scheme was compared with that
of the conventional OFDMA system. It was shown that the

View publication stats