COMMUNICATIONS THEORIES & SYSTEMS

Antenna Selection and Power Allocation Design for

5G Massive MIMO Uplink Networks
Hongyuan Gao1,*, Yumeng Su1, Shibo Zhang1, Ming Diao1
1
College of Information and Communication Engineering, Harbin Engineering University, Harbin 150001, China
* The corresponding author, email: [email protected]

Abstract: Massive MIMO is one of the key wireless networks, the carbon emissions and
technologies in future 5G communications operating costs caused by wireless commu-
which can satisfy the requirement of high nication are increasing day by day [1]. Now-
speed and large capacity. This paper considers adays, promoting energy conservation then
antenna selection and power allocation design building intensive community has become a
to promote energy conservation then provide heated research topic [2-4]. Therefore, careful
good quality of service (QoS) for the whole planning and deployment of the base station
massive MIMO uplink network. Unlike pre- (BS) infrastructure is necessary to decrease
vious related works, hardware impairment, the energy consumption in line with green
transmission efficiency, and energy consump- objectives [5-7]. As a key technology of 5G
tion at the circuit and antennas are involved in communication system, massive MIMO has
massive MIMO networks. In order to ensure received substantial attention in both academic
the QoS, we consider the minimum rate con- and industry domain since 2010 [8-12]. Com-
straint for each user and the system, which pared to traditional communication networks,
increases the complexity of power allocation massive MIMO has great advantages of higher
problem for maximizing energy and spectral data rate, larger capacity, lower latency and
efficiency in massive MIMO system. To this greater throughput [13].
end, a quantum-inspired social emotional op- Large number of antennas at the base sta-
timization (QSEO) algorithm is proposed to tion is the major characteristic of massive
obtain the optimal power control strategy in MIMO compared with conventional MIMO
massive MIMO uplink networks. Simulation technology [14]. The promising technology
results assess the great advantages of QSEO has been studied in many aspects [15-22].
which previous strategies do not have. Distributed MIMO has been illustrated to be
Keywords: 5G; massive MIMO; antenna se- capable of significantly increasing the sys-
lection; power allocation; quantum-inspired tem capacity [15]. In [16], the author studied
social emotional optimization the sum rate in different signal-to-noise ratio
(SNR) conditions and illustrated how dense
I. INTRODUCTION multiple antenna arrays can be designed in
massive MIMO system. In [17], a coordination
Received: Apr. 19,2018
With the increasing demands of high-speed approach was proposed to reduce the negative
Revised: Aug. 23, 2018
communication and rapid development of influence of pilot contamination on channel es- Editor: Da Chen

China Communications • April 2019 1

 timation. Importantly, the author demonstrated author of [25] proposed an antenna selection
A quantum-inspired that the effect of pilot contamination can be strategy based on the theory of rectangular
social emotional op- completely vanished under certain conditions maximum volume submatrices. However, this
timization (QSEO) on the channel covariance. To reduce the com- strategy is invalid if a square matrix with max-
algorithm is proposed
putational complexity of channel estimation, imum-volume is not given.
to obtain the optimal
power control strategy
[18] decomposed the space of the received Due to the growing demand of environ-
in massive MIMO up- signals into three subspaces according to fac- mental protection, energy efficiency (EE) and
link networks. tor analysis, and an interference-free subspace spectral efficiency (SE) have become two
was created to obtain accurate channel esti- important concerns in massive MIMO net-
mation. In [19], a joint pilot design and power works [26-28]. Power allocation is an essential
allocation strategy was considered to mitigate technique to enhance the system performance
pilot contamination and provide good service and promote energy conservation. To further
in multi-cell massive MIMO system, which exploit the benefits of power allocation, more
can be used as a benchmark for pilot design and more works have been proposed for pow-
in ideal or non-ideal hardware scenarios. Due er allocation in massive MIMO networks [29-
to the high hardware complexity, linear pro- 32]. For maximizing the achievable uplink
cessing methods are widely used in massive rate in multi-cell massive MIMO systems, a
MIMO system [20]. In [21], a low-complexity pilot power allocation strategy was proposed
hybrid precoding method named phased-ZF in [29]. Considering pilot allocation, hardware
(PZF) was proposed to approach the perfor- impairments and other system parameters, the
mance of the optimal linear precoding scheme resource allocation problem for maximizing
in massive MIMO systems. In [22], two inter- SE in multi-cell massive MIMO system was
ference-suppressed precoding methods were discussed in [30]. In [31], an approximate
proposed and significantly suppressed the power allocation scheme for maximizing EE
mutual interference between the users with was proposed in massive MIMO networks.
statistical and imperfect instantaneous channel The author considered power amplifier effi-
state information (CSI). Pilot-based channel ciency, transmission power and circuit pow-
estimation can be avoided by utilizing statisti- er. To ensure the QoS of the whole massive
cal CSI. MIMO system, minimum rate constraint of
With the dramatic increase of antenna quan- the system was considered in [32]. The paper
tity, antenna selection is an effective technique developed a unified framework for EE op-
to decrease the operating cost and the number timization and proposed a power allocation
of radio frequency chains in massive MIMO method based on fractional programming the-
networks. It’s impractical to obtain the opti- ory. However, all schemes mentioned above
mal antenna selection by exhaustive search could not guarantee the QoS for each user and
because of the high computational complexity. the system simultaneously. Especially in more
Some researches on suboptimal antenna se- complex conditions, it’s hard for mathematical
lection methods are emerged in recent years approximation to obtain the optimal solution.
[23-25]. In [23], a successive removal strategy
1.1 Motivation and contributions
for antenna selection was proposed. The suc-
cessive elimination strategy was performed In this paper, we propose an effective anten-
according to the received channel coefficients na selection and power allocation strategy to
from the previous users. Considering the trade- optimize the EE and SE in massive MIMO
off between the performance and computation- uplink networks. To reduce the computation
al complexity, a mixed-integer programming complexity and satisfy the changing require-
approach was proposed to jointly optimize ments of the practical system in real time, an
antenna selection and precoding scheme [24]. effective antenna selection strategy is designed
In order to maximize the channel capacity, the to select a group of antennas from the avail-

2 China Communications • April 2019

able ones at the BS. For promoting energy Through the simulation results, QSEO has
conservation, we assume that the unselected the advantages which previous strategies do
antennas do not consume any energy. Besides, not have.
hardware impairment, transmission efficien- • QSEO combines the advantages of both
cy, energy consumed at active antennas and quantum evolution and social emotional
circuit energy consumption are considered in optimization algorithm, with the strong
massive MIMO system. Different from former searching ability and fast convergence. Be-
articles, minimum rate constraints of each sides, the QSEO algorithm proposed in this
user and the system are satisfied to guarantee paper can not only be used to obtain the
the QoS for each user and the whole system, optimal power allocation strategy in mas-
which highly increase the complexity of the sive MIMO uplink networks, it’s a general
power allocation problem. A quantum-inspired algorithm which can also be widely used
social emotional optimization (QSEO) algo- to solve continuous, discrete and hybrid
rithm is designed to obtain the optimal power optimization problems in many engineering
allocation result in massive MIMO uplink domains.
networks. The main contributions of this paper
1.2 Organization and notation
are summarized as follows:
• An effective antenna selection strategy is Other sections of this paper are organized as
proposed to reduce the radio frequency follows. The model of the massive MIMO
chains and promote energy conservation in uplink network is shown in the next section.
massive MIMO uplink networks. The num- In Section III, we propose an effective antenna
ber of selected antennas varies according selection strategy and derive exact expressions
to the variation of system requirements in of EE and SE considering the hardware im-
real time. Larger amount of antennas are pairment of the BS. Power allocation strategy
selected when higher transmission rate is based on QSEO for maximizing EE and SE
required in practical system. are proposed in Section IV. The simulation
• Exact expressions for EE and SE are giv- results are given in Section V and the last sec-
en considering the hardware impairment, tion concludes this paper.
transmission efficiency, circuit power con- Notation: In this paper, matrix and vector
sumption and power consumed by active are denoted by boldface upper and lower case
antennas in massive MIMO networks. To symbols. Superscripts (.)T and (.) H represent
optimize the EE and SE while guarantee the the transpose and conjugate transpose, respec-
QoS in massive MIMO uplink networks, tively. || . || and | . | denote the Euclidean norm
minimum rate constraints of each user and of a vector and a scalar, respectively. diag(b)
the system are considered in this paper. represents the diagonal matrix with the ele-
• A quantum-inspired social emotional op- ments of vector b on its main diagonal, and
timization (QSEO) algorithm is proposed E{.} represents the expectation.
to solve the multi-constrained non-con-
vex power allocation problem in massive
MIMO uplink networks. Two different
II. SYSTEM MODEL
quantum evolution rules are designed to
In this paper, a single-cell multiuser massive
improve the diversity of solutions and
MIMO system is considered. One base station
guarantee the strong searching ability of
equipped with M antennas and K users with
the algorithm. The interactions among the
single antenna constitute the system. B is the
quantum individuals are fully utilized to
bandwidth of the system. To promote energy
obtain the optimal solution of the feasible
conservation, there is not necessary to utilize
region, which avoids the local convergence
all the antennas at the BS during the period of
compared with traditional SEO algorithm.

China Communications • April 2019 3

 data transmission. According to the require- with
ments of practical system, we select a part of α k =| gkH gk |2 ,(5)
antennas to participate in communication. The σ k2 = gkH gk σ 2 ,(6)
antenna selection strategy is discussed in the ωk , j =| gkH g j |2 ,(7)
following section. Assuming that M s antennas
where α k and ωk , j are positive quantities de-
are selected at the BS, M s ≤ M . In massive
pend on transmission channels, σ 2 denotes the
MIMO uplink system, the CSI matrix from the
noise power. In particular, α k is only related
users to the BS can be expressed by
to the CSI of user k, while the interference
G = HD,(1)
coefficient ωk , j depends on the CSI from other
w h e r e G i s a n Ms × K m a t r i x a n d
users to the base station.
G = [ g1 , g2 ,..., g K ], gk denotes the CSI between
Hardware impairment is inevitable in prac-
the k th (k = 1, 2,..., K ) user and the BS. H is
tical massive MIMO system. The degree of
the small-scale fading matrix with the size of the reduction for the desired signals causing
M s × K, while D = diag( β1 , β 2 ,..., β K ) is by the hardware impairment is 1− ε 2, where
the large-scale fading matrix with the size of ε represents the magnitude of error. For sim-
K × K. β k represents the shadow and geomet- plicity, we assume that the BS is hardware-im-
ric fading between the k th user and the BS, paired, and ε BS denotes the impairment mag-
which is shown by the following formula nitude. The SINR experienced by user k at the
z BS takes the following general form [32]:
β k = k v ,(2)
 rk  (1 − ε BS
2
) pk α k
  γk = K
,(8)
 r0  σ k + pk φk + ∑ p j wk , j
2

where rk is the distance from user k to the BS, j =1

j≠k

r0 denotes the radius of the BS, v represents with

the path loss exponent, and zk is a log-normal φk = ε BS2
gkH Dk gk ,(9)
distributed variable with zero mean and stan- wk , j | gkH g j |2 +ε BS
= 2
gkH D j gk ,(10)
dard deviation. w h e r e Dk = diag ({| g m , k |2 }mM=s 1 ) , g m , k d e -
During the data transmission phase, each
notes the CSI between the user k and
user transmits its signal to the BS while inter-
the m th antenna of the BS. Similarly,
fering with other users. The signal received at
D j = diag ({| g m , j |2 }mM=s 1 ), g m , j denotes the CSI
the BS is given by
K between the user j and the m th antenna of the
y =pk gk sk + ∑ p j g j s j + n ,(3)
j =1
BS. α k and σ k2 are the same form as in (5) and
j≠k
(6).
where pk and p j denote the transmission power
The data transmission rate of user k is given
of user k and user j (j = 1,..., K, j ≠ k). sk and s j by
denote the signal vector of user k and user j, = Rk B log 2 (1 + γ k ).(11)
which satisfy E{|| sk ||2 } = 1 and E{|| s j ||2 } = 1, Then, the achievable rate of the massive
respectively. n represents the additive white MIMO uplink system can be expressed as
Gaussian noise vector. K K

In the case of all CSI can be obtained and

=R
=
k
k 1=
∑
= R ∑ B log
k 1
2 (1 + γ k ).(12)

maximum ratio combining (MRC) is used for

signal processing at the BS, the signal-to-inter- III. OPTIMIZATION FUNCTIONS IN
ference-plus- noise ratio (SINR) experienced MASSIVE MIMO UPLINK NETWORKS
by user k at the BS is shown as
pk α k In this section, we first propose an effective
γk = K
,(4)
σ + ∑ p j ωk , j
2
k
antenna selection strategy. Then, we derive
j =1
j≠k

4 China Communications • April 2019

exact expressions of SE and EE which are two Algorithm 1. Steps of antenna selection strategy.
major concerns in massive MIMO system.
1 Input:
Considering the impact of various parameters, 2 GM × K = [ g1 , g2 ,..., g K ] // CSI matrix;
these expressions provide practical design in-
3 M // the number of available transmission antennas;
sights on practical massive MIMO system. 4 ξ // antenna selection coefficient;

3.1 Antenna selection 5 Initialize:

6 num = 1 // the first iteration;
According to the requirements of massive 7 A = Φ // the set of selected antennas;
MIMO system during the period of data trans- 8 Obtain M s by (13);
mission in real time, we select M s(M s ≤ M) 9 Calculate the achievable sum-rate from the users to each available antenna by
antennas out of the available M ones at the (12);
10 while num ≤ M s do
BS. The specific expression of the relationship
between M s and M is shown by 11 Select the best one which a higher sum-rate can be obtained from M antennas;
12 Add the selected antenna to the set of A;
M= s ξ ⋅ M  ,(13) 13 M= M −1;
where . denotes the ceiling function, ξ de- 14 num
= num +1;
notes the antenna selection coefficient which 15 end while
is related to the needs of the practical system. 16 Output: The set of selected antennas A.

Assuming that the CSI for all antennas is 

obtained at the BS, the optimal solution for
antenna selection can be obtained according Rk
ϕ=
k = log 2 (1 + γ k ).(14)
to the exhaustive search under a given per- B
formance metric. However, with the large For user k, the following power constraint
amount of antennas in practical system, huge must be satisfied
computational cost is inevitable by exhaustive pk ≤ pkmax ,(15)
search. It is necessary to design a new strategy where pkmax denotes the maximum transmission
since the existing solutions cannot match the power of the user k. To ensure the commu-
practical system in real time. For this reason, nication quality of each user, the minimum
an effective antenna selection strategy to re- achievable rate of each user should be satis-
duce the calculation complexity is proposed. fied, which is shown by
At the stage of antenna selection, we as- B log 2 (1 + γ k ) ≥ Rkmin ,(16)
sume that all users have the same transmission where Rkmin is the minimum achievable rate of
power and the antennas at the BS are inde-
user k.
pendent of each other. Then, we calculate the
In order to ensure the quality of service
achievable rate from the users to each antenna
(QoS) for the massive MIMO system, the fol-
by (12). Antennas which provide higher sum-
lowing expression needs to be satisfied
rate are selected at the BS. K

The basic idea of the antenna selection ∑ B log 2 (1 + γ k ) ≥ Rmin ,(17)

k =1
strategy in this paper is to select the best M s
where Rmin represents the minimum achievable
antennas which afford the highest transmission
rate of the system.
rate. The steps of the antenna selection strate-
The total SE (in bit/s/Hz) of the system is
gy are summarized in Algorithm 1.
given as follows
3.2 Spectral efficiency
K
=ϕ ∑ log
k =1
2 (1 + γ k ).(18)
According to the expression of transmission
According to (15), (16) and (17), we de-
rate for user k in (11), the achievable SE (in
fine P as the feasible set of power allocation
bit/s/ Hz) ϕ k for user k is defined by
results, and p = [ p1 , p2 ,..., pK ] represents one

China Communications • April 2019 5

 feasible power allocation result of P. Accord- networks. However, the maximization of
ing to the settings mentioned above, the prob- SE and EE are two competing goals in mas-
lem of SE maximization can be denoted as sive MIMO system. Both (19) and (22) are
K
non-convex and complicated optimization
ϕ ∗ max
= = ϕ max ∑ log 2 (1 + γ k ).(19)
p∈P p∈P
k =1 problems. To tackle the tough problem, an
efficient power control strategy is proposed in
3.3 Energy efficiency the next section.
The network EE (in bit/J) is related to the
sum-rate and the power consumed in the sys- IV. POWER CONTROL STRATEGY IN
tem, which is defined by the following formu- MASSIVE MIMO UPLINK NETWORKS
la
K
As is shown above, the aim of the problems
∑ B log 2 (1 + γ k )
mentioned ahead is to find the optimal power
ψ = k =1
,(20)
PΣ control strategies for maximizing SE and EE
where P∑ represents the total consumption of in massive MIMO networks. Unfortunately,
energy. The circuit power consumption always the feasible region is difficult to determine.
exists in practical massive MIMO networks. Especially in (22), the objective function does
Then we consider the power consumed at the not have a concave numerator, and it’s difficult
users and the BS respectively. For the perspec- to find the optimal solution with affordable
tive of users, the power consumption is larger complexity. Since the multi-constrains nonlin-
than total transmission power of each user ear optimization problem is difficult to resolve,
since the transmission efficiency cannot reach the existing power allocation method based
100%. Active antennas at the BS also consume on fractional programming is hard to find the
energy to ensure their works for receiving and optimal solution, and the traditional intelligent
processing signals from the users. Thus, the algorithms cannot be simply applied to the
total consumed power is given by problem because of local convergence. In this
Ms section, a quantum-inspired social emotional
K
1 K
PΣ = ∑ pc , m + ∑ pc , k + ∑ pk ,(21) optimization (QSEO) algorithm is proposed
=m 1=k 1 η =k 1
to solve the power allocation problems men-
where pc , m denotes the power consumed at the
tioned before.
m th antenna, pc , k denotes the circuit power
dissipated from user k to the BS, and η rep- 4.1 Quantum-inspired social
resents the transmission efficiency. emotional optimization algorithm
The feasible set of power allocation results QSEO is a discrete optimization algorithm
P is the same as in 3.2 according to the con- which is inspired from the social emotional
straints in (15), (16) and (17). The problem of optimization (SEO) algorithm [33] and the
EE maximization can be formulated as follows thought of quantum intelligence computation
K
[34]. In the L-dimensional space (L represents
∑ B log 2 (1 + γ k )
the number of dimensions for the optimization
ψ
= =*
max ψ max Ms
k =1
.
p∈P p∈P K
1 K
problem), there are N quantum individuals
=m 1=k 1
∑ pc,m + ∑ pc,k +
η =k 1
∑ pk in quantum population of QSEO. The n th
(22) (n = 1, 2,..., N) quantum individual of the t th
SE and EE are two major concerns in the iteration is defined as
performance of practical massive MIMO sys- α nt ,1 , α nt ,2 ,..., α nt , L 
t
tem. In particular, ϕ in (18) denotes a variable x = t n  ,(23)
 β n ,1 , β n ,2 ,..., β n , L 
t t

in proportion to the sum-rate while ψ in (20)

w h e r e | α nt ,l |2 + | β nt,l |2 = 1 , l = 1, 2,..., L .
is the benefit-cost in massive MIMO uplink
We define 0 ≤ α nt ,l ≤ 1 and 0 ≤ β nt,l ≤ 1 , so

6 China Communications • April 2019

 β nt= 1 − (α nt ,l ) 2 . Then, xnt can be simplified as  1 − ( xt )2 θ nt ,+l1 = 0 and ρ nt +,l1 ≤ κ1
,l
t +1 n ,l
x =
n ,l
=xnt [α
= n ,1 , α n ,2 ,..., α n , L ]
t t t
[ xnt ,1 , xnt ,2 ,..., xnt , L ], | xnt ,l ⋅ cos θ nt ,+l1 − 1 − ( xnt ,l ) 2 ⋅ sin θ nt ,+l1 | else,
(24)
(27)
where l = 1, 2,..., L and 0 ≤ xnt ,l ≤ 1 . Each xnt ,l
where θ nt ,+1l denotes the l th quantum rotation
denotes a quantum bit. The measurement state
angle of the n th quantum individual; c1, c2,
xnt = [ xnt ,1 , xnt ,2 ,..., xnt , L ] of the n th quantum indi-
and c3 are three constants which represent im-
vidual is gotten by the following rules, which
pact factors of the global optimal solution of
is shown by
the quantum population, the historical optimal
1, λ t > ( xnt ,l ) 2 solution of the n th quantum individual, and
xnt ,l =  nt,l ,(25)
0, λn ,l ≤ ( xn ,l )
t 2
the worst solution set to the quantum individ-
where λnt .l is a random number evenly distrib- ual n, respectively. κ1 denotes the mutation
uted between 0 and 1. probability and ρ nt +,l1 is a random variable even-
The social evaluations of all quantum indi- ly distributed between 0 and 1.
viduals are calculated by the function of spe- In particular, for the first iteration, the emo-
cific problem. The historical optimal solution tional index of any quantum individual n is
of the n th quantum individual is expressed by set to 1. Each quantum individual obeys the
µnt = [ µnt ,1 , µnt ,2 ,..., µnt , L ]. For the first iteration, following updating process
N′
the historical optimal solution of each quantum
c3 ⋅ ∑ (ϖ nt ′,l − xnt ,l ),(28)
ϑnt ,+1l =
individual is initialized as its measurement state. n ′=1

The quantum individual who gets the highest where ϑnt ,+1l denotes the l th quantum rotation
social evaluation is denoted as the global optimal angle of the n th quantum individual. The ele-
solution ugt = [u gt ,1 , u gt ,2 ,..., u gt , L ] among t itera- ments of the quantum individuals are updated
tions. We define ϖ t as the set of the worst quan- by (27), and the corresponding measurement
tum individuals which consists of N ′ (N ′ ≤ N) states of the quantum individuals are up-
quantum individuals with the lowest social dated by (25). Then, we calculate the social
evaluations, where ϖ nt ′ = [ϖ nt ′,1 ,ϖ nt ′,2 ,...,ϖ nt ′, L ] evaluations of all quantum individuals under
represents the measurement state of the n′th the same function of the problem mentioned
(n′ = 1, 2,..., N ′) quantum individual. ahead. To make it easy, the emotional index
Each quantum individual has different emo- and the historical optimal solution of each
tions which greatly affect the process of evo- quantum individual are updated by the follow-
lution. The emotional index of quantum indi- ing
vidual n is a variable between 0 and 1, which 1 f ( xnt +1 ) > f ( µ nt )

is denoted by ent. e1 and e2 are two emotional +1
ent= t t +1 t
en − ∆e f ( xn ) ≤ f ( xn ) (29)
thresholds of three different emotions, that is,  t
en else,
upset, calm and enthusiastic. Assuming that
 x t +1 f ( xnt +1 ) > f ( µ nt )
e1 < e2, the process of updating the quantum µnt+1 =  nt (30)
 µ n f ( xn ) ≤ f ( µ n ),
t +1 t
individual n according to different emotions
can be expressed as follows where f (.) denotes the social evaluation
c1 ⋅ (u gt ,l − xnt ,l ) ent < e1 function, ∆e denotes the reducing index of
 emotion. Especially, when ent+1 < 0, we define
c1 ⋅ (u g ,l − xn ,l ) + c2 ⋅ ( µn ,l − xn ,l )
t t t t

 N′ ent+1 = 0. Then, we update the worst solution set

θ nt ,+1l = +c ⋅ (ϖ t − x t )
3 ∑ e1 ≤ ent < e2
 n ′ =1
n′,l n ,l
and the global optimal solution of the quan-
 N′
c2 ⋅ ( µ nt ,l − xnt ,l ) + c3 ⋅ ∑ (ϖ nt ′,l − xnt ,l ) ent ≥ e2 , tum population. The measurement state of
 n ′=1 the quantum individual who gets the highest
(26) social evaluation value in history is selected as

China Communications • April 2019 7

 the global optimal solution among all previous social evaluations of the new quantum indi-
iterations. viduals with the computational complexity of
The process of iteration ends until the algo- O( N ). The historical optimal solutions, the
rithm achieves the terminal condition (usually emotional indexes of all quantum individuals,
depend on the predetermined number of itera- the worst solution set and the global optimal
tion). solution of the quantum population are up-
dated according to the social evaluations after
4.2 Computational complexity of
each iteration. The computational complexity
the QSEO algorithm
is O(4 N ).
As described in 4.1, the QSEO algorithm For the termination of the QSEO algorithm
needs to update the quantum rotation angles after running t iterations, the computational
in all dimensions of all quantum individuals complexity is O(t (5 N + 3 NL)).
at each iteration, which has a computational
complexity of O( NL), where N is the popula- 4.3 Power allocation strategy based
tion size and L is the maximum dimensional on QSEO
number. The quantum individuals are updated In this subsection, we consider the implemen-
by the quantum rotation angles and the mea- tation of power allocation strategy based on
surement states can be obtained by (25), with QSEO in massive MIMO uplink networks.
the computational complexity of O (2 NL ). In order to solve the non-convex continuous
After the updating process, we calculate the optimization problem, we encode the esti-
mated power of the each user by L binary
bits. Hence, the dimension of each quantum
Algorithm 2. Power control strategy for SE and EE maximization. individual is L= K × L . At the stage of ini-
1 Input system parameters proposed in Section II and Section III; tialization, the measurement state of the n th
2 t = 1 // the first iteration; quantum individual xnt = [ xnt ,1 , xnt ,2 ,..., xnt , K × L ] is
3 Initialize parameters of QSEO, quantum population, measurement states of
obtained by (25). Since the transmission pow-
quantum individuals and historical optimal solutions µ1t , µ 2t ,..., µ Nt ;
er of the each user is encoded by L binary bits,
4 Obtain the power allocation strategy of each quantum individual according to the
the power allocation result can be obtained by
mapping rules in (31);
5 Calculate the social evaluations // select (19) or (22) for SE or EE maximization,
the following formula
respectively;  kL 2( kL −l ) xnt ,l  max
pnt , k =  ∑  p ,(31)
6 Obtain the global optimal solution ugt and the set of worst solutions ϖ t of the  l =( k −1) L +1 2 L − 1  k
 
quantum population; max
where k = 1, 2,..., K, pk denotes the maximum
7 while t ≤ maximum number of iterations
transmission power of user k, and pnt , k denotes
8 if t = 1
9 Update all quantum individuals by (28) and (27) the n th feasible power allocation result of
10 else user k. The corresponding power allocation
11 Update the quantum individuals with different emotions by (26) and (27) result of quantum individual n is denoted by
12 end if pnt = [ pnt ,1 , pnt ,2 ,..., pnt , K ].
13 Obtain measurement state of each quantum individual;
For the power allocation strategy based on
14 Obtain power allocation result of each quantum individual by (31);
QSEO, (19) and (22) are considered as the
15 Calculate the social evaluations of the new quantum individuals;
16 Update the worst solution set, the global optimal solution of the quantum social evaluation functions to optimize SE and
population, historical optimal solutions and the emotional index for any n; EE, respectively. The implementation steps of
17 t = t +1; the power control strategy based on QSEO for
18 end while SE and EE maximization are summarized in
19 Obtain the optimal power allocation result according to the global optimal Algorithm 2.
solution of the quantum population;
20 Output: The optimal power allocation result.


8 China Communications • April 2019

V. SIMULATION RESULTS hardware-impaired massive MIMO networks,
the performance of power control strategies
In this section, the simulation results of dif- based on QSEO, SEO, fractional programming
ferent communication scenarios are given to theory, and the maximum power allocation
examine the capability of the power control strategy is studied in this section. For compar-
strategy based on QSEO. To this end, we study ison purposes, the population size and the iter-
two major concerns in massive MIMO net- ation number of SEO are the same of QSEO.
works, namely, energy efficiency and spectral For SEO, the other parameter settings can
efficiency. refer to [33], while the power control strategy
based on fractional programming theory and
5.1 System parameters
the maximum power allocation strategy are
We consider the hardware-impaired single-cell discussed in [32]. All results are the average of
multiuser massive MIMO uplink network 100 simulations.
as depicted in Section II, where K = 3 and The variation of the system energy efficien-
M = 125. Under the assumption of perfect cy with the number of iterations is shown in
CSI, MRC is adopted for data recovery. We figure 1. The results show that: 1) the max-
assume that only the BS is hardware-impaired imum power allocation strategy causes the
and ε BS represents the degree of impairment. lowest EE. The reason is that though the larg-
M s antennas are selected at the BS, and the est energy is obtained for the desired signal,
antenna selection coefficient ξ varies ac- the SINR in (8) is seriously decreased by the
cording to the requirements of the practical
system. The power of Gaussian white noise is Table I. Parameter settings in massive MIMO system.
σ 2 = FBN 0. Without loss of generality, assum- Parameter Value
ing that the location of the BS is (0,0)m, the System bandwidth B 1MHz
users are distributed in three circles where the Radius of the BS r0 100m
centers are located at (125,200)m, (320,0)m, Path loss exponent v 3.8
(253,103)m with the radius of 2m, 2m, 5m, re- Noise factor F 3dB
spectively. The parameters in massive MIMO Noise power spectral density N 0 −170dBW / Hz
uplink system are shown in table 1. Transmission efficiency η 50%
Circuit power of user k pc , k 0.01W
5.2 Energy efficiency
Power consumption at antenna m pc , m 0.001W
To make it easy, assuming each antenna con- Minimum achievable rate of user k Rkmin 1Mbit / s
sumes the same energy pc=,m pc ∀m in data Minimum achievable rate of the system Rmin 5Mbit / s
processing. All users have the same maximum 1
where 1dB = 101/10
, 1dBW = 101/10
W.
max
transmission power p= k pmax ∀k, the same
circuit power pc=
,k pc ∀k, and the same mini- Table II. Parameter settings of QSEO.
min
mum achievable rate constraint R= Rmin ∀k. Parameter Value
k

For QSEO, all quantum individuals have Population size N 50

different emotions and the emotional index Number of the worst quantum individuals N ′ 3

of each quantum individual is distributed be- Number of binary bits encoded for each estimated variable L 15

tween 0 and 1. At the stage of initialization, Emotional thresholds e1, e2 0.14, 0.58

each element of the quantum individual is set Emotional reducing index ∆e 0.05

to 1/ 2 and the emotional indexes are set to Impact factor of the optimal solution of the quantum population c1 0.3

1. The other parameters of QSEO are shown Impact factor of the historical optimal solution of each quantum
0.6
individual c2
in Table II.
Impact factor of the worst solution set c3 −0.1
For the problem of EE maximization in
Mutation probability κ1 0.1/L

China Communications • April 2019 9

 strong non-negligible interference between the and Lagrangian constraints are needed for
users. 2) For power control strategies based finding a suitable power allocation result. It’s
on QSEO, SEO, and fractional programming only applicable to convex optimization prob-
theory, the EE in massive MIMO uplink net- lem with equality constraints. For the non-con-
works increases with the number of iterations. vex power allocation problem with minimum
3) Compared with other methods, a higher EE rate constraints in massive MIMO uplink net-
is obtained by the power allocation strategy works, it couldn’t get a higher performance,
based on QSEO. However, for the power al- while the resolving method of traditional SEO
location method based on fractional program- is easy to get into a local optimum in the ear-
ming theory, a lot of mathematical derivations ly searching stage. In QSEO, two different
quantum evolution strategies are designed to
guarantee the diversity of solution and the
250
strong searching ability of the algorithm. The
designed evolution rules can make full use of
200 the advantages of quantum theory and SEO,
Power allocation based on QSEO and the interactions of quantum individuals
Power allocation based on SEO
are fully utilized to obtain the optimal solution
efficiency(Mbit/J)

Power allocation based on fractional programming

150 Maximum power allocation of the feasible region, which avoids the local
convergence of traditional algorithms. To sum
100 up, QSEO can overcome the shortcomings of
Energy ef

traditional algorithms and show its advantages

in convergence speed and convergence accu-
50
racy.
Fig. 2. shows the curves of energy efficien-
0 cy for the variation of antenna selection coef-
0 100 200 300 400 500 600 700 800 900 1000
Iteration number
ficient ξ in no hardware impaired system. The
antenna selection coefficient increases as a
Fig. 1. Energy efficiency versus iteration number with ξ = 0.7, pmax = −20dBW,
higher requirement is needed for the commu-
nication system. However, the EE decreases
ε BS = 0.
with the increase of ξ. The reason is that each
antenna consumes energy for data receiving
350
and processing. As the increasing number of
Power allocation based on QSEO
Power allocation based on SEO selected antennas, the power consumed at the
300
Power allocation based on fractional programming
BS increases and the total power consumed in
Maximum power allocation
the massive MIMO system goes up. Besides,
250 the increase of achievable rate is far less than
efficiency(Mbit/J)

energy consumption. Compared with other

200
power control strategies, the power control
strategy based on QSEO has the best perfor-
150
mance.
Energy ef

100
Fig. 3. shows the variation curves of ener-
gy efficiency with hardware impairment coef-
50 ficient at the BS. ε BS is a variable distributed
between 0 and 1, where ε BS = 0 represents no
0
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 hardware impairment at the BS and ε BS = 1 de-
Antenna selection coef
coefficient
notes the BS is completely damaged. In order
to study the effect of the hardware impairment
Fig. 2. Energy efficiency as a function of ξ, with pmax = −20dBW, ε BS = 0.

10 China Communications • April 2019

coefficient on EE, we define dB as the unit parameter settings of power control strategies
of ε BS. We can learn from the curves that EE based on QSEO, SEO, fractional programming
decreases as the increase of hardware impair- theory, and the maximum power allocation
ment. In particular, when ε BS = 0, the EE of all strategy are the same of 5.2 in this section.
curves goes to zero. That is because the power Fig. 6. shows the variation of the system
of the desired signal decreases as ε BS increas- spectral efficiency with the number of itera-
es. When the BS is totally destroyed, the en- tions under different power control strategies.
ergy of the desired signal is zero. The power The simulation results show that SE increases
control strategy based on QSEO is more sen- with the iteration number. The trend of the
sitive to hardware impairment and has a better curves is similar to the curves in figure 1.
performance than other strategies.
The variation of energy efficiency with the 250
Power allocation based on QSEO
maximum transmission power of each user is Power allocation based on SEO
Power allocation based on fractional programming
shown in figure 4. We observe that the QSEO Maximum power allocation
200
power allocation strategy has an obvious
advantage over the other power control strat-
efficiency(Mbit/J)

egies. For the power allocation problem in 150

massive MIMO networks, the solution space
expands with the increase of the maximum
100
transmission power. And it’s difficult for the
Energy ef

strategies based on fractional programming

theory and SEO to find the optimal solutions 50
due to their own limitations of randomness
and local convergence. But for QSEO, the
quantum evolution mechanic and the variety 0
-50 -45 -40 -35 -30 -25 -20 -15 -10 -5 0
of evolution rules can overcome the disadvan- Hardware impairment coef
coefficient(dB)

tages of traditional SEO algorithm and the ap-

proximation method. Hence, the QSEO-based Fig. 3. Energy efficiency as a function of ε BS, with pmax = −20dBW, ξ = 0.7.
power allocation strategy can get an excellent
performance in massive MIMO networks.
In order to see the relationship of two dif-
250
ferent parameters, we show the simulation Power allocation based on QSEO
Power allocation based on SEO
results in another way. The simulation result Power allocation based on fractional programming
Maximum power allocation
which is related to the variation of two system 200

parameters, hardware impairment coefficient

efficiency(Mbit/J)

at the BS ε BS and the maximum transmission

150
power pmax is shown in figure 5. It is clear
to see the impact of ε BS and pmax on energy
100
Energy ef

efficiency based on QSEO power allocation

strategy. A higher pmax and a lower hardware
impairment coefficient are good for improving 50
the EE in massive MIMO networks.

5.3 Spectral efficiency 0

-60 -50 -40 -30 -20 -10 0 10
p max (dBW)
The simulation results for SE maximization in
hardware-impaired massive MIMO networks
are presented in this part. To make it easy, the Fig. 4. Energy efficiency as a function of pmax, with ε BS = 0, ξ = 0.7.

China Communications • April 2019 11

 Compared with power allocation strategies the degree of hardware impairment at the BS
based on SEO, fractional programming theory with the unit of dB. We can learn from the
and the maximum power allocation method, curves that SE decreases with the increase of
the proposed power allocation strategy based hardware impairment coefficient. When the
on QSEO can get a better performance for the BS is totally destroyed, ε BS = 0, the SE of all
SE maximization in massive MIMO networks. power allocation strategies goes to zero. The
Fig. 7. shows the variation curves of spec- power control strategy based on QSEO is
tral efficiency with the hardware impairment more sensitive to hardware impairment and
coefficient. Same as in figure 2, ε BS denotes has a higher SE than other strategies under the

30
200

180
Energy efficiency(Mbit/J)

25
300 160 Power allocation based on QSEO

efficiency(bit/s/Hz)
Power allocation based on SEO
200 140 Power allocation based on fractional programming
20
Maximum power allocation
100 120

100 15
0
Spectral ef

20 80
10
0 60

-20 0 40
5
p max (dBW) -10
-20 20
-40
-30
-40 0 0
-60 -50 0 100 200 300 400 500 600 700 800 900 1000
Hardware impairment coefficient(dB) Iteration number

Fig. 5. Energy efficiency as a function of pmax and ε BS, with ξ = 0.7. Fig. 6. Spectral efficiency versus iteration number with ξ = 0.7,
pmax = −20dBW, ε BS = 0.

30 40
Power allocation based on QSEO Power allocation based on QSEO
Power allocation based on SEO Power allocation based on SEO
Power allocation based on fractional programming 35 Power allocation based on fractional programming
25 Maximum power allocation Maximum power allocation

30
efficiency(bit/s/Hz)
efficiency(bit/s/Hz)

20
25

15 20
Spectral ef
Spectral ef

15
10
10
5
5

0 0
-50 -45 -40 -35 -30 -25 -20 -15 -10 -5 0 -60 -50 -40 -30 -20 -10 0 10
Hardware impairment coef
coefficient(dB) p max (dBW)

Fig. 7. Spectral efficiency as a function of ε BS, with pmax = −20dBW, Fig. 8. Spectral efficiency as a function of pmax, with ε BS = 0, ξ = 0.7.
ξ = 0.7.

12 China Communications • April 2019

condition of BS is not paralyzed.
The variation of spectral efficiency with the
maximum transmission power of each user is 35

shown in figure 8. The SE of the power control

strategy based on QSEO increases with the

Spectral efficiency(Mbit/J)
30
40
increase of the maximum transmission power
30 25
of each user, while other strategies almost no
20
change with the variation of pmax. Power allo-
20
cation strategy based on SEO mechanism and 10

fractional programming theory are trapped 0 15

into the local optimal solution because SEO 20

is vulnerable to local convergence and frac- 0 10

tional programming theory is only suitable

-20 0
for the convex optimization problem. For the -10 5
p max (dBW) -40 -20
maximum power allocation strategy, each -30
-40 0
user transmits signals with the power of pmax, -60 -50
the SINR in (8) tends to a fixed value. But for Hardware impairment coefficient(dB)

QSEO, the quantum evolution mechanic and

the variety of evolution rules can improve the
global searching ability and get the optimal Fig. 9. Spectral efficiency as a function of pmax and ε BS, with ξ = 0.7.
power allocation strategy for maximizing SE
in massive MIMO networks. the efficiency of the transmitter are considered
In order to see the effects of two differ- in practical systems. To guarantee the QoS in
ent parameters on the spectral efficiency, we massive MIMO uplink networks, minimum
show the simulation results in another way. rate constraints of each user and the system are
In figure 9, two system parameters, namely considered in this paper. QSEO algorithm is
hardware impairment coefficient at the BS ε BS proposed to solve the non-convex multi-con-
and the maximum transmission power pmax is strained power allocation problem for EE and
considered. It is clear to see the impact of ε BS SE maximization. Simulation results demon-
strate the superiority of the QSEO-based pow-
and pmax on spectral efficiency based on QSEO
er allocation strategy which previous strategies
power allocation strategy from the simulation
do not have.
results. A higher pmax and a lower hardware
Although power control strategy based on
impairment coefficient are good for improving QSEO enjoys excellent performance in EE
the SE of massive MIMO networks. and SE, there are still some certain commu-
nication systems which can be investigated.
VI. CONCLUSION In the future, we will study the application of
massive MIMO with imperfect CSI, multi-cell
In this paper, we analyze the EE and SE under multiuser networks, cognitive radio, heteroge-
the condition of specific requirement for data neous networks, and co-time co-frequency full
transmission in massive MIMO uplink system. duplex systems.
It is assumed that perfect CSI is obtained, and
MRC is used for data recovery at the BS. For ACKNOWLEDGEMENT
promoting energy conservation, we select part
of antennas at the BS to participate in com- This work was supported by the National
munication according to the requirements of Natural Science Foundation of China (No.
the system. Hardware impairments, energy 61571149), the Special China Postdoctoral
consumed at each antenna, circuit power, and Science Foundation (2015T80325), the Fun-

China Communications • April 2019 13

 damental Research Funds for the Central Uni- [12] T. L. Marzetta, “Noncooperative cellular wireless
with unlimited numbers of base station anten-
versities (HEUCFP201808), the China Post-
nas,” IEEE Transactions on Wireless Communica-
doctoral Science Foundation (2013M530148). tions, vol. 9, no. 11, 2010, pp. 3590-3600.
[13] T. L. Marzetta, “Massive MIMO: An introduc-
References tion,” Bell Labs Technical Journal, vol. 20, 2015,
[1] P. Gandotra, and R. K. Jha, “A survey on green pp. 11-22.
communication and security challenges in 5G [14] X. H. Ge, J. Yang, H. Gharavi, and Y. Sun, “Energy
wireless communication networks,” Journal of efficiency challenges of 5G small cell networks,”
Network and Computer Applications, vol. 96, IEEE Communications Magazine, vol. 55, no. 5,
2017, pp. 39-61. 2017, pp. 184-191.
[2] J. G. Andrews, S. Buzzi, W. Choi, S. V. Hanly, A. [15] X. M. Li, T. Jiang, S. G. Cui, J. P. An, and Q. A.
Lozano, A. C. K. Soong, and J. C. Zhang, “What Zhang, “Cooperative communications based on
will 5G be?,” IEEE Journal on Selected Areas in rateless network coding in distributed MIMO
Communications, vol. 32, no. 6, 2014, pp. 1065- systems,” IEEE Wireless Communications, vol. 17,
1082. no. 3, 2010, pp. 60-67.
[3] A. Murugadass, and A. Pachiyappan, “Pow- [16] P. Pan, H. Wang, Z. Zhao, and Wei Zhang, “How
er-aware fuzzy based joint base station and many antenna arrays are dense enough in mas-
relay station deployment scheme for green sive MIMO systems,” IEEE Transactions on Vehic-
radio communication,” Sustainable Computing: ular Technology, vol. 67, no. 4, 2018, pp. 3042-
Informatics and Systems, vol. 13, 2017, pp. 1-14. 3053.
[4] L. W. Chen, and C. C. Chang, “Cooperative traf- [17] H. Yin, D. Gesbert, M. Filippou, and Y. Liu, “A
fic control with green wave coordination for coordinated approach to channel estimation
multiple intersections based on the internet of in large-scale multiple-antenna systems,” IEEE
vehicles,” IEEE Transactions on Systems, Man, Journal on Selected Areas in Communications,
and Cybernetics, vol. 47, no. 7, 2017, pp. 1321- vol. 31, no. 2, 2013, pp. 264-273.
1335. [18] X. Wei, W. Peng, D. Chen, R. Schober, and T.
[5] W. C. Cheng, X. Zhang, and H. L. Zhang, “Statis- Jiang, “Uplink channel estimation in massive
tical-QoS driven energy-efficiency optimization MIMO systems using factor analysis,” IEEE Com-
over green 5G mobile wireless networks,” IEEE munications Letters, vol. 22, no. 8, 2018, pp.
Journal on Selected Areas in Communications, 1620-1623.
vol. 34, no. 12, 2016, pp. 3092- 3107. [19] T. V. Chien, E. Björnson, and E. G. Larsson, “Joint
[6] P. Gandotra, R. K. Jha, and S. Jain, “Green com- pilot design and uplink power allocation in
munication in next generation cellular networks: multi-cell massive MIMO systems,” IEEE Trans-
A survey,” IEEE Access, vol. 5, 2017, pp. 11727- actions on Wireless Communications, vol. 17, no.
11758. 3, 2018, pp. 2000-2015.
[7] M. J. Feng, S. W. Mao, and T. Jiang, “BOOST: [20] A. K. Papazafeiropoulos, H. Q. Ngo, and T. Rat-
Base station on-off switching strategy for green narajah, “Performance of massive MIMO uplink
massive MIMO HetNets,” IEEE Transactions on with zero-forcing receivers under delayed chan-
Wireless Communications, vol. 16, no. 11, 2017, nels,” IEEE Transactions on Vehicular Technology,
pp. 7319- 7332. vol. 66, no. 4, 2017, pp. 3158-3169.
[8] E. Basar, “Index modulation techniques for 5G [21] L. Liang, W. Xu, and X. Dong, “Low-complexity
wireless networks,” IEEE Communications Maga- hybrid precoding in massive multiuser MIMO
zine, vol. 54, no. 7, 2016, pp. 168-175. systems,” IEEE Wireless Communications Letters,
[9] E. Bjornson, E. G. Larsson, and T. L. Marzetta, vol. 3, no. 6, 2014, pp. 653-656.
“Massive MIMO: Ten myths and one critical [22] S. Qiu, D. Chen, K. Luo, D. Qu, and T. Jiang,
question,” IEEE Communications Magazine, vol. “Downlink precoding with mixed statistical and
54, no. 2, 2016, pp. 114-123. imperfect instantaneous CSI for massive MIMO
[10] C. Mollen, J. Choi, E. G. Larsson, and R. W. systems,” IEEE Transactions on Vehicular Tech-
Heath, “Uplink performance of wideband mas- nology, vol. 67, no. 4, 2018, pp. 3028-3041.
sive MIMO with one-bit ADCs,” IEEE Transac- [23] M. Benmimoune, E. Driouch, W. Ajib, and D.
tions on Wireless Communications, vol. 16, no. Massicotte, “Novel transmit antenna selection
1, 2017, pp. 87-100. strategy for massive MIMO downlink channel,”
[11] X. Yang, W. J. Lu, N. Wang, K. Nieman, C. K. Wen, Wireless Networks, vol. 23, no. 8, 2017, pp.
C. Zhang, S. Jin, X. M. Mu, I. Wong, Y. M. Huang, 2473- 2484.
and X. H. You, “Design and implementation of [24] P. V. Amadori, and C. Masouros, “Large scale an-
a TDD-based 128-antenna massive MIMO pro- tenna selection and precoding for interference
totype system,” China Communications, vol. 14, exploitation,” IEEE Transactions on Communica-
no. 12, 2017, pp. 162-187. tions, vol. 65, no. 10, 2017, pp. 4529-4542.
[25] H. Tang, and Z. P. Nie, “RMV antenna selection

14 China Communications • April 2019

 algorithm for massive MIMO,” IEEE Signal Pro- He is currently an Associate Professor with the Col-
cessing Letters, vol. 25, no. 2, 2018, pp. 239-242. lege of Information and Communication Engineering,
[26] Z. Z. Zhang, Z. Y. Chen, M. Y. Shen, and B. Xia, Harbin Engineering University, China. Areas of his
“Spectral and energy efficiency of multipair current interests include wireless energy harvesting
two-way full-duplex relay systems with massive communications, intelligent computing, software ra-
MIMO,” IEEE Journal on Selected Areas in Com- dio, signal recognition and classification, cognitive
munications, vol. 34, no. 4, 2016, pp. 848-863. radio, array signal processing, LTE-Unlicensed, artifi-
[27] K. N. R. S. V. Prasad, E. Hossain, and V. K. cial intelligence, HetNets in 5G, communication the-
Bhargava, “Energy efficiency in massive MI- ory and image processing, and massive MIMO.
MO-based 5G networks: Opportunities and
challenges,” IEEE Wireless Communications, vol. Yumeng Su, received her B.E.
24, no. 3, 2017, pp. 86-94. degree in Electronic Informa-
[28] G. Z. Xu, A. Liu, W. Jiang, H. G. Xiang, and W. tion Engineering from Harbin
Luo, “Energy-efficient beamforming for two-tier Engineering University, Harbin,
massive MIMO downlink,” China Communica- Heilongjiang, P. R. China, in
tions, vol. 12, no. 10, 2015, pp. 64-75. June 2016. She is currently a
[29] P. Liu, S. Jin, T. Jiang, Q. Zhang, and M. Mat- doctoral student in Harbin En-
thaiou, “Pilot power allocation through user gineering University. Her cur-
grouping in multi-cell massive MIMO systems,” rent research interests include intelligent computing,
IEEE Transactions on Communications, vol. 65, relay selection, power allocation, massive MIMO, se-
no. 4, 2017, pp. 1561-1574. cure communications, co-frequency co-time full du-
[30] E. Björnson, E. G. Larsson, and M. Debbah, plex systems and beyond 5G technologies.
“Massive MIMO for maximal spectral efficiency:
How many users and pilots should be allocat- Shibo Zhang, received his B.E.
ed?” IEEE Transactions on Wireless Communica- degree in Electronic Informa-
tions, vol. 15, no. 2, 2016, pp. 1293-1308. tion Engineering from Harbin
[31] L. Zhao, H. Zhao, F. L. Hu, K. Zheng, and J. X. Engineering University, Harbin,
Zhang, “Energy efficient power allocation al- Heilongjiang, P. R. China, in
gorithm for downlink massive MIMO with MRT June 2016. He is currently a
precoding,” Proc. Veh. Technol. Conf. (Vtc Fall), doctoral student in Harbin En-
IEEE, 2013, pp. 1-5. gineering University. His cur-
[32] A. Zappone, L. Sanguinetti, G. Bacci, E. Jor- rent research interests include relay communication,
swieck, and M. Debbah, “Energy-efficient power intelligent computing, energy harvesting, fog com-
control: A look at 5G wireless technologies,” puting, heterogeneous networks, device-to-device
IEEE Transactions on Signal Processing, vol. 64, communications, Internet of Things and future wire-
no. 7, 2016, pp. 1668- 1683. less networks.
[33] Z. L. Guo, X. Z. Yue, H. G. Yang, K. Liu, and X. S.
Liu, “Enhancing social emotional optimization Ming Diao, was born in 1960.
algorithm using local search,” Soft Computing, He received his B. E. Degree
vol. 21, no. 24, 2017, pp. 7393-7404. and M. E. degree in 1982 and
[34] H. Y. Gao, S. B. Zhang, Y. N. Du, Y. Wang, and 1987 respectively, from Harbin
M. Diao, “Relay selection scheme based on Engineering University. He is
quantum differential evolution algorithm in re- currently a Professor with the
lay networks,” KSII Transactions on Internet and College of Information and
Information Systems, vol. 11, no. 7, 2017, pp. Communication Engineering,
3501-3523. Harbin Engineering University, a senior member of
China Institute of Communications, a board member
Biographies of Committee of Deep Space Exploration Technology,
Hongyuan Gao, received the Chinese Society of Astronautics, and also a member
Ph.D. degree from the Depart- of China Society of Image and Graphics. His current
ment of Communication and research interest includes wide-band signal detec-
Information Systems, College tion, processing and recognition, and signal process-
of Information and Communi- ing for communications.
cation Engineering, Harbin En-
gineering University, China, in
2010. He has been a Visiting
Research Professor with the Department of Comput-
er and Information Science, Korea University, Sejong
Metropolitan City, South Korea, from 2015 to 2016.

China Communications • April 2019 15