A UNIFORM BEAM SELECTION ALGORITHM FOR BEAMSPACE

MIMO-NOMA IN MILLIMETER-WAVE COMMUNICATION SYSTEM

KAI YANG, XIAO YAN, KAIYU QIN, QIAN WANG

School of Aeronautics and Astronautics, University of Electronic Science and Technology of China, Chengdu 611731,
China
E-MAIL: [email protected], [email protected]

Abstract: power-domain on the transmitter side and multi-user signals

The concept of beamspace multiple-input multiple-out can be separated with a successive interference canceller
non-orthogonal multiple access (MIMO-NOMA) is able not (SIC) even in the same beam on the receiver side [4], by
only to reduce the number of radio frequency (RF) chains but which the number of the users in the system can be greater
also to increase the spectrum efficiency in the millimeter-wave than the number of RF chains. Then the sparsity and
(mmWave) massive MIMO communication system. However,
cross-correlation of the user beamspace channels are
it is difficulty to realize the existing beam selection scheme in
the system since the user side will spend too much time and considered to construct a uniform beam selection scheme.
computation in decoding messages by performing successive Simulation results show that not only the spectrum and
interference canceller (SIC). To solve this problem, we energy efficiency but also the computation of the users can
propose a Uniform beam selection algorithm. Specifically, be improved quite a lot.
according to the sparsity of the beamspace channel, the users The rest of the paper is organized as follows. In Section 2,
are classified into two sets firstly. Then the beams are selected the system model is introduced for studying the mmWave
uniformly by the users with the cross-correlation of the MIMO-NOMA scheme. In Section 3, the Uniform beam
beamspace channels. The simulation results verify that not selection algorithm is proposed. In Section 4, the simulation
only the spectrum and energy efficiency but also the
results are shown. Finally, the conclusion is written in
computation of the user side in the system can be improved
quite a lot with the Uniform beam selection algorithm. Section 5.

Keywords: 2. Beamspace MIMO-NOMA System

MmWave communication; Massive MIMO; beamspace;
MIMO-NOMA; Uniform beam selection In this paper, we considers the Saleh-Valenzuela
channel model for mmWave communication [1-3], so the
1. Introduction channel vector h k can be expressed as formula (1)
L
Millimeter-wave (mmWave) massive multiple-input
multiple-output (MIMO) is a promising technology for 5G hk  
l 1
(l ) (l )
k u( k ) (1)
communication system [1]. However it is too complex to be
implemented since it will cost too many Radio-Frequency In (1), L is the total number of the propagation paths. For
(RF) chains [2]. With the concept of Beamspace
the User k in the Path l ,  k(l ) denotes the complex
MIMO-NOMA (Non-orthogonal Multiple Access) being
proposed recently [3-4], the number of RF chains can be gain, u(k(l ) ) , its size N  1 , is a steering vector and u( )
decreased significantly by using a lens antenna array [5], can be expressed as formula (2)
and the number of the users in the system can also be
increased because more than one user can be served by a 1
beam. But this brings a new problem of which beam is the u( )  e j 2 (2)
N
optimal one chosen by the user.
In this paper, we focus on users’ beam selection And we consider a typical downlink massive MIMO
scheme and propose a Uniform beam selection algorithm. communication system, where one base station (BS)
Specifically, multiple users are multiplexed in the equipped with N antennas serves K single-antenna

978-1-7281-1536-8/18/$31.00 ©2018 IEEE 166

Authorized licensed use limited to: Higher College of Technology. Downloaded on March 22,2021 at 07:54:38 UTC from IEEE Xplore. Restrictions apply.
users. The spatial signal domain in the MIMO system can the principle of SIC, the received signal at um ,n in the
be transformed to the beamspace signal domain by a set of
beamspace MIMO-NOMA system is shown as formula (5)
bases, expressed as U  [u(1 ), u( 2 ), , u( N )]H [3], in
m 1
N 1
1

H H
which n  (n  ) ( n  1, 2, , N ) denote the ym,n  h m, n w n pm,n xm,n  h m,n w n pi ,n xi ,n
N 2 i 1
spatial directions. Thus, K users' received signal vector |S j |
(5)

 w
H
y  [ y1 , y2 , , yK ]T in the beamspace MIMO-NOMA  h m, n w n j pi , j xi , j  nm,n
system can be expressed as formula (3) j  n i 1

H In (5), pm,n is the power allocation coefficient and xm,n

y  H WPx  n  H H U H WPx  n (3)
is the message for um , n ; and nm,n ~ CN (0,  2 ) is the
In (3), x  [ x1 , x2 , , xK ]T , its size K 1 , denotes the
AWGN signal at um , n .
signal vector for all the users launched by the BS with
E(xx H )=I K ; P  diag{p} , its size K  K , denotes the According to (5), the received Signal to Interference
plus Noise Ratio (SINR) can be shown as formula (6)
transmit power matrix with p  [ p1 , p2 , , pK ]
H 2
K h m, n w n pm,n
satisfying pk 1
k  P , where P denotes the total  m, n 

2
(6)
m, n
transmit power at the BS; W  [w1 , w 2 , , w K ] , its
The interference is denoted as formula (7)
size N  K , denotes the precoding matrix with ‖w k‖2  1
( k  1, 2, , K ); H  [h1 , h 2 , , h K ] , its size N  K , |S j |
2 m 1 2
  p
H H
denotes the spatial channel matrix where h k , its size  m, n  h m, n w n pi,n  h m, n w j   m2 ,n (7)
i, j
N  1 , denotes the channel vector between the BS and the 2 i 1 j n 2 i 1

User k ; n , its size K 1 , denotes the additive white

Gaussian noise (AWGN); H , its size N  K , denoted as
The data rate at um ,n can be denoted as formula (8)
H  [h1 , h 2 , , h K ] , is the beamspace channel matrix. h k Rm,n  log2 (1   m,n ) (8)
has sparse nature which means an efficient reduction in the
number of RF chains. Therefore, K users' received signal Then the sum-throughput of the beamspace MIMO-NOMA
vector in (3) can be re-expressed as formula (4) system can be expressed as formula (10)
H N RF |Sn |
y  H r Wr Px  n (4) R  R
n 1 m 1
m, n (9)
In (4), H r , its size | M | K , is the reduced beamspace
channel matrix. M is the index set of the beam-selected 3. UNIFORM BEAM SELECTION
number and we have N RF =|M| ; Wr , its size | M | K , is
the reduced precoding matrix of H r . As shown in Section 2, the beam selection scheme has
In beamspace MIMO-NOMA system, the User m in a significant influence on the performance of the
beamspace MIMO-NOMA system. In this section, we study
Beam n is denoted as um , n ( m  1, 2, ,| Sn | ). Let the Uniform beam selection algorithm, in which
h m,n and w n denote the beamspace channel vector and approximated equal number of users are served by each
beam. Not only the spectrum and energy efficiency but also
the corresponding uniform precoding vector of um , n , the computation of the user side can be improved with the
respectively. And without loss of generality, we algorithm, which will be shown in two subsections.
H H H
have h1,n wn  h2,n w n   h|Sn |,n w n . According to

167

Authorized licensed use limited to: Higher College of Technology. Downloaded on March 22,2021 at 07:54:38 UTC from IEEE Xplore. Restrictions apply.
3.1. Classifying Users
Redundant u1,5 u1,6 u1,7 u1,8
We start from determining which beams should be user

Selected Beam index

Selected Beam index
u1,5 u1,7
Fixed user
chosen to construct the reduced beamspace matrix. Sort N u1,6 u7,5

elements in User k ’s beamspace channel vector No user

u1,8
hk  [hk ,1 , hk ,2 , , hk , N ] ( k  1, 2, , K ) in the descending New user

order of magnitude and select the index with the maximal u7,5

value, denoted by bk* ( bk*  (1, 2, , N ) ). Since h k has

User index User index
sparse nature, it is easy for us to know that beam bk* (a) (b)

carries the main power for User k . Merge the same indexes Fig.1 Uniform beam selection algorithm (a) Stage 1:
in the beam set {bk* }kK1 , then we can get the reduced Classifying Users; (b) Stage 2: Beam Selection
beamspace channel vector set denoted as As w  [ w1 , w2 , , wg ] ( g  K ) and these users’
* K
M ( M  {bk }k 1 , N RF  M ), which can be utilized by the beamspace channel matrix of size ( N RF  |  w |) can be
reduced beamspace matrix.
expressed as Hw  (hw1 , hw2 , , hwg ) . The beams in which
According to the description above, it is possible for
the number of the users is less than D are denoted as
different users to select the same beam ( bi*  b*j , i  j ).
  (n1 , n2 , , n f ) ( f  K ) and their equivalent beamspace
Let's consider the case where k1 users are served by Beam
channel matrix of size ( N RF  |  |) can be expressed
B1 and k 2 users are served by Beam B2 ( k1 k2 ).
as H1,  (h1,n1 , h1,n2 , , h1,n f ) . By the way of calculating
According to the SIC principle in NOMA, the users in
Beam B1 will spend more time and energy in decoding the cross-correlation function between the channel vectors
messages than the users in Beam B2 . from H w and the channel vectors from H1, , the
To balance the decoding time and computation of the cross-correlation function matrix, its size |  |  |  w | , can be
users in different beams, the maximum number of the users denoted as formula (10)
in each beam should be restricted to D   K / | M | . H H H H
C  H1, Hw  [H1, h w1 , H1, h w2 , , H1, h wg ] (10)
Therefore, we divide the K users into two sets. One set
belongs to the restriction and the other set doesn’t belongs The index of the maximum value in each column of
to the restriction. They are called as “fixed users”, denoted C denotes the sub-optimal beam for the selected user. And
by  s and “redundant users”, denoted by  w , if one beam is selected by two users simultaneously, the
user with the greater value will be served. And algorithm
respectively. Specifically, for the Beam n , the D users
can be shown as Fig.1.
H
which have the most value of hm,n w n are distributed to
4. SIMULATION RESULTS
the  s set and the others are distributed to the  w set if
| Sn | D ; and if | Sn | D , all the users in the Beam n In this section, numerical results are provided to
are distributed to the  s set. illustrate the performance of the Uniform beam selection
algorithm proposed. We compare the proposed algorithm
with the “Fully digital beam selection” algorithm, the
3.2. Beam Selection “Sparsity beam selection” algorithm [3], the “IA beam
selection” algorithm [5], and the “MIMO-OMA beam
In this stage, the N RF beams are selected by the selection” algorithm [6]. The parameters of the
users according to their channels’ sparsity and correlation. communication system are set up as: N  256 , K  32 ,
The original beams are selected by the users in the  s set, L  3 , k(1) ~ CN (0,1) , k(l ) ~ CN (0,101 ) ( l  2,3 ), and
and the other sub-optimal beams are selected by the users in
the  w set according to the channel cross-correlation
k(l ) U (1/ 2,1/ 2) ( l  1, 2,3 ).
Fig.2 (a) depicts Spectrum efficiency against SNR,
matrix. Specifically, the users in  w set can be expressed

168

Authorized licensed use limited to: Higher College of Technology. Downloaded on March 22,2021 at 07:54:38 UTC from IEEE Xplore. Restrictions apply.
 250
Fully digital
20
Fully digital 5. Conclusions
IA beam IA beam
200 Sparsity beam Sparsity beam
MIMO-OMA 15 MIMO-OMA In this paper, we have proposed a new Uniform beam
Uniform beam Uniform beam
selection algorithm for the mmWave massive
Spectral efficiency

Energy efficiency
150
MIMO-NOMA communication system. Specifically, we
10
exploit the channel sparity and cross-correlation of the
100
users and find the suitable channels for them. The
5 numerical results have illustrated that not only the spectrum
50
and energy efficiency but also the computation of the users
0
can be improved with the proposed algorithm.
0
0 10 20 0 10 20
SNR (dB) SNR (dB)
(a) (b) Acknowledgements
Fig.2 (a) Spectrum efficiency versus SNR; (b) Energy
efficiency versus SNR This paper is supported by the National Nature
Science Foundation of China under Grant No.61601091
180
Sparsity beam and No.61801093.
160 Uniform beam
Number of SIC calculations

140
References
120

100
[1] Wenyan Ma, and Chenhao Qi, “Beamspace Channel
80 Estimation for Millimeter Wave Massive MIMO
60 System With Hybrid Precoding and Combining”,
40 IEEE Transactions on Signal Processing, Vol. 66, No.
20
30 40 50 60 70 80 90 100
18, pp. 4839-4853, Sep. 2018.
Number of Users [2] Bichai Wang, Linglong Dai, Zhaocheng Wang, Ning
Fig.3 The number of SIC calculations versus the number of Ge, and Shidong Zhou, “Spectrum and
Users Energy-Efficient Beamspace MIMO-NOMA for
and Fig.2(b) depicts Energy efficiency against SNR. In the Millimeter-Wave Communications Using Lens
two figures, we can find that the system with the “Fully Antenna Array”, IEEE Journal on Selected Areas in
digital beam selection” algorithm has the highest spectrum Communications, Vol. 35, No. 10, pp. 2370-2382, Oct.
efficiency, but its energy efficiency is the lowest compared 2017.
with the system with the other algorithms. And the system [3] Bichai Wang, Linglong Dai, Xiqi Gao, and Lajos
with the proposed Uniform beam selection algorithm has Hanzo, “Beamspace MIMO-NOMA for
higher spectrum efficiency and energy efficiency than the Millimeter-Wave Communications Using Lens
system with the “MIMO-OMA beam selection” and the “IA Antenna Arrays”, Proceeding of VTC-Fall, Toronto,
beam selection” algorithms. We can also find that the pp. 1-5, Nov. 2017.
system with the “Sparsity beam selection” algorithm has a [4] Zhiguo Ding, Fumiyuki Adachi, H. Vincent Poor,
higher spectrum and energy efficiency than the system with “The Application of MIMO to Non-Orthogonal
the proposed algorithm. This is because all the users with Multiple Access”, IEEE Transactions on Wireless
the “Sparsity beam selection” algorithm select the strongest Communications, Vol. 15, No. 1, pp. 537-552, Jan.
beams for themselves. However this will increase the 2016.
computation burden on the user sides, which can be shown [5] Xinyu Gao, Linglong Dai, Zhijie Chen, Zhaocheng
in Fig.3. Wang, and Zhijun Zhang, “Near-Optimal Beam
Fig.3 depicts the Number of SIC calculations against Selection for Beamspace MmWave Massive MIMO
the Number of Users in the system. In the figure, we can Systems”, IEEE Communications Letters, Vol. 20, No.
find that the number of SIC calculations on the user sides 5, pp. 1054-1057, May. 2016.
with the Uniform beam selection algorithm is lower than [6] Md Shipon Ali, Hina Tabassum, and Ekram Hossain
that with the “Sparsity beam selection” algorithm on the “Dynamic User Clustering and Power Allocation for
user sides. Thus, by using the Uniform beam selection Uplink and Downlink Non-Orthogonal Multiple
algorithm, the computation burden on the user sides can be Access (NOMA) Systems”, IEEE Access, Vol. 4, pp.
reduced quite a lot in the system. 6325-6343, 2013.

169

Authorized licensed use limited to: Higher College of Technology. Downloaded on March 22,2021 at 07:54:38 UTC from IEEE Xplore. Restrictions apply.