You are on page 1of 5

Capacity-Based User Selection Algorithm for

Downlink Beamforming Non-Orthogonal Multiple


Access System

Abdelsalam Sayed-Ahmed1 , Maha Elsabrouty2


Department of Electronics and Communications
Egypt-Japan University of Science and Technology
New Borg El-Arab, Alexandria, Egypt
Email: 1 abdelsalam.ahmed@ejust.edu.eg, 2 maha.elsabrouty@ejust.edu.eg

Abstract—Non-Orthogonal Multiple Access (NOMA) is a key Recently, many researchers have directed their attention
multiple access technique for 5th generation (5G) mobile commu- towards merging the concepts of NOMA and ZFBF. Existing
nication systems. In this paper, we propose a capacity-based user user selection techniques for NOMA systems such as NOMA-
selection (NOMA-CUS) algorithm for multiuser multiple input based zero-forcing beamforming (NOMA-ZFBF) scheme [6]
single output (MISO) downlink NOMA system with zero-forcing and NOMA-based ZFBF using semi-orthogonal user selection
beamforming. NOMA-CUS algorithm improves the sum rate of
the NOMA system by selecting the users in each cluster based
and matching user selection (NOMA-SUS-MU) scheme [7]
on the system sum rate improvement. Simulation results indicate firstly select the ZFBF users, i.e. users whose channel coef-
that NOMA-CUS algorithm gives a very close performance to that ficients are used for obtaining the zero-forcing beamforming
obtained by the exhaustive search with acceptable complexity and (ZFBF) vectors, greedily based on their channel norm param-
provides better system sum rate and average matched user data eters using Frobenius norm-based algorithms and then select
rate than that of the other existing user selection algorithms. their NOMA matched users.
Keywords—Non-orthogonal multiple access; zero-forcing beam- In this paper, we propose a user selection algorithm to
forming; capacity-based user selection; multiple input single output maximize the NOMA system sum rate using the concepts
(MISO); 5G mobile communication systems of the capacity-based user selection beamforming techniques.
The proposed NOMA capacity-based user selection (NOMA-
I. I NTRODUCTION CUS) algorithm selects the users in each cluster such that
the NOMA system sum rate is maximized and becomes very
One of the promising technologies for increasing the close to exhaustive search user selection sum rate but with
system capacity in upcoming 5G communication systems is much lower computational complexity as confirmed by the
non-orthogonal multiple access (NOMA) [1]. In downlink complexity estimation developed.
NOMA systems, the message signals of the different users
are multiplexed in the power domain with different power The rest of the paper is organized as follows: Section II
levels, i.e. large power to users with weak channel conditions presents the system model of the NOMA with beamform-
and less power to users with good channel conditions, at the ing system. Section III shows the concepts of the existing
transmitter and successive interference cancellation (SIC) is NOMA with beamforming schemes. In section IV, the pro-
used for multiuser detection (MUD) at the receivers [2]. posed NOMA-CUS algorithm is presented, and its compu-
tational complexity analysis is performed. Simulation results
On the other hand, another trend to increase the system sum are presented in section V. Finally, conclusions are drawn in
rate in orthogonal multiple access (OMA) systems is zero- section VI.
forcing beamforming (ZFBF). In ZFBF, the cell is divided
into a number of clusters equal to the number of transmit Notations: Matrices and vectors are referred to by upper-
antennas and each user message is multiplied by a beamform- case and lowercase boldface letters, respectively. The operators
ing (BF) vector and transmitted through multiple antennas. (.)T and (.)∗ denote the transpose and conjugate transpose,
If the beamforming vectors are optimally selected, the mu- respectively. khk denotes the Euclidean vector norm of h. H†
tual interference between users’ signals will be eliminated. stands for the Moore-Penrose pseudo-inverse. |U | stands for
However, obtaining the optimal beamforming vectors is a the cardinality of the candidate user set U . U \ S means the
complicated nonconvex optimization problem [3]. Therefore, set difference that deletes the elements of S from U . R(x, y)
many suboptimal beamforming techniques were proposed for denotes the system sum rate for the ZFBF user set x and the
effectively selecting the beamforming user set that maximizes matched user set y.
the sum rate. These beamforming techniques are divided into
two main categories: 1) Frobenius norm-based algorithms such II. S YSTEM M ODEL
as the semi-orthogonal user selection (SUS) algorithm [4] and
2) capacity-based algorithms such as the greedy user selection In this paper, we consider a multiple input single output
with swap (GUSS) algorithm [5]. (MISO) downlink multiuser mobile communication system

978-1-5386-0643-8/17/$31.00 ©2017 IEEE

Authorized licensed use limited to: Higher College of Technology. Downloaded on March 23,2021 at 07:06:43 UTC from IEEE Xplore. Restrictions apply.
that contains a single base station (BS) that is equipped with Table I: The proposed NOMA-CUS Algorithm
M transmit antennas and K users whose mobile equipment
1) Let SZ = ∅, SM = ∅, S = SZ ∪ SM and RS = 0.
have one receive antenna (K ≥ 2M ). The BS is assumed Let Add = 1 and Del = 1.
to have complete channel state information (CSI) for all users. while (Add == 1) and (|SZ | < M ) or
The modeled system uses zero-forcing beamforming (ZFBF) in (Del == 1) and (|SZ | > 1) do
which the BS sends M simultaneous beams and each beam is while (Add == 1) and (|SZ | < M ) do
directed to a cluster of L users. The leader is called ZFBF user 2) Add Stage:
whose channel vector is used to find the beamforming vectors a) Search for a new ZFBF user :
and the others are called matched users. Let the ZFBF user set KZ = arg max R(SZ ∪ {kz } , SM )
kz ∈U \S
be SZ and the matched user set be SM . Denote the candidate b) Search for a new matched user to KZ :
user set
Pas U whose cardinality is K, i.e. |U | = K. The BS KM = arg max R(SZ ∪ {KZ } , SM ∪ {km })
M km ∈U \(S∪{KZ })
sends k=1 wk,1 xk where xk is the sum of the transmitted c) Take a decision:
signals for the ZFBF and matched users in cluster k, i.e. if R(SZ ∪ {KZ } , SM ∪ {KM }) > R(SZ ∪ {KZ } , SM )
L p
xk =
P
αk,i P Sk,i , and wk,1 is (M × 1) ZFBF vector and R(SZ ∪ {KZ } , SM ∪ {KM }) > RS then
i=1 RS = R(SZ ∪ {KZ } , SM ∪ {KM })
for the k-th cluster. Sk,i , i = 1, 2, ..., L is the normalized SZ ← SZ ∪ {KZ }, SM ← SM ∪ {KM } and Del = 1
transmitted signal of the i-th user in the k-th cluster, i.e. else if
with unit energy, and P is the total transmit power in each R(SZ ∪ {KZ } , SM ) > R(SZ ∪ {KZ } , SM ∪ {KM })
cluster. αk,i is the power allocation factor for the i-th user and R(SZ ∪ {KZ } , SM ) > RS then
in the k-th cluster and 0 < αk,i < 1. The ZFBF vectors are RS = R(SZ ∪ {KZ } , SM )
SZ ← SZ ∪ {KZ }, and Del = 1
generated Tbased on the
T channel vectors of the ZFBF users, else
T
H1 = h1,1 ... hM,1 where hm,1 is the 1 × M channel Add = 0
vector for the ZFBF user in the m-th cluster, as follows: end if
end while
W1 = [w1,1 ... wM,1 ] = H†1 = H∗1 (H1 H∗1 )−1 (1) while (Del == 1) and (|SZ | > 1) do
3) Delete Stage:
where hk,1 wm,1 = 0 for 1 ≤ m, k ≤ M, k 6= m and wm,1 is a) Search for a redundant matched user:
obtained by normalizing the m-th column of W1 . L = arg max R(SZ , SM \ {l})
l∈SM
b) Search for a redundant cluster:
In this paper, the channel model is assumed to be Rayleigh {LZ , LM } = arg max R(SZ \ {lz } , SM ∪ {lm })
fading with zero mean and unit variance. The noise is assumed lz ∈SZ ,lm ∈SM
to be independent and identically distributed (i.i.d) additive c) Take a decision:
white complex Gaussian noise with zero mean and variance if R(SZ \ {LZ } , SM ∪ {LM }) > R(SZ , SM \ {L})
and R(SZ \ {LZ } , SM ∪ {LM }) > RS then
of σn2 .
RS = R(SZ \ {LZ } , SM ∪ {LM })
SZ ← SZ \ {LZ }, SM ← SM \ {LM } and Add = 1
else if R(SZ , SM \ {L}) > R(SZ \ {LZ } , SM \ {LM })
III. E XISTING NOMA WITH BEAMFORMING SCHEMES
and R(SZ , SM \ {L}) > RS then
In [6], the NOMA-ZFBF algorithm reduces the inter- RS = R(SZ , SM \ {L})
SM ← SM \ {L}, and Add = 1
cluster interference by selecting the highly correlated users as
else
candidates to be set in the same cluster. The algorithm tries Del = 0
to reduce the inter-user interference by selecting from these end if
highly correlated users those who have the largest channel end while
gain difference to be clustered together. The strong user in end while
each cluster is called ZFBF user and the weak user is called
the matched user. The ZFBF vectors are generated based on IV. P ROPOSED NOMA-CUS T ECHNIQUE
the channel vectors of the strong users as described in (1).
Consequently, the inter-cluster interference is removed. A. NOMA-CUS Algorithm Structure
We propose a NOMA user selection algorithm (NOMA-
In [7], the NOMA-SUS-MU algorithm divides the cluster-
CUS) utilizing the concepts of the capacity-based user se-
ing process into two stages: 1) Selecting the ZFBF users using
lection algorithms that were proposed for OMA with ZFBF
the conventional SUS algorithm [4] and the ZFBF vectors
systems such as GUSS algorithm [5]. The proposed algorithm
are calculated as described in (1). 2) Choosing the matched
tries to greedily select the ZFBF and matched user sets that
users using the MU algorithm by selecting users that bear the
will provide the maximum sum rate to be clustered together.
least inter-cluster interference. Consequently, the inter-cluster
During our development, we consider two users in each cluster,
interference is removed at the ZFBF users and minimized at
similar to the existing schemes in [6], [7]. The NOMA-CUS
the matched users. In all NOMA with beamforming schemes,
algorithm is presented in Table I. Let U = {1, ..., K} be the
SIC is applied to remove the inter-user interference. Simulation
index set of all the candidate users, SZ and SM be the index
results in [7] show that the system sum rate and matched users
sets of the selected ZFBF and matched users, respectively and
sum rate of NOMA-SUS-MU scheme are better than those of
S be the index set of all the selected users, i.e. S = SZ ∪ SM .
NOMA-ZFBF scheme. Therefore, in this paper, the proposed
NOMA-CUS scheme performance will be compared against The algorithm starts with empty selected user sets, i.e.
that of NOMA-SUS-MU scheme. SZ = ∅ and SM = ∅, and zero sum rate, i.e. RS = 0.

Authorized licensed use limited to: Higher College of Technology. Downloaded on March 23,2021 at 07:06:43 UTC from IEEE Xplore. Restrictions apply.
The algorithm is divided into two stages: add and delete. The computational complexity of the NOMA-CUS algo-
During the add stage, the algorithm searches for a new ZFBF rithm is equal to the sum of the computational complexity of
user KZ that when added to SZ provides the highest sum the add stage and that of the delete stage. Let n1 = |SZ | and
rate R(SZ ∪ {kz } , SM ) for kz ∈ U \ S and then searches n2 = |SM | be the size of the ZFBF and matched user index
for a new matched user KM that when added to SM gives sets, respectively, at the start of any step in the NOMA-CUS
the highest sum rate R(SZ ∪ {KZ } , SM ∪ {km }) for km ∈ algorithm. The complexity of one add iteration is due to:
U \(S ∪{KZ }). The algorithm compares R(SZ ∪{KZ } , SM ), 1) (K − (n1 + n2 )) rate calculations for selecting the ZFBF
R(SZ ∪ {KZ } , SM ∪ {KM }) and RS to take one of the user index that when added to SZ could maximize the sum
following decisions: 1) if R(SZ ∪ {KZ } , SM ) is the highest rate. Each sum rate calculation is done in two steps:
data rate, then the user index KZ is added to the index set i) Calculating the beamforming vectors based on the ZFBF
SZ . 2) if R(SZ ∪ {KZ } , SM ∪ {KM }) is the highest data users channel vectors Wm,1 and the matched users channel
rate, then the user indices KZ and KM are added to the index vectors Wm,2 that requires one ((n1 + 1) × M ) and one
sets SZ and SM , respectively. 3) if RS is the highest one, no (n2 × M ) matrix pseudo-inversion operations, respectively.
users are added. The add stage is repeated until M ZFBF users ii) Calculating (n1 + 1) power allocation factors. The com-
are selected, i.e. |SZ | = M or the sum rate decreases and the plexity of calculating each power allocation factor is due to
algorithm goes to the delete stage. (n1 + 3) (1 × M ) × (M × 1) vector multiplications.
2) One ((n1 +1)×M ) matrix pseudo-inversion for calculating
In the delete stage, the algorithm tries to delete the redun-
Wm,1 for the selected ZFBF index set.
dant users and maximize the sum rate. The algorithm searches
3) (K − (n1 + n2 ) − 1) rate calculations for selecting the
for a redundant matched user L that when deleted from SM
matched user index that when added to SM could maximize
gives the highest sum rate R(SZ , SM \ {l}) for l ∈ SM and
the sum rate. Each sum rate calculation is done in two steps:
then searches for a redundant cluster of users {LZ , LM } that
i) Calculating the beamforming vectors Wm,2 that requires
when deleted from SZ and SM , respectively gives the highest
one ((n2 + 1) × M ) matrix pseudo-inversion.
sum rate R(SZ \ {lz } , SM \ {lm }) for lz ∈ SZ and lm ∈ SM
ii) Calculating (n1 + 1) power allocation factors. The com-
and lz and lm belong to the same cluster. The algorithm com-
plexity of calculating each power allocation factor is due to
pares R(SZ , SM \ {L}), R(SZ \ {LZ } , SM \ {LM }) and RS
(n1 + 3) (1 × M ) × (M × 1) vector multiplications.
to take one of the following decisions: 1) if R(SZ , SM \ {L})
The maximum computational complexity of one add stage is
is the highest data rate, then the user index L is deleted from
the sum of the complexity of the M add iterations under the
the index set SM . 2) if R(SZ \ {LZ } , SM \ {LM }) is the
assumption that n1 = n2 .
highest data rate, then the user indices LZ and LM are deleted
from the index sets SZ and SM , respectively. 3) if RS is the
highest one, no users are deleted. The delete stage is repeated
The complexity of one delete iteration is due to:
until there is only one remaining ZFBF user in the system, i.e.
1) One (n1 × M ) matrix pseudo-inversion for calculating
|SZ | = 1 or the sum rate decreases. The algorithm oscillates
Wm,1 .
between the add and delete stages until there is no sum rate
2) n2 rate calculations to select the redundant matched user
improvement in both stages.
index that when deleted from SM may maximize the sum rate.
Hint: The power allocation factors and data rates calculations
Each sum rate calculation is done in two steps:
used in the NOMA-CUS algorithm are performed using the
i) Calculating the beamforming vectors Wm,2 that requires
equations deriven in [7].
one ((n2 − 1) × M ) matrix pseudo-inversion.
B. Computational Complexity Analysis ii) Calculating (n1 ) power allocation factors. The complexity
of calculating each power allocation factor is due to (n1 + 2)
In this section, the computational complexity of the pro- (1 × M ) × (M × 1) vector multiplications.
posed NOMA-CUS scheme and that of the NOMA-SUS-MU 3) n1 rate calculations to select the redundant cluster that when
scheme are evaluated. The basic operations used in these deleted may maximize the sum rate. Each sum rate calculation
two schemes are vector multiplication, vector 2-norm, vector- is done in two steps:
matrix multiplication and beamforming vectors calculation [8]. i) Calculating the beamforming vectors Wm,1 and Wm,2 that
The flop count for each of the (1 × M ) × (M × 1) vector requires one ((n1 − 1) × M ) matrix pseudo-inversion and one
multiplication, and (1 × M ) or (M × 1) vector 2-norm is ((n2 − 1) × M ) matrix pseudo-inversion, respectively.
M flops. The flop count for the (1 × M ) × (M × M ) ii) Calculating (n1 − 1) power allocation factors. The com-
vector-matrix multiplication is M 2 flops. The beamforming plexity of calculating each power allocation factor is due to
vectors calculation is performed in two steps: First, obtaining (n1 + 1) (1 × M ) × (M × 1) vector multiplications.
the channel matrix pseudo-inversion and second, normalizing The maximum computational complexity of one delete stage
the resulting matrix columns. The (N × M ) matrix pseudo- is the sum of the complexity of the possible (M − 1) delete
inversion consists of 2 matrix multiplications and one matrix iterations under the assumption that n1 = n2 .
inversion. The normalizing operation requires calculating N
(M × 1) vector 2-norm operations. Thus, the flop count for
the beamforming vectors calculation is 2M N 2 + M N + N 3 Since the algorithm may oscillate between the add and
flops. delete stages, the complexity of one add stage and one delete
The computational complexity of the NOMA-SUS-MU stage should be multiplied by a and b factors which are the
scheme was analyzed in [9] and its flop count is given by average number of add and delete stage repetitions till the
algorithm is finished, respectively. The values of a and b
FS ≈ O(KM 3 ) (2) depend on M , K, P and the candidate users channel model.

Authorized licensed use limited to: Higher College of Technology. Downloaded on March 23,2021 at 07:06:43 UTC from IEEE Xplore. Restrictions apply.
20 90
NOMA-Ex
NOMA-CUS
TDMA-Ex
80 TDMA-CUS
NOMA-CUS
NOMA-SUS-MU
TDMA-CUS
TDMA-SUS-MU
NOMA-SUS-MU 70
M=8
TDMA-SUS-MU
Sum rate (bps/Hz)

Sum rate (bps/Hz)


15 60

50

40
M=4

10 30

20 M=2

10

5 0
20 30 40 50 60 70 80 90 100 20 40 60 80 100 120 140 160
Number of users Number of users

Fig. 1: Sum rate performance of NOMA-Ex and NOMA-CUS algorithms Fig. 2: Sum rate performance of NOMA-CUS, NOMA-SUS-MU algorithms
versus the number of users K at M = 2, and SN R = 15 dB. versus the number of users K at M = 2, 4 and 8 clusters, and SN R = 15
dB.
Therefore, NOMA-CUS algorithm flop count is
5
M −1 NOMA-CUS
M=4 M=2
X TDMA-CUS
2 M=8
FC = a [(K − (n1 + n2 ))(2M (n1 + 1)

Average matched user rate (bps/Hz)


NOMA-SUS-MU
4.5
TDMA-SUS-MU
n1 =n2 =0
+ M (n1 + 1) + (n1 + 1)3 + (n1 + 1)(n1 + 3)M ) 4

+ 2M (n1 + 1)2 + M (n1 + 1) + (n1 + 1)3


3.5
+ (K − (n1 + n2 ) − 1)((2M (n2 + 1)2
+ M (n2 + 1) + (n2 + 1)3 ) + (n1 + 1)(n1 + 3)M )] 3

XM
+b [2M n21 + M n1 + n31 + n2 (2M (n2 − 1)2 2.5
n1 =n2 =2
+ M (n2 − 1) + (n2 − 1)3 + n1 (n1 + 2)M ) 2
20 40 60 80 100 120 140 160
+ n1 (2M (n1 − 1)2 + M (n1 − 1) + (n1 − 1)3 Number of users

+ 2M (n2 − 1)2 + M (n2 − 1) + (n2 − 1)3 Fig. 3: Average matched user rate performance of NOMA-CUS, NOMA-SUS-
MU algorithms versus the number of users K at M = 2, 4 and 8 clusters,
+ (n1 − 1)(n1 + 1)M )] and SN R = 15 dB.
≈ O(3.4aKM 4 + (2.6b − 5.2a)M 5 ) (3)
for all values of K. The sum rate of the NOMA-CUS scheme
From this complexity analysis, the proposed NOMA-CUS is higher than that of its TDMA-CUS version and this proves
scheme has polynomial time computational complexity and the contribution of the NOMA concepts. Furthermore, NOMA-
its complexity is slightly higher than that of NOMA-SUS- SUS-MU scheme performance is leading its TDMA-SUS-MU
MU scheme. Although the NOMA-SUS-MU scheme has lower version only for M = 2. At M = 4 and 8, NOMA-SUS-MU
computational complexity as most of the Frobenius norm scheme does not provide any sum rate improvement over its
algorithms, it does not guarantee a substantial rate increase in TDMA-SUS-MU version. In Fig.3, the NOMA-CUS is shown
each iteration. Therefore, the higher computational complexity to have a higher average matched user rate than that of NOMA-
of NOMA-CUS scheme is acceptable as a price to guarantee SUS-MU scheme for all values of K.
a positive rate increment in each iteration and to have a very Fig.4 indicates that the sum rate of the proposed NOMA-
close performance to that of the exhaustive search. CUS scheme is better than that of NOMA-SUS-MU scheme
for all values of M . The sum rate of the NOMA-CUS scheme
V. S IMULATION R ESULTS is higher than that of its TDMA-CUS version and this proves
In this section, the simulation results of the proposed that the proposed NOMA-CUS scheme works efficiently at any
NOMA-CUS algorithm performance compared to the NOMA- number of clusters. Furthermore, NOMA-SUS-MU scheme
SUS-MU and exhaustive search user selection (NOMA-Ex) are performance is leading its TDMA-SUS-MU version only for
presented versus the number of candidate users K, the number M = 2. At M > 2, NOMA-SUS-MU scheme does not provide
of clusters M and the transmit signal to noise ratio (SN R). any sum rate improvement over its TDMA-SUS-MU version.
Fig.1 shows the sum rate performance of the three schemes In Fig.5, the NOMA-CUS average matched user rate is shown
as K increases at M = 2. The results show that the proposed to be higher than that of NOMA-SUS-MU scheme for all
NOMA-CUS scheme sum rate performance is very close to values of M . It is noticed that for M > 2, NOMA-SUS-MU
that of the NOMA-Ex scheme. average matched user rate is decreased with increasing M .
Fig.2 shows that the sum rate of the proposed NOMA- Fig.6 indicates that the sum rate of the proposed NOMA-
CUS scheme is better than that of NOMA-SUS-MU scheme CUS scheme is better than that of NOMA-SUS-MU scheme

Authorized licensed use limited to: Higher College of Technology. Downloaded on March 23,2021 at 07:06:43 UTC from IEEE Xplore. Restrictions apply.
90 120
NOMA-CUS
TDMA-CUS NOMA-CUS
80
NOMA-SUS-MU 100 TDMA-CUS
TDMA-SUS-MU NOMA-SUS-MU
70 TDMA-SUS-MU
M=8
Sum rate (bps/Hz)

Sum rate (bps/Hz)


80
60

50 60
M=4
40
40

30 M=2

20
20

10 0
2 3 4 5 6 7 8 9 10 5 10 15 20 25 30
Number of clusters SNR (dB)

Fig. 4: Sum rate performance of NOMA-CUS, NOMA-SUS-MU algorithms Fig. 6: Sum rate performance of NOMA-CUS, NOMA-SUS-MU algorithms
versus the number of clusters M at K = 100 users and SN R = 15 dB. versus the SNR at K = 100 users and M = 2, 4 and 8 clusters.

5 8
NOMA-CUS NOMA-CUS
TDMA-CUS TDMA-CUS
7
Average matched user rate (bps/Hz)

Average matched user rate (bps/Hz)


NOMA-SUS-MU NOMA-SUS-MU
4.5 M=8
TDMA-SUS-MU TDMA-SUS-MU
6
M=4
4
5 M=2

3.5 4

3
3

2
2.5
1

2 0
2 3 4 5 6 7 8 9 10 5 10 15 20 25 30
Number of clusters SNR (dB)

Fig. 5: Average matched user rate performance of NOMA-CUS, NOMA-SUS- Fig. 7: Average matched user rate performance of NOMA-CUS, NOMA-SUS-
MU algorithms versus the number of clusters M at K = 100 users and MU algorithms versus the SNR at K = 100 users and M = 2, 4 and 8
SN R = 15 dB. clusters.

for all values of the SN R. In Fig.7, the average matched user [2] Z. Ding, Z. Yang, P. Fan, and H. V. Poor, “On the Performance of Non-
rate for NOMA-CUS scheme is higher than that of NOMA- Orthogonal Multiple Access in 5G Systems with Randomly Deployed
SUS-MU scheme for all values of the SN R. It is noticed Users,” IEEE Signal Process. Lett., vol. 21, no. 12, pp. 1501–1505, Dec
2014.
that NOMA-SUS-MU average matched user performance for
[3] S. Venkatesan and H. Huang, “System Capacity Evaluation of Multiple
M = 8 is less than that at M = 2 and 4. Antenna Systems Using Beamforming and Dirty Paper Coding,” Bell
Labs, 2003.
VI. C ONCLUSION [4] T. Yoo and A. Goldsmith, “On the Optimality of Multiantenna Broadcast
Scheduling Using Zero-Forcing Beamforming,” IEEE J. Sel. Areas
A capacity-based clustering algorithm for beamforming Commun., vol. 24, no. 3, pp. 528–541, March 2006.
NOMA is presented. The proposed NOMA-CUS algorithm [5] S. Huang, H. Yin, J. Wu, and V. C. M. Leung, “User Selection for
efficiently selects the ZFBF and matched users that will be Multiuser MIMO Downlink with Zero-Forcing Beamforming,” IEEE
clustered together to maximize the system sum rate with Trans. Veh. Technol., vol. 62, no. 7, pp. 3084–3097, Sept 2013.
acceptable complexity. The algorithm provides very close [6] B. Kimy, S. Lim, H. Kim, S. Suh, J. Kwun, S. Choi, C. Lee, S. Lee,
performance to the exhaustive search clustering formation and and D. Hong, “Non-Orthogonal Multiple Access in a Downlink Mul-
a much improved performance, especially in system sum rate tiuser Beamforming System,” in MILCOM 2013 - 2013 IEEE Military
Communications Conference, Nov 2013, pp. 1278–1283.
and average matched user rate over the existing algorithms.
[7] S. Liu, C. Zhang, and G. Lyu, “User Selection and Power Schedule for
Downlink Non-Orthogonal Multiple Access (NOMA) System,” in 2015
ACKNOWLEDGMENT IEEE International Conference on Communication Workshop (ICCW),
June 2015, pp. 2561–2565.
This work has been supported by Egypt Japan University of [8] C. D. Meyer, Matrix Analysis and Applied Linear Algebra. Philadelphia,
Science and Technology (E-JUST), and the Egyptian Ministry PA, USA: SIAM, 2000.
of Higher Education (MOHE). [9] A. Sayed-Ahmed and M. Elsabrouty, “User Selection and Power Al-
location for Guaranteed SIC Detection in Downlink Beamforming Non-
R EFERENCES Orthogonal Multiple Access,” in 2017 Wireless Days (WD), March 2017,
pp. 188–193.
[1] Q. C. Li, H. Niu, A. T. Papathanassiou, and G. Wu, “5G Network
Capacity: Key Elements and Technologies,” IEEE Veh. Technol. Mag.,
vol. 9, no. 1, pp. 71–78, March 2014.

Authorized licensed use limited to: Higher College of Technology. Downloaded on March 23,2021 at 07:06:43 UTC from IEEE Xplore. Restrictions apply.

You might also like