Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

User Capacity in Downlink MISO-NOMA Systems

Shahira H. Amin1 , Ahmed H. Mehana1 , Samy S. Soliman1,2 , Yasmine A. Fahmy1


1 Electronics and Electrical Communications Engineering Department, Cairo University, Egypt
2 Zewail City - University of Science and Technology, Egypt

shahirahany16@eng1.cu.edu.eg, ahesham.mehana@cu.edu.eg, ssoliman@zewailcity.edu.eg, yfahmy@cws-cufe.org

Abstract—Non-orthogonal multiple access (NOMA) emerges allocation algorithms to maximize the ergodic sum capacity,
as a strong candidate for the next wireless generation as it under a total power constraint and only a weak user’s rate
promises large spectral efficiency. However, as the number of constraint. The authors in [10] propose an optimal precoding
the users increases, the interference increases. This limits the
rate per user. In this paper, the maximum number of users scheme, based on the duality between the broadcast channel
that can be admitted in a NOMA system, under a total power and the multiple access channel, in addition to a suboptimal
constraint and a minimum rate requirement per user, is sought. precoding scheme, based on singular value decomposition,
A lower bound for the achievable sum-rate is first obtained. A to maximize the instantaneous sum capacity, subject to a
power allocation scheme that maximizes the sum-rate, under total power constraint as well as a minimum rate constraint
the aforementioned constraints, is then proposed. Using the
proposed scheme, an efficient algorithm to obtain the maximum for the weak user. In [11], a power allocation scheme that
number of users is developed. The performance is compared maximizes the instantaneous sum capacity, subject to a total
with the orthogonal multiple access (OMA) schemes. Our results power requirement and minimum user-rate requirements, was
show that the proposed scheme efficiently computes the number obtained, for M -users in a SISO-NOMA system. The authors
of users and provides significant gain compared to the OMA in [12] solve a similar problem to that in [11] for an orthog-
schemes.
onal frequency division multiplexing (OFDM) SISO-NOMA
I. I NTRODUCTION system. In [13], an optimal power allocation strategy is pro-
Traditionally, the orthogonal multiple access (OMA) vided to maximize the instantaneous weighted sum capacity,
schemes have been adopted in many wireless applications to subject to a total power requirement and minimum user-rate
serve each user in one or more orthogonal resource blocks. requirements, for N -users in a SISO-NOMA system. In [14],
Resource blocks could be time slots as in time division mul- power is allocated to the users to maximize the instantaneous
tiple access (TDMA), sub-carriers as in orthogonal frequency sum capacity, under a total power constraint and minimum
division multiple access (OFDMA), or spreading codes as in rate constraints, for G users in a beamforming (BF) MISO-
code division multiple access (CDMA). Unlike OMA, the NOMA system. The authors in [15] group the users into
principle idea in non-orthogonal multiple access (NOMA) clusters, and unlike [7]- [14], study MIMO-NOMA systems,
is to serve more than one user within the same resource and provide a power allocation strategy that maximizes the
block [1]. number of users per cluster, subject to a total power constraint
At the transmitter, multiple users’ signals are superposed. and minimum individual instantaneous signal-to-interference-
These users are differentiated in the power-domain, such that plus-noise ratio (SINR) constraints.
the strong user with the best channel conditions is assigned the In literature, it is clear that the problem of maximization
least amount of transmitting power, while the weak user with of the number of users has not been fully investigated.
the worst channel conditions is assigned the largest amount Additionally, the ergodic capacity has not been frequently
of power. Hence, the name power-domain NOMA. At the addressed. Unlike [14], where the authors solve the problem of
receiving terminal, successive interference cancellation (SIC) maximization of the instantaneous sum capacity in a downlink
is performed to remove the interference from the users with MISO-NOMA system, and [15], where the authors formulate
the worse channel conditions [1]. Compared to OMA, NOMA the problem of maximization of the number of users per
improves the spectral efficiency [1]–[6]. cluster, under individual instantaneous SINR requirements
Several researchers have studied the downlink power- in a downlink MIMO-NOMA system, in this paper, the
domain NOMA systems. In [7], the authors obtain the op- maximization of the number of users in a downlink MISO-
timum power allocation that maximizes the instantaneous NOMA system subject to a total power constraint and ergodic
sum capacity, in a single-input single-output (SISO)-NOMA user-rate constraints is considered. The paper contributes: 1)
system with two users, under a total power requirement a closed form equation for the achievable ergodic rate, 2) a
and minimum rate requirements for both users. A similar power allocation scheme to guarantee that the rate constraints
optimization problem to that in [7] is formulated in [8] for or- are achieved for all users, and 3) an algorithm that obtains
thogonal frequency division multiple access (OFDMA) SISO- the maximum number of users such that the rate constraints
NOMA systems. In [9], the authors consider multiple-input are met with equality.
multiple-output (MIMO)-NOMA systems with two users, and The rest of this paper is organized as follows: the system
provide optimal as well as low complexity suboptimal power model is illustrated in Section II. A lower bound for the
This work was funded by telecommunications regulatory agency of Egypt
(NTRA-Egypt).

978-1-5386-4727-1/18/$31.00 ©2018 IEEE


Authorized licensed use limited to: Universita degli Studi di Roma La Sapienza. Downloaded on May 31,2022 at 20:48:55 UTC from IEEE Xplore. Restrictions apply.
ergodic capacity is then obtained in Section III. In Section IV, The received signal at the g th user, assuming a flat fading
the optimization problem under study is formulated, the power channel is given by
allocated to each user is derived in the case of NOMA and
OMA, and an algorithm to find the maximum possible number yg = hH
g x + ng
of users is proposed. The simulation results are presented and G
hg X hl
discussed in Section V. Finally, the paper is concluded in = hH
g
H
αg sg + hg αl sl +ng , (5)
Section VI. khg k khl k
| {z } | l=1,l6={z
g
}
desired signal
interference signal
II. S YSTEM M ODEL
where the term ng represents the complex additive white
A downlink MISO-NOMA system, with G users and a Gaussian noise (AWGN) of the g th user, with zero mean and
single base station (BS) is considered. The number of antennas variance N0 , i.e., ng ∼ CN (0, N0 ). The second term in (5)
at the BS is M , and each user has one antenna. According to represents the interference on the g th user resulting from the
the NOMA principle, the G users are served within the same other users.
orthogonal resource block, but are distinguished in the power Following the NOMA principle, the g th user performs SIC
domain. The transmitted signal at the BS is given by to remove the interference from the following weaker users,
i.e., users g + 1, g + 2, . . . , G. For each user to successfully
G
X perform SIC, the achievable data rate for the g th user as
x= pg αg sg , (1) obtained in [14] is
g=1
Rg = min(Rg,g , Rg−1,g , ..., R2,g , R1,g ), (6)
where sg is the transmitted symbol of the g th user and is
assumed
 to have unit energy, i.e., E |sg |2 = 1, where where Rm,g is the achievable data rate for the g th user at the
E . is the expectation operator; αg2 is the power allocated receiver of the mth user, and could be written as in [14] as
to the g th user, and pg is the M × 1 precoding vector of the follows
g th user. Assuming perfect channel state information at the 2
transmitter (CSIT), and that the BS performs matched filtering |hH
m hg |
αg2
( !)
khg k2
(MF) precoding, the precoding vector of the g th user is given Rm,g = E log2 1+ P
|hH 2 ,
g−1 m hl |
by [14] l=1 αl2 khl k2 + N0
hg m = 1, 2, . . . , g. (7)
pg = , g = 1, 2, ..., G, (2)
khg k
Lemma 1: Assuming ∆  1, the achievable data rate for
where hg is the M × 1 complex Gaussian channel vector the g th user could be expressed as
between the BS and the g th user. The elements of hg are
independent identically distributed (i.i.d.) complex normal Rg = Rg,g . (8)
random variables with zero mean and variance, σg2 , i.e.,
{m} {m} Proof: Refer to Appendix A
hg ∼ CN (0, σg2 ), where hg is the mth element of
hg , m = 1, 2, ....M . For a total transmitting power of Pmax , In this paper, it is assumed that ∆  1. Hence, (8) is valid.

G
X III. A L OWER B OUND FOR THE E RGODIC C APACITY
αg2 ≤ Pmax . (3)
g=1 In this section, a lower bound for the
 ergodic
 capacity
 is
obtained. Using Jensen’s inequality, E φ x ≥ φ E x
At the transmitter, the G users are sorted from the strongest  
user to the weakest user according to their mean channel gains when φ is a convex function. Since log2 1 + x1 is a convex
such that σ12 > σ22 > · · · > σG 2
. The channel gain gap is function, the ergodic capacity of the g th user can be lower
defined as in [14] as bounded as
2
σg2 |hH
g hg |
αg2
( !)
∆g = 2 , g = 1, 2, ..., G − 1, (4) khg k2
σg+1 CgN OM A =E log2 1+ 2
Pg−1 |hH
g hl |
l=1 αl2 khl k2 + N0
where ∆g > 1. In this paper, it is assumed that the number
!
1
of users within the coverage of the BS is large enough. ≥ log2 1+ 2 ) . (9)
g−1 2 | g l |
hH h
(P
Therefore, it can be guaranteed with high probability that there l=1α khl k2
l +N 0

exists G users such that ∆g = ∆. A similar assumption was E 2


|
hH
g hg
α2g kh k2
|
considered in [14]. g

Authorized licensed use limited to: Universita degli Studi di Roma La Sapienza. Downloaded on May 31,2022 at 20:48:55 UTC from IEEE Xplore. Restrictions apply.
G
Lemma 2: Using the laws of probability, the ergodic capac- X
αg2 ≤ Pmax (15b)
ity of the g th user could be further lower bounded as
g=1
CgN OM A ≥ C g ≥ Cmin , (15c)
!
1
log2 1+ ( ) ( ) where Ig is the indicator function (Ig = 1 if C g ≥ Cmin
2
2| hH
g hl | and zero otherwise), Cmin is the minimum rate requirement
Pg−1 1
E l=1 αl khl k2 + N0 E 2
α2g
|hH
g hg | assuming all users have the same minimum rate requirement,
khg k2
and C g = C N g
OM A
, in the case of NOMA, while C g =
= CN
g
OM A
, (10) CgOM A
, in the case of OMA. The constraint C g ≥ Cmin
guarantees a minimum rate per user.
where C Ng
OM A
is the lower bound for the ergodic capacity
th It worth pointing out that the constraints in (15a), (15b),
of the g user.
and (15c) are linear functions of the power allocated to each
Proof: Refer to Appendix B
user. However, the objective function in (15) could be viewed
Lemma 3: Using (10), the lower bound of the ergodic
as the summation of zero/one indicator functions, which are
capacity of the g th user could be written in a closed-form
neither differentiable nor convex. The problem in (15) is,
as follows
! therefore, a non-convex problem. Similar to the approach
N OM A 1 in [16], where the authors obtain an algorithm from the
Cg = log2 1 + ! .
solution of a non-convex optimization problem, in the fol-
!
Pg−1 2 N0 1
l=1 αl + σ 2 α2 (M −1) g g
lowing, a user admission algorithm that iteratively calculates
the maximum number of admissible users is provided. The
(11)
proposed algorithm obtains a locally optimal solution.
Proof: Refer to Appendix C
Solving to satisfy the user-rate constraints in (15c), in the
In the case of OMA, the resources are divided among the
case of NOMA and OMA, using (11), the power allocated to
G users. The ergodic capacity for the g th user is given in [15]
each user in the case of NOMA is given by
as 2
|hHg hg |
! g−1 !
αg2 kh
( !)
1 k 2 2 2Cmin − 1 X
2 N0
CgOM A
=E log2 1 + g
. (12) αg,N OM A ≥ αl + 2 . (16)
G N0 M −1 σg
G l=1

Using Jensen’s inequality, a lower bound for the ergodic Using (17), the power allocated to the g th user in the case of
capacity is obtained as in the NOMA case. The lower bound OMA is given by
for the ergodic capacity of the g th user is ! !
2 2GCmin − 1 N0
g hg |
|hH
2
αg,OM A ≥ . (17)
αg2 M −1 Gσg2
( !)
1 khg k2
CgOM A = E log2 1+ N0

G G Similar to [15], where the authors propose an algorithm
to calculate the maximum number of users per cluster, the
!
1 1
log2 1+ ( ) = C OM
g
A
. (13) maximum number of users in this paper is calculated using
G
N0
E 1
2
Algorithm 1 below. The maximum number of users is ob-
G |hH
g hg | tained by allocating the minimum amount of power to each
α2g khg k2
user such that the user-rate constraints in Problem (15) are
Using (24) in Lemma 3, (13) could be simplified as met with equality.
! Starting at G = 1, the minimum amount of power allocated
OM A 1 1
Cg = log2 1 + N0
. (14) to the first user is calculated using (16), in the case of NOMA
G Gα2 (M −1)σ 2 or (17), in the case of OMA. The assigned power to the first
g g
user is checked to be less than Pmax . If true, a second user
IV. T HE P OWER A LLOCATION S CHEME
is admitted and assigned the minimum amount of power. The
In this section, the power allocated to each user is derived total assigned power is then checked to be less than Pmax .
and a user admission algorithm that calculates the maximum If true, a third user is admitted and allocated the minimum
number of users is proposed to meet the user-rate constraints amount of power. Again, the total allocated power is checked
with equality. The optimization problem is formulated as to be less than Pmax , and the process is repeated iteratively
follows X until the total assigned power becomes greater than or equal
arg max G = Ig to Pmax . At this point, the algorithm terminates.
α
g In each iteration of Algorithm 1, (16) in the case of
α = [α12 , α22 , ..., αG
2 T
] NOMA, is evaluated once, and the computational complexity
of the algorithm is O(G). The proposed algorithm is more
subject to: αg2 ≥ 0 (15a) efficient than the exhaustive search where the computational

Authorized licensed use limited to: Universita degli Studi di Roma La Sapienza. Downloaded on May 31,2022 at 20:48:55 UTC from IEEE Xplore. Restrictions apply.
complexity is O(G2ig ), where Gig is the initial guess of the
maximum number of users that can be served. It is noteworthy
that both the proposed algorithm and exhaustive search would
result in the same maximum number of admissible users.

Algorithm 1 Maximum Number of Users


Initialize: G = 1
PGαG using (16) or (17)
Calculate
while g=1 αg2 < Pmax do
G=G+1
Calculate αG using (16) or (17)
end while
G=G−1

Fig. 1. The actual sum capacity (Csum ) and the lower bound of the sum
capacity (C sum ) when ∆ = 10 (dB), Cmin = 1.5 (bps/Hz), and M = 2
V. N UMERICAL R ESULTS AND D ISCUSSION
In this section, the numerical results are presented and
discussed. The noise variance, N0 is set to 0 (dB). The total
transmitting power Pmax , is varied within the range 0-30 (dB).

TABLE I
VALUES OF ∆ WHICH MEET Rg ≤ Rg,g FOR CERTAIN M AND Cmin

PP Cmin
PP 1.5 (bps/Hz) 2 (bps/Hz) 2.5 (bps/Hz)
M P
P
2 ≥ 10 (dB) ≥ 11.5 (dB) ≥ 12.5 (dB)
4 ≥ 12.5 (dB) ≥ 14 (dB) ≥ 16 (dB)

Fig. 1 shows the actual sum capacity (Csum ) and the lower
bound of the sum capacity (C sum ) versus the average signal-
to-noise (SNR). The actual and loer bound of the sum capacity
are calculated as follows Fig. 2. The maximum number of users (G) when Cmin = 1.5 (bps/Hz),
M = 2, and for different values of ∆
G
X G
X
Csum = CgN OM A , C sum = CN
g
OM A
, (18)
g=1 g=1 In Fig. 2, the maximum number of users (G) is plotted
for different values of the channel gain gap (∆ = 10, 12, 13
respectively, where CgN OM A is calculated using the left hand
(dB)) when Cmin = 1.5 (bps/Hz) and M = 2. For certain
side of (9), and C N g
OM A
is calculated using the right hand side
average SNR, as ∆ decreases, G increases. The effect of ∆
of (9). C sum is linear in the average SNR because C sum = 2 2
PG N OM A on αg,N OM A is implied in (16) through σg as follows
g=1 g C = GC min as algorithm 1 allocates each user ! g−1 !
the minimum power, i.e., solves (11) with equality, which 2Cmin − 1 X ∆g−1 N0
2 2
guarantees that C g equals to Cmin . Similarly, Csum is also αg,N OM A ≥ αl + . (19)
M −1 σ12
linear in the average SNR. l=1

It is clear in Fig. 1 that Csum is closer to C sum at the low 2


From (19), it can be seen that as ∆ decreases, αg,N OM A
average SNR values than at the high average SNR values. This decreases, and more users could be allowed in the system
is because at low average SNR, G is small. Therefore, the for the same transmitting power. This is expected as when ∆
difference between Csum and C sum is small. As the average decreases, the users become closer to the BS, and therefore,
SNR increases, G gets bigger, and the difference between require less power to be admitted to the system.
Csum and C sum gets larger. Fig. 3 shows the maximum number of users (G) for differ-
The following set of figures (Figures 1- 4) are to show the ent values of the number of transmitting antennas (M = 2, 4)
maximum number of admitted users versus the average SNR when ∆ = 13 (dB) and Cmin = 1.5 (bps/Hz). As M
varying first ∆, second, M , and third, Cmin . In Figures 2- increases, G increases. This is clear form (16) as when M
4, it can be observed that G increases in discrete steps with 2
increases, αg,N OM A decreases, and consequently, more users
increasing the average SNR. The increase in G occurs every could be admitted in the system for the same available power
∆ increase in the average SNR because the power allocated at the BS.
to user g − 1 is approximately ∆ times greater than the power In Fig. 4, the maximum number of users (G) is plotted for
allocated to user g. This is clear from (19). different values of the minimum rate requirements (Cmin =

Authorized licensed use limited to: Universita degli Studi di Roma La Sapienza. Downloaded on May 31,2022 at 20:48:55 UTC from IEEE Xplore. Restrictions apply.
Fig. 3. The maximum number of users (G) when ∆ = 13 (dB), Cmin = 1.5 Fig. 5. The maximum number of users (G) in the case of NOMA and OMA
(bps/Hz), and for different values of M when Cmin = 1.5 (bps/Hz), ∆ = 10 (dB), and M = 2

VI. C ONCLUSION
A closed-form lower bound for the ergodic capacity in the
downlink MISO-NOMA system is obtained. The optimum
power allocated to each user is calculated such that the user-
rate constraints are satisfied. An algorithm that obtains the
maximum number of users that can be served in the MISO-
NOMA system such that the user-rate constraints are met with
equality is proposed. The computational complexity of this
algorithm is linear with the maximum number of users ad-
mitted. The numerical results show that the maximum number
of users increases with increasing the number of transmitting
antennas, whereas the maximum number of users decreases
with increasing the channel gain gap or the minimum rate
requirements. Also, the maximum number of admitted users
Fig. 4. The maximum number of users (G) when ∆ = 15 (dB), M = 2, in the case of NOMA is greater than in the case of OMA.
and for different values of Cmin
A PPENDIX A
P ROOF OF L EMMA 1
Using (4), (7) could be re-written as
1.5, 2, 2.5 (bps/Hz)) when ∆ = 10 (dB) and M = 2. As |hH
2
m hg |
αg2
( !)
Cmin decreases, G increases. This is clear from (16) as when khg k2
2
Cmin decreases, αg,N Rm,g = E log2 1+ P ,
OM A decreases. Hence, greater number
2
g−1 |hH
m hl | ∆m−1 N0
l=1 αl2 khl k2 + σ12
of users could be served in the system given a certain amount
(20)
of transmitting power.
where the elements of hm are i.i.d. complex standard
Finally, Fig. 5 compares the maximum number of users (G) normal distributed. From (20), it can be observed that
in the case of NOMA and OMA. It is clear that the NOMA R1,g , R2,g , ..., Rg−1,g have a norm term in the numerator and
allows a greater maximum number of users than the OMA. a Frobenius norm term in the denominator; this is due to the
This is because the total bandwidth in OMA is divided among MF precoding, where |hH m hg | is the Frobenius norm when
the users, unlike NOMA. Hence, in OMA, greater amount of m = g, and |hH m hg | is the norm, ∀m 6= g. The Frobenius
power is required to serve an additional user than in NOMA. norm terms in the denominator of R1,g , R2,g , ..., Rg−1,g are
It is clear from (16) and (17) that although the ratio between multiplied by α12 , α22 , ..., αg−1
2 2
, respectively, where αg−1 >
2 2
the amounts of assigned power to any two consecutive users ... > α2 > α1 , due to the NOMA power allocation. Also,
is approximately ∆, in the case of NOMA, yet, in the case ∆g−2 > · · · > ∆2−1 > ∆1−1 . Therefore, Rg−1,g < ... <
of OMA, the ratio between the assigned powers to any two R2,g < R1,g .
consecutive users is greater than ∆. This could be observed Unlike the rates (R1,g , R2,g , ..., Rg−1,g ), Rg,g has a Frobe-
in Fig. 5. Rationally, NOMA allows the users to share the nius norm term in the numerator and norm terms in the
same bandwidth, and consequently, achieves greater number denominator. By numerically solving the inequality Rg,g <
of users than OMA. Rg−1,g , it is shown in Table I that, for certain values of M

Authorized licensed use limited to: Universita degli Studi di Roma La Sapienza. Downloaded on May 31,2022 at 20:48:55 UTC from IEEE Xplore. Restrictions apply.
and Cmin , Rg,g ≤ Rg−1,g is met when ∆  1, where Cmin Using (23) and (24) in (10), the lower bound of the ergodic
is the minimum user-rate requirement. Hence, (8) is valid. capacity of the g th user could be simplified in a closed-form
A PPENDIX B as in (11).
P ROOF OF L EMMA 2 R EFERENCES
Using the laws of probability, [1] Z. Ding, X. Lei, G. K. Karagiannidis, R. Schober, J. Yuan, and
( g−1 2 ! !) V. K. Bhargava, “A survey on non-orthogonal multiple access for 5G
X hH g hl
1 networks: Research challenges and future trends,” IEEE Journal on
2
E αl + N0 2 Selected Areas in Communications, vol. 35, no. 10, pp. 2181–2195,
khl k2 |hHg hg |
l=1 αg2 kh g k 2 2017.
( g−1 2 ) ( ) [2] L. Dai, B. Wang, Y. Yuan, S. Han, I. Chih-Lin, and Z. Wang, “Non-
X hH h
g l
1 orthogonal multiple access for 5G: solutions, challenges, opportunities,
2
=E αl + N0 E 2 and future research trends,” IEEE Communications Magazine, vol. 53,
khl k2 |hH g hg | no. 9, pp. 74–81, 2015.
l=1 αg2 kh k 2
g [3] S. R. Islam, N. Avazov, O. A. Dobre, and K.-S. Kwak, “Power-domain
g−1 H 2 !
non-orthogonal multiple access (NOMA) in 5G systems: Potentials and
1 X
2
hg lh
challenges,” IEEE Communications Surveys & Tutorials, vol. 19, no. 2,
+ COV 2 , αl + N0 . (21)
g hg |
|hH khl k2 pp. 721–742, 2017.
αg2 kh gk
2 l=1 [4] A. Benjebbour, Y. Saito, Y. Kishiyama, A. Li, A. Harada, and T. Naka-
2 mura, “Concept and practical considerations of non-orthogonal multiple
Pg−1 |hH
g hl | 1 access (NOMA) for future radio access,” in Proc. of IEEE Intelligent
The terms l=1 αl2 khl k2 + N0 and 2 could be
α2
|hH
g hg | Signal Processing and Communications Systems (ISPACS), 2013 Inter-
g g kh k2 national Symposium on. IEEE, 2013, pp. 770–774.
Pg−1 kh k2 khl k2 cos2 (θg,l ) 1 [5] Y. Liu, G. Pan, H. Zhang, and M. Song, “On the capacity comparison
simplified as l=1 αl2 g khl k2 + N 0 and α2 khg k2 , between MIMO-NOMA and MIMO-OMA,” IEEE Access, vol. 4, pp.
g
respectively, where θg,l is the angle between the two vectors 2123–2129, 2016.
hg and hl . As khg k2 increases, the first random variable [6] Z. Wu, K. Lu, C. Jiang, and X. Shao, “Comprehensive study and
comparison on 5G NOMA schemes,” IEEE Access, vol. 6, pp. 18 511–
increases while the second random variable decreases. There- 18 519, 2018.
fore, the two random variables are negatively correlated, and [7] C.-L. Wang, J.-Y. Chen, and Y.-J. Chen, “Power allocation for a
hence, downlink non-orthogonal multiple access system,” IEEE Wireless Com-
( g−1 2 ! !) munications Letters, vol. 5, no. 5, pp. 532–535, 2016.
X hH g h l
1 [8] Z. Q. Al-Abbasi and D. K. So, “Power allocation for sum rate
E αl2 + N0 2
maximization in non-orthogonal multiple access system,” in Proc. of
khl k2 g hg |
|hH IEEE Personal, Indoor, and Mobile Radio Communications (PIMRC),
l=1 αg2 kh gk
2
2015 IEEE 26th Annual International Symposium on. IEEE, 2015, pp.
( g−1 2 ) ( ) 1649–1653.
X hH g h l
1 [9] Q. Sun, S. Han, I. Chin-Lin, and Z. Pan, “On the ergodic capacity
≤E αl2 + N0 E 2 . (22)
khl k2 |hHg hg |
of MIMO NOMA systems,” IEEE Wireless Communications Letters,
l=1 αg2 kh gk
2 vol. 4, no. 4, pp. 405–408, 2015.
[10] Q. Sun, S. Han, Z. Xu, S. Wang, I. Chih-Lin, and Z. Pan, “Sum rate
Using (22), the ergodic capacity of the g th user could be optimization for MIMO non-orthogonal multiple access systems,” in
further lower bounded as in (10). Proc. of IEEE Wireless Communications and Networking Conference
(WCNC), 2015 IEEE. IEEE, 2015, pp. 747–752.
A PPENDIX C [11] Z. Yang, W. Xu, C. Pan, Y. Pan, and M. Chen, “On the optimality
P ROOF OF L EMMA 3 of power allocation for NOMA downlinks with individual QoS con-
hH
straints,” IEEE Communications Letters, 2017.
g hl
It was given in [17] that khl k is a complex Gaussian [12] W. Cai, C. Chen, L. Bai, Y. Jin, and J. Choi, “Subcarrier and power
hH
allocation scheme for downlink OFDM-NOMA systems,” IET Signal
g hl
random variable with zero mean and variance σg2 , i.e., khl k ∼ Processing, vol. 11, no. 1, pp. 51–58, 2016.
[13] J. Wang, Q. Peng, Y. Huang, H.-M. Wang, and X. You, “Convexity
CN (0, σg2 ). Hence, of weighted sum rate maximization in NOMA systems,” IEEE signal
( ) processing letters, vol. 24, no. 9, pp. 1323–1327, 2017.
|hH
g hl |
2
[14] C. Chen, W. Cai, X. Cheng, L. Yang, and Y. Jin, “Low complexity
E = σg2 . (23)
khl k2 beamforming and user selection schemes for 5G MIMO-NOMA sys-
tems,” IEEE Journal on Selected Areas in Communications, vol. 35,
|hH hg |2 no. 12, pp. 2708–2722, 2017.
Additionally, khg
gk
2 = σg2 hH
g hg , where in the left hand side, [15] M. Zeng, A. Yadav, O. A. Dobre, G. I. Tsiropoulos, and H. V. Poor,
the elements of hg are complex normal distributed with zero “Capacity comparison between MIMO-NOMA and MIMO-OMA with
multiple users in a cluster,” arXiv preprint arXiv:1706.02731, 2017.
mean and variance, σg2 , and in the right hand side, the elements [16] Y. Sun, D. W. K. Ng, Z. Ding, and R. Schober, “Optimal joint power and
of hg are complex standard normal distributed and hH g hg is subcarrier allocation for MC-NOMA systems,” in Proc. of IEEE Global
a 1 × 1 Wishart matrix with M (M > 1) degrees of freedom, Communications Conference (GLOBECOM), 2016 IEEE. IEEE, 2016,
pp. 1–6.
i.e., hH
g hg ∼ W1 (M, IM ). Hence, [17] [17] H. Q. Ngo, E. G. Larsson, and T. L. Marzetta, “Energy and spectral
efficiency of very large multiuser MIMO systems,” IEEE Transactions
( )
1 1 on Communications, vol. 61, no. 4, pp. 1436–1449, 2013.
E Hh 2
= 2 , M ≥ 2. (24)
2 | hg g | α g (M − 1)σg2
αg khg k2

Authorized licensed use limited to: Universita degli Studi di Roma La Sapienza. Downloaded on May 31,2022 at 20:48:55 UTC from IEEE Xplore. Restrictions apply.

You might also like