You are on page 1of 29

Accepted Manuscript

Optimal joint subcarrier and power allocation for MISO-NOMA satellite


networks

Mostafa Alhusseini, Paeiz Azmi, Nader Mokari

PII: S1874-4907(18)30119-8
DOI: https://doi.org/10.1016/j.phycom.2018.11.001
Reference: PHYCOM 613

To appear in: Physical Communication

Received date : 6 March 2018


Revised date : 17 September 2018
Accepted date : 9 November 2018

Please cite this article as: M. Alhusseini, P. Azmi and N. Mokari, Optimal joint subcarrier and
power allocation for MISO-NOMA satellite networks, Physical Communication (2018),
https://doi.org/10.1016/j.phycom.2018.11.001

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form.
Please note that during the production process errors may be discovered which could affect the
content, and all legal disclaimers that apply to the journal pertain.
Phy Comm
Physical Communication 00 (2018) 1–28

Optimal Joint Subcarrier and Power Allocation for MISO-NOMA


Satellite Networks
Mostafa Alhusseini a , Paeiz Azmi a,1 , and Nader Mokari a

a: Electrical and Computer Engineering Department, Tarbiat Modares University, Tehran Iran.

Abstract
In this paper, we propose a novel MISO multi-spot-beam satellite system design based on jointly optimizing subcarrier and power
allocation for the downlink of a power domain non-orthogonal multiple access (PD-NOMA) network. The proposed non-convex
NP-hard optimization problem minimizes the total transmit power taking into account subcarrier and transmit power constraints.
An alternating method is used to solve this problem. For a fixed power allocation, we update the integer nonlinear programming
(INLP) subcarrier allocation by using NOMAD toolbox. For a given subcarrier assignment, in order to update the transmit power
values, we use the successive convex approximation approach (SCA). In addition, the proposed problem is solved by using the
optimal method called “Monotonic Optimization” in order to find the optimality gap of the alternating method. Simulation results
show that the total transmit power by applying the alternating method is close to the total transmit power by applying the monotonic
approach. Moreover, we observe that the MISO and PD-NOMA techniques reduce the total transmit power of the MISO multi-
spot-beam satellite system by approximately 20% with respect to the SISO and OMA techniques and the effectiveness of the
proposed scenario and methods is considered in clear and rainy sky situations.

Keywords: Satellite system, Multi-spot-beam, MISO system, Subcarrier allocation, Power allocation.

1. Introduction

1.1. State of the art

Satellite communication is an important part of the communication systems around the world since they present

services, which are limited or not found in terrestrial networks [1]. The modern satellite communication systems use

geosynchronous earth-orbit (GEO) satellite communication networks including multiple narrow spot-beams in their

coverage area. In these systems, large number of satellite antennas and narrow spot-beams can provide higher rate

services to users due to the large satellite antenna gain [1]-[3]. High throughput satellite (HTS) systems are the newest

satellite communication generation and the most used around the world. HTS system networks divide the service area

1 Corresponding author.

E-mail address: Pazmi@modares.ac.ir (P. Azmi).


1
Mostafa Alhusseini, Paeiz Azmi, Nader Mokari / Physical Communication 00 (2018) 1–28 2

Table 1: List all abbreviation

MISO Multi Input Single Output DVB-SH DVB - Satellite services to Handhelds
QoS Quality of Service MIMO-OFDM Orthogonal Frequency Division Multiple-OFDM
INLP Integer Non Linear Program SCA Successive Convex Approximation
AF Amplify-and-Forward NOMA Non-Orthogonal Multiple Access
AFR Array-Fed Reflector GFDM-NOMA Generalized Frequency Division Multiplexing-NOMA
SISO Single Input Single Output PD-NOMA Power Domain- NOMA
OMA Orthogonal Multiple Access AWGN Additive White Gaussian Noise
GEO Geosynchronous Earth-Orbit ITU-R International Telecommunication Union - Radiocomm
HTS High Throughput Satellite SINR Signal-to-interference-plus-noise Ratio
MIMO Multi Input Multi Output SIC Successive Interference Cancellation
SE Spectral Efficiency DC Difference of two Convex function
PSO Particle Swarm Optimization SCALE Successive Convex Approximation for Low Complexity
HetNet Heterogonous Network DVB-NGH DVB-Next Generation broadcasting system to Handheld
AFR Array-Fed Reflector

into multi-spot-beam service areas in order to use high order of frequency reuse over the service area and employ

the satellite antenna gain to obtain more capacity for given satellite resources [4], [5]. With the rapid development of

satellite communication systems, a new multiple access technique for next generation communications has attracted

the attention of researchers. Power domain non-orthogonal multiple access (PD-NOMA) is a new multiple access

scheme proposed for the next generation of HTS systems [6]-[8]. By applying the PD-NOMA technique, the same

subcarrier can be shared by multiple users [9]. Furthermore, future mobile satellite systems can take advantage of

the MIMO techniques. MIMO satellites are set to provide spatial diversity gain over shadowing by installing them

in highly separated orbital positions [10], [11]. In this context, for modern satellite systems, an efficient and adaptive

resource allocation is essential for a high capacity and spectral efficiency (SE) considering the traffic demands and

channel conditions. In this sense, it is necessary to develop efficient adaptive resource allocation schemes for allocating

resources adaptively among the beams of a multi-beam satellite system by considering the multi-beam interference

varying dynamically, as well the different channel conditions of each user and the traffic demands [1], [12]-[13].

1.2. Related works

1.2.1. Power allocation

In the related literature [1], [14]-[16], power allocation algorithms for multi-beam satellite communication sys-

tem are suggested. In [1], a heuristic particle swarm optimization (PSO) problem is applied. In [14], a non-linear

optimization with constraints that describes the impact of the inter-beam satellite system interference is proposed.

2
Mostafa Alhusseini, Paeiz Azmi, Nader Mokari / Physical Communication 00 (2018) 1–28 3

In [15], optimum power allocation for the multi-beam broadband multimedia satellite systems based on the different

traffic demand of each beam is described. In [16], the power minimization problem in NOMA-based HetNet is dis-

cussed. In [11], [17]-[19], recent investigations are reported on more practical ways to exploit MIMO technology for

enhancing mobile satellite broadcasting power and/or spectral efficiency. In [11], four potential satellite and hybrid

MIMO scenarios for DVB-SH are investigated and analyzed. In [17], the performance, capabilities, and design of a

comprehensive laboratory testbed for MIMO hybrid mobile digital broadcasting is considered. In [18], the relation-

ship between the capacity performance of MIMO satellite communication and using MIMO technology in satellite

communication system is discussed. The QoS-guaranteed energy-efficient power allocation in downlink multi-user

MIMO-OFDM systems is investigated in [19]. Several sides of the PD-NOMA scheme have been recently presented

in different works [20], [21]. In [20], the power allocation problem in a PD-NOMA wireless network with a one-way

OFDM AF relay in multibeam satellite systems is applied. In [21], the problem of non orthogonal multiple access

(NOMA) in multibeam satellite systems is discussed.

1.2.2. Bandwidth Allocation

In [22], a dynamic bandwidth allocation scheme based on the traffic distributions and subcarrier conditions for a

parallel multi-beam satellite system is suggested. In [23], the optimum power allocation algorithm for the multi-beam

broadband multimedia satellite systems is proposed.

1.2.3. Joint Power and Bandwidth Allocation

In [24], two joint bandwidth and power allocation schemes to optimize power and subcarrier allocation for beam-

hopping user downlinks multibeam satellite system are assumed. In [25], a downlink power allocation for heteroge-

neous traffic in GFDM-NOMA-based heterogeneous cellular networks is investigated.

MIMO multi-spot-beam satellite system is a novel architecture for the next generation satellite systems [4]. Power

domain - non-orthogonal multiple access (PD-NOMA) is a novel multiple access scheme which is a promising can-

didate for the next cellular communication systems [8]. In this paper we combine these two techniques which has not

investigated so far jointly, next we propose a novel MISO multi-spot-beam satellite system design based on jointly

optimizing subcarrier and power allocation for the downlink of a PD-NOMA network. Our aim in this work is to

minimize total transmit power subject to subcarrier and transmit power constraints; therefore, we formulate an opti-

mization problem to achieve this aim. The formulated optimization problem is a mixed of integer non linear problem

and high complex, it cannot be solved directly by using the conventional methods. To tackle this issue, there are many

strategies to solve this problem listed in Table 2. In order to solve the formulated optimization problem, we adopt the

alternating algorithm, i.e., we convert the optimization problems to two subproblems power allocation and subcarrier
3
Mostafa Alhusseini, Paeiz Azmi, Nader Mokari / Physical Communication 00 (2018) 1–28 4

Table 2: Methods

Method Reference explanation


Mesh Adaptive Direct [R3] To solve non linear problem with binary optimization
Search (MADS) algorithm variables or integer optimization variables or continues
optimization variables.
SCA [R4] To convert non convex optimization problem to convex
one, by employing mathematical approximations.
Alternating method [R5] To convert the optimization problem to several subprob-
lems and solve them separately and alternatively.
Matching theory [R6] Matching theory is a powerful tool to study the forma-
tion of dynamic and mutually beneficial relations among
different types of rational and selfish agents.
Monotonic method [R7] To achieve optimal solution by employing high complex
algorithm which proposed in [7].
Metaheuristics method [R8] Search methods are used to enhance the efficiency of the
multi-objective problem.

allocation and solve them alternatively. In each iteration, for a given subcarrier allocation, we allocate power, then,

we allocate subcarrier for a given power allocation. In each iteration of this method, power allocation problem is

non-convex and subcarrier allocation is NLP. We solve the power allocation by the SCA approach. To this end, we

use the SCALE approximation. Also, we solve the subcarrier allocation problem, by using the Mesh Adaptive Direct

Search (MADS) algorithm. Moreover, we obtain optimal solution and the optimality gap of the proposed solution

method by converting the optimization problems to a canonical form of the monotonic optimization.

1.3. Contributions

The main contributions of this paper are summarized as follows:

(1) A mathematical formulation of the joint subcarrier and power allocation for downlink MISO multi-spot-beam

satellite system based on PD-NOMA is proposed.

(2) To solve the considered problem, we divide the optimization problem into two subproblems: subcarrier allo-

cation subproblem, and power allocation subproblem. Then we apply an alternating method in order to find

the subcarrier allocation and power allocation values. The subcarrier allocation problem is an integer nonlinear

programming (INLP), is solved by using the NOMAD toolbox, and the power allocation subproblem is solved

by applying the successive convex approximation (SCA) method. Moreover, we prove the convergence of the

proposed alternating method.

(3) To find the optimality gap of the low complexity alternating method, we solve the proposed problem by applying

the optimal method, namely, ”Monotonic Optimization”, which has high computation complexity.
4
Mostafa Alhusseini, Paeiz Azmi, Nader Mokari / Physical Communication 00 (2018) 1–28 5

1.4. Paper organization

The rest of the paper is organized in the following manner: In Section 2, the system model and problem formulation

for the multi-spot-beam satellite system scenario are proposed. In Section 3, the proposed solution and algorithm

to solve the problem are investigated. The proposed optimal solution is presented in Section 4. The computational

complexity is investigated in Section 5. Numerical results are analyzed in Section 6. Finally, the important conclusions

are given in Section 7.

2. The multi-spot-beam satellite system architecture and the resource allocation problem

2.1. Multi-spot-beam satellite system scenario

We consider a downlink MISO multi-spot-beam satellite system. The schematic diagram of the system model

is shown in Fig. ??. The considered HTS system includes Na multiple array antennas serving the coverage area.

Each antenna serves Ka users and consists of Nb spot-beams, where the coverage zone is served via (Na × Nb ) spot-

beams shaped by on-board array-fed reflector (AFR) antenna. Each small cell is covered by one spot-beam. The

geostationary satellite employs a transparent architecture designed to operate in the Ka-band. Multiple gateways are

used to accommodate the high bandwidth required at the feeder link. We focus on downlink resource allocation and

we suppose that noise in Gaussian and the channel is modeled as Rayleigh fading. The NOMA technique is applied

on each subcarrier, where more than one user can be assigned on each subcarrier. Herein, the total bandwidth of the

system Btot , is equally divided into Nc subcarriers where the bandwidth of each subcarrier is Bsc = Btot /Nc .

The set of all antennas, users, beams, and subcarriers are denoted by Na , Ka , Nb , and Nc , respectively. The total

available system power is define by Ptot . If pka,b,c is the transmit power for user k on subcarrier c over beam b from

antenna a, the following inequalities hold,

X Ka X
Na X Nb X
Nc
ρka,b,c pka,b,c ≤ ptot , (1)
a=1 k=1 b=1 c=1

Ka X
X Nb X
Nc
ρka,b,c pka,b,c ≤ pmax
a , ∀a, (2)
k=1 b=1 c=1

where, pmax
a , a ∈ {1, 2, ..., Na } is the maximum transmit power of each antenna a. The subcarrier assignment is denoted

by a binary variable ρka,b,c , where ρka,b,c = 1 if user k is allocated on subcarrier c over beam b from antenna a, otherwise

ρka,b,c = 0.

Let hka,b,c be the channel gain that includes: (i) satellite beam antenna gain toward the purposed coverage area, (ii)

gain of the received antenna, (iii) free space loss and rain attenuation (according to ITU-R recommendation [12]. Let
5
Mostafa Alhusseini, Paeiz Azmi, Nader Mokari / Physical Communication 00 (2018) 1–28 6

Figure 1: System model for MISO multi-spot-beam satellite system.

hka,b,c be the channel gain for user k ∈ {1, 2, ..., Ka } on subcarrier c ∈ {1, 2, ..., Nc } over beam b ∈ {1, 2, ..., Nb } from
k
antenna a ∈ {1, 2, ..., Na }. The received SINR γa,b,c for user k on subcarrier c over beam b from antenna a is expressed

as:
2
k
|hka,b,c | pka,b,c
γa,b,c = k,CCI k,ACI k,NOMA
, (3)
N0 Bc + Ia,b,c + Ia,b,c + Ia,b,c

where, N0 is the noise power spectral density, which depends on the receiver antenna and equivalent noise temperature,

and the interference terms can be obtained as follows:

Nb
X
k,CCI
Ia,b,c = ρka,i,c |hka,i,c |2 pka,i,c , (4)
i=1,i,b

Na X
X Nb
k,ACI
Ia,b,c = ρkl,i,c |hkl,i,c |2 pkl,i,c , (5)
l=1,l,a i=1

K
X
k,NOMA
Ia,b,c = ρka,b,c |hka,b,c |2 pva,b,c , (6)
k=1,|hka,b,c |2 ≤|hva,b,c |2

6
Mostafa Alhusseini, Paeiz Azmi, Nader Mokari / Physical Communication 00 (2018) 1–28 7

k,CCI
here, Ia,b,c represents the intra antenna interference on subcarrier c of beam b caused by co-channel carriers in other
k,ACI k,NOMA
beams. Ia,b,c explains the inter antenna interference caused by the adjacent antennas. Ia,b,c illustrates the inter

user interference power received from the PD-NOMA technique. The coverage region of the antennas using the same

frequency is managed to be far enough so that inter antenna interference is ignored. Therefore, the received S INRka,b,c

for user k on subcarrier c over beam b from antenna a is reformulated as follows:

2
|hka,b,c | pka,b,c
S INRka,b,c = k,CCI k,NOMA
. (7)
N0 Bc + Ia,b,c + Ia,b,c

The instantaneous data rate attained by user k on subcarrier c over beam b from antenna a can then be written as:

k
ra,b,c = Bc log2 (1 + S INRka,b,c ). (8)

In order to obtain successful interference cancellation, the SINR of the user which has better channel gain on itself

must be greater than or equal the SINR of it on a user with worse channel gain, then:

ρm n m m
a,b,c ρa,b,c (S INRa,b,c (m) − S INRa,b,c (n)) ≤ 0, ∀a ∈ Na , b ∈ Nb , c ∈ Nc , m, n ∈ Ka ,
(9)
|hm 2 n 2
a,b,c | ≤ |ha,b,c | , m , n,

where, S INRm
a,b,c (n) is the SINR of user m on subcarrier c over beam b from antenna a at user n [26].

In our work, the users are static because the number of users in our system is fix, but the QoS of each user is

dynamic which is allocated due to the optimal solution of the problem.

2.2. Resource allocation problem

In this paper, we aim to propose an optimal joint subcarrier and power allocation algorithm. Static users are

considered for power and bandwidth allocation problem. The objective function is to minimize the total transmit

power, specifically, the objective function is represented as follows:

Na X
X Ka X
Nb X
Nc
f (ρ, p) = ρka,b,c pka,b,c . (10)
a=1 k=1 b=1 c=1

We examine the design of subcarrier and power allocation over satellite downlink in terms of Shannon capacity. For

this system model, we find an optimized design by minimizing a general function. We focus on the problem of

transmit power, and subcarrier allocation:

7
Mostafa Alhusseini, Paeiz Azmi, Nader Mokari / Physical Communication 00 (2018) 1–28 8

Na X
X Ka X
Nb X
Nc
min ρka,b,c pka,b,c , (11a)
P,ρ
a=1 k=1 b=1 c=1
Na X
X Ka X
Nb X
Nc
s.t. ρka,b,c pka,b,c ≤ ptot , (11b)
a=1 k=1 b=1 c=1
Ka X
X Nb X
Nc
ρka,b,c pka,b,c ≤ pmax
a , ∀a, (11c)
k=1 b=1 c=1
Nb X
X Nc
ρka,b,c ra,b,c
k
≥ Rmin , ∀a, k, (11d)
b=1 c=1
Ka
X
ρka,b,c ≤ Umax , ∀a, b, c, (11e)
k=1

ρm n m m
a,b,c ρa,b,c (S INRa,b,c (m) − S INRa,b,c (n)) ≤ 0, ∀a ∈ Na , b ∈ Nb , c ∈ Nc , m, n ∈ Ka ,

|hm 2 n 2
a,b,c | ≤ |ha,b,c | , m , n, (11f)

ρka,b,c ∈ {0, 1}, ∀a, b, c, k, (11g)

pka,b,c ≥ 0, ∀a, b, c, k, (11h)

where, constraints (11b) and (11c) are the power consumption of the total satellite system and the power consumption

for each antenna, respectively. (11d) denotes the traffic users rate constraint, (11e) and (11g) indicate the PD-NOMA

assumptions on each subcarrier, (11f) defines the SIC constraint.

Problem (11) is a non-linear program containing both integer and continuous variables. To solve (11), any direct

search method would involve an exhaustive search of all the possible subcarrier assignment ρ, followed by finding the

optimal power allocation p for each of these assignments. It is apparent that such an approach incurs an exponential

complexity with respect to the number of subcarriers, which can be large in practice. Moreover, the non-convexity

of the rate function implies that jointly optimizing all transmit powers is in itself a challenging problem, even for a

fixed subcarrier allocation. Therefore, problem (11) is a non-convex, intractable, and NP-hard problem. Hence, we

divide problem (11) into two subproblems: subcarrier allocation subproblem, and power allocation subproblem. Then

we apply an alternating algorithm to optimize alternatively the subcarrier and power allocation. Moreover, to find

the optimality gap of the proposed algorithm, the proposed problem is also solved using the optimal method called

”Monotonic Optimization”.

8
Mostafa Alhusseini, Paeiz Azmi, Nader Mokari / Physical Communication 00 (2018) 1–28 9

Table 3: Resource Allocation Optimization Algorithm

INITIALIZE: Let t = 0, find the initial values pt = p0 and


ρt = ρ0
REPEAT:
• Find a solution for (14) with ρ = ρt
then give it to pt+1 using the SCA approach.
• Find a solution for (13) with p = pt+1 for ρ
then give it to ρt+1 using NOMAD toolbox.
•t =t+1
UNTIL ||pt − pt−1 || < Ω1 or t = T
OUTPUT: pt and ρt are the output of the system

3. Joint subcarrier and power allocation algorithm evolution

3.1. Algorithm overview

For solving the problem (11), which is a constrained non-linear program incorporating both integer and continuous

variables, we utilize an alternating algorithm that optimizes the subcarrier and power allocation alternatively [27]-[28].

This algorithm is presented in Table 3, where t denotes the current iteration number t ≥ 0. First, we must compute the

initial power and the subcarrier allocation ρ0 and p0 . At the beginning of each iteration, we calculate p for (14) with

ρ = ρt . Next, we evaluate ρ for (13) with p = pt+1 . We take this subcarrier allocation as ρt+1 . Later, it is shown that

ρ can be found by using the nomad toolbox. The algorithm is stopped when t = T that is feasible for practice, or if

||pt − pt−1 || is smaller than a prescribed small value Ω1 . ||.|| represents the Euclidean norm vector. Finally, pt and ρt of

the last iteration is the solution to problem (11).

Proposition 1: The algorithm, presented in Table 3, after each iteration, improves the objective function f (ρ, p)

value and converges.


Proof 1. For the overall algorithm characterized in Table 3, after applying the first step, obtaining the power alloca-
tion of iteration t + 1 with the given subcarrier allocation ρ = ρt , we will have f (ρt , pt ) ≥ f (ρt , pt+1 ), which is proved
in the Propositions (2) for the SCA approach. Moreover, after applying the second step, obtaining the subcarrier
allocation of iteration t + 1 with the given power allocation p = pt+1 , we will have f (ρt , pt+1 ) ≥ f (ρt+1 , pt+1 ) because
for a given feasible power solution, the subcarrier allocation problem (13) improves the objective function after each
iteration. Finally, we will have:

... ≥ f (ρt , pt ) ≥ f (ρt , pt+1 ) ≥ f (ρt+1 , pt+1 ) ≥ ... ≥ f (ρ∗ , p∗ ), (12)

where, ρ∗ and p∗ is the final solution obtained at the last iteration. Therefore, after each iteration, the duality gap
G = || f (ρ∗ , p∗ ) − f (ρt , pt )|| value decreases, and which the duality gap value at the last iteration will be very small,
which guarantees the convergence. Therefore, the optimal total transmit power is bounded below, due to the finite set
of transmit sum powers and channel gains and the procedure must converge [28].

9
Mostafa Alhusseini, Paeiz Azmi, Nader Mokari / Physical Communication 00 (2018) 1–28 10

3.2. Subcarrier Allocation Optimization

For obtaining the subcarrier allocation ρt+1 when p = pt+1 , the optimization problem (11) can be expressed as

follows:

Na X
X Ka X
Nb X
Nc
min ρka,b,c pka,b,c , (13a)
ρ
a=1 k=1 b=1 c=1
Na X
X Ka X
Nb X
Nc
s.t. ρka,b,c pka,b,c ≤ ptot , (13b)
a=1 k=1 b=1 c=1
Ka X
X Nb X
Nc
ρka,b,c pka,b,c ≤ pmax
a , ∀a, (13c)
k=1 b=1 c=1
Nc
X
ρka,b,c ra,b,c
k
≥ Rmin , ∀a, k, b, (13d)
c=1
Ka
X
ρka,b,c ≤ Umax , ∀a, b, c, (13e)
k=1

ρm n m m
a,b,c ρa,b,c (S INRa,b,c (m) − S INRa,b,c (n)) ≤ 0, ∀a ∈ Na , b ∈ Nb , c ∈ Nc , m, n ∈ Ka ,

|hm 2 n 2
a,b,c | ≤ |ha,b,c | , m , n, (13f)

ρka,b,c ∈ {0, 1}; ∀a, b, c, k. (13g)

Presently, the subcarrier allocation problem is an INLP, in order to solve it, we use nonlinear optimization solver with

the mesh adaptive direct search algorithm (NOMAD).

10
Mostafa Alhusseini, Paeiz Azmi, Nader Mokari / Physical Communication 00 (2018) 1–28 11

3.3. Power Allocation Optimization

By using the subcarrier allocation ρt , the power allocation is obtained by solving the following problem:

Na X
X Ka X
Nb X
Nc
min ρka,b,c pka,b,c , (14a)
p
a=1 k=1 b=1 c=1
Na X
X Ka X
Nb X
Nc
s.t. ρka,b,c pka,b,c ≤ ptot , (14b)
a=1 k=1 b=1 c=1
Ka X
X Nb X
Nc
ρka,b,c pka,b,c ≤ pmax
a , ∀a, (14c)
k=1 b=1 c=1
Nb X
X Nc
ρka,b,c ra,b,c
k
≥ Rmin, ∀a, k, (14d)
b=1 c=1

ρm n m m
a,b,c ρa,b,c (S INRa,b,c (m) − S INRa,b,c (n)) ≤ 0, ∀a ∈ Na , b ∈ Nb , c ∈ Nc , m, n ∈ Ka ,

|hm 2 n 2
a,b,c | ≤ |ha,b,c | , m , n, (14e)

pka,b,c ≥ 0; ∀a, b, c, k. (14f)

Clearly, this optimization problem is non-convex. Therefore, different approaches are proposed to solve this problem.

In this section, we use the SCA approach to approximate the non-convex problem with a convex problem and near

optimal solution is obtained. Moreover, it can be analytically proved that the alternating algorithm based on this

approach have convergence point [29].


Lemma 1. Constraint (14e) is a linear constraint with respect to variables P.

Proof 2. To prove Lemma (2), we have

ρm n m m
a,b,c ρa,b,c (S INRa,b,c (m) − S INRa,b,c (n)) ≤ 0
2 m
|hm
a,b,c | pa,b,c
⇒ PNb PK v

N0 Bc + i=1,i,b ρm m 2 m m
a,i,c |ha,i,c | pa,i,c + |ha,b,c |
2
k=1,|hm 2 v 2 ρm
a,b,c pa,b,c
a,b,c | ≤|ha,b,c |

|hna,b,c |2 pm
a,b,c
PNb PK v
N0 Bc + i=1,i,b ρm n 2 m n
a,i,c |ha,i,c | pa,i,c + |ha,b,c |
2
k=1,|hm 2 v 2 ρm
a,b,c pa,b,c (15)
a,b,c | ≤|ha,b,c |

K
X
⇒ N0 Bc |hm 2 n 2 m 2 n 2
a,b,c | − N0 Bc |ha,b,c | + |ha,b,c | |ha,b,c | [ ρm v
a,b,c pa,b,c ]
v
k=1,|hm 2
a,b,c | ≤|ha,b,c |
2

Nb
X Nb
X
+ |hm
a,b,c |
2
ρm n 2 m n
a,i,c |ha,i,c | pa,i,c − |ha,b,c |
2
ρm m 2 m
a,i,c |ha,i,c | pa,i,c ≤ 0.
i=1,i,b i=1,i,b

11
Mostafa Alhusseini, Paeiz Azmi, Nader Mokari / Physical Communication 00 (2018) 1–28 12

After simplifying (15), the constraint (14e) is written as


Nb X
X K
ρm n m 2 n 2 m
a,b,c ρa,b,c (N0 Bc |ha,b,c | − N0 Bc |ha,b,c | + |ha,b,c |
2
ρva,i,c |hna,i,c |2 pva,i,c
i=1,i,b n=1
(16)
Nb
X K
X
− |hna,b,c |2 ρva,i,c |hm 2 v
a,i,c | pa,i,c ) ≤ 0,
i=1,i,b m=1

which is a linear constraint.

3.3.1. Successive convex approx. for low-complexity

We apply the following lower bound to find the relaxation of the non-convex problem (14) [30], [31]:

α̂ log2 z + β̂ ≤ log2 (1 + z),


z0 z0 (17)
α̂ = , β̂ = log2 (1 + z0 ) − log2 z0 ,
1 + z0 1 + z0

which is tight with equality under chosen value z0 when the constants α̂ and β̂ are chosen as specified above. Applying

approximation (17) to the rate function in constraint (14d), we obtain the following problem:

Na X
X Ka X
Nb X
Nc
min ρka,b,c pka,b,c , (18a)
p
a=1 k=1 b=1 c=1
X Ka X
Na X Nb X
Nc
s.t. ρka,b,c pka,b,c ≤ ptot , (18b)
a=1 k=1 b=1 c=1
Ka X
X Nb X
Nc
ρka,b,c pka,b,c ≤ pmax
a , ∀a, (18c)
k=1 b=1 c=1
Nb X
X Nc
ρka,b,c (αka,b,c log(S INRka,b,c ) + βka,b,c ) ≥ Rmin ; ∀a, k, (18d)
b=1 c=1

(16), (18e)

pka,b,c ≥ 0; ∀a, b, c, k, (18f)

where, the rate function becomes:

R̂(k) (k) (k) (k)


a,b,c (t) = β̂a,b,c (t) + α̂a,b,c (t) log2 (γa,b,c (t)) (19)

S INR(k)
a,b,c (t−1)
α̂(k)
a,b,c (t) = (k) and β̂(k) (k) (k) (k)
a,b,c (t) = log2 (1 + S INRa,b,c (t − 1)) − α̂a,b,c (t) log2 (S INRa,b,c (t − 1)).
1+S INRa,b,c (t−1)

Therefore, we use the SCALE method to convert the non-convex problem(10) to a convex one. This problem (18) is
12
Mostafa Alhusseini, Paeiz Azmi, Nader Mokari / Physical Communication 00 (2018) 1–28 13

still non-convex because the constraint (18d) is not convex. By using the change of variable pka,b,c = exp( p̂ka,b,c ), we

have:

Na X
X Ka X
Nb X
Nc
min ρka,b,c exp( p̂ka,b,c ), (20a)

a=1 k=1 b=1 c=1
Na X
X Ka X
Nb X
Nc
s.t. ρka,b,c exp( p̂ka,b,c ) ≤ ptot , (20b)
a=1 k=1 b=1 c=1
Ka X
X Nb X
Nc
ρka,b,c exp( p̂ka,b,c ) ≤ pmax
a , ∀a, (20c)
k=1 b=1 c=1
Nb X
X Nc
ρka,b,c (αka,b,c log(S INRka,b,c ) + βka,b,c ) ≥ Rmin ; ∀a, k, (20d)
b=1 c=1

(16), (20e)

exp( p̂ka,b,c ) ≥ 0; ∀a, b, c, k. (20f)

After applying pka,b,c = exp( p̂ka,b,c ), the constraint (20e) becomes non-convex. To convert it to a convex constraint, we

use the difference of two convex function (DC) approach [32]. Therefore, the constraint (20e) is replaced by

Nb X
X K
ρm n m
a,b,c ρa,b,c (N0 Bc |ha,b,c |
2
− N0 Bc |hna,b,c |2 + |hm
a,b,c |
2
ρva,i,c |hna,i,c |2 pva,i,c
i=1,i,b n=1
(21)
Nb X
X K
− |hna,b,c |2 ρva,i,c (exp p̂v,(t−1)
a,i,c + exp p̂v,(t−1) v,t
a,i,c ( p̂a,i,c − p̂v,(t−1) m 2
a,i,c ))|ha,i,c | ≤ 0.
i=1,i,b m=1

Thus, problem (20) is rewritten as follows:

X Ka X
Na X Nb X
Nc
min ρka,b,c exp( p̂ka,b,c ), (22a)

a=1 k=1 b=1 c=1
Na X
X Ka X
Nb X
Nc
s.t. ρka,b,c exp( p̂ka,b,c ) ≤ ptot , (22b)
a=1 k=1 b=1 c=1
Ka X
X Nb X
Nc
ρka,b,c exp( p̂ka,b,c ) ≤ pmax
a , ∀a, (22c)
k=1 b=1 c=1
Nb X
X Nc
ρka,b,c (αka,b,c log(S INRka,b,c ) + βka,b,c ) ≥ Rmin ; ∀a, k, (22d)
b=1 c=1

(21), (22e)

exp( p̂ka,b,c ) ≥ 0; ∀a, b, c, k, (22f)

13
Mostafa Alhusseini, Paeiz Azmi, Nader Mokari / Physical Communication 00 (2018) 1–28 14

which is a convex problem [29],[33]. To solve it, we define the Lagrangian function as:

Na X
X Ka X
Nb X
Nc Na X
X Ka X
Nb X
Nc
L(p̂, λ, θ a , ϕ a,k , ν) = ρka,b,c exp( p̂ka,b,c ) + λ( ρka,b,c exp( p̂ka,b,c ) − ptot )
a=1 k=1 b=1 c=1 a=1 k=1 b=1 c=1
Na
X Ka X
X Nb X
Nc Na X
X Ka Nb X
X Nc
+ θa ( ρka,b,c exp( p̂ka,b,c ) − pmax
a )− ϕ a,k ( ρka,b,c (αka,b,c log(S INRka,b,c )
a=1 k=1 b=1 c=1 a=1 k=1 b=1 c=1

(23a)
Nb X
X K
+ βka,b,c ) − Rmin ) + ν(ρm n m 2 n 2 m
a,b,c ρa,b,c (N0 Bc |ha,b,c | − N0 Bc |ha,b,c | + |ha,b,c |
2
ρva,i,c |hna,i,c |2 exp( p̂va,i,c )
i=1,i,b n=1
Nb X
X K
− |hna,b,c |2 ρva,i,c (exp( p̂v,(t−1) v,(t−1) v,t v,(t−1) m 2
a,i,c ) + exp( p̂a,i,c )( p̂a,i,c − p̂a,i,c ))|ha,i,c | ).
i=1,i,b m=1

After simplifying the Lagrangian function, we obtain the solution:

ϕa,k αka,b,c
pka,b,c = Num
, (24)
1 + λ + θa + Den×ln(2)

where:

Nb
X Na
X
Num = ϕa,k αka,b0 ,c |ha,b,c |2,(k) + ϕa0 ,k αka0 ,b,c |ha,b,c |2,(k) +
b0 =1,,b a0 =1,,a
Na
X Nb
X
ϕa0 ,k αka0 ,b0 ,c |ha,b,c |2,(k) .
a0 =1,,a b0 =1,,b

Nb X
X Nc Na X
X Nb X
Nc
Den = (N0 Bc + |ha,i, j |2,(k) + |hl,i, j |2,(k) ).
i=1,,b j=1 l=1,,a i=1 j=1

Algorithm 4 portrays the transmit power allocation algorithm procedures for each iteration in the iterative algorithm

where the output is pt+1 , for the input pt of iteration t. The process of Algorithm 4 ends when s = S or if ||pt,s − pt,s−1 ||

is smaller than Ω2 .

Proposition 2: The SCA approach with the SCALE approximation creates a sequence of improved solutions which

will converge to a local optimum.


t
Proof 3. pt , α̂t and β̂ are the values after iteration t.
PNa PNb PNc k k
Let fk (p; α̂, β̂) = a=1 b=1 c=1 ρa,b,c pa,b,c , ∀k ∈ Ka . It can be demonstrated that the solution of iteration (t − 1),
pt−1 , is a feasible solution for the next iteration t (see [29] and Theorem 1 in [31]). For every user k ∈ Ka , we
t−1
have Rmink = gk (pt−1 ; α̂t−1 , β̂ ) since all minimum rate constraints are active at the optimal solution of problem (24)
14
Mostafa Alhusseini, Paeiz Azmi, Nader Mokari / Physical Communication 00 (2018) 1–28 15

Table 4: Transmit Power Allocation Algorithm

(k),s (k),s
INITIALIZE: s = 0, pt,s = pt , all α̂a,b,c = 1 and all β̂a,b,c =0
REPEAT
• Solve (22) then give the solution to pt,s
(k),s+1 (k),s+1
• Update all α̂a,b,c and all β̂a,b,c at pt,s
• s= s+1
UNTIL ||pt,s − pt,s−1 || ≤ Ω2 or s = S
OUTPUT pt+1 = pt,S

t−1
(see Lemma 2 in [32]), then we have Rmin
k = gk (pt−1 ; α̂t−1 , β̂ ) ≥ gk (pt−1 ) which comes from (19), then Rmin
k =
t−1 t−1 t−1 t−1 t−1 t t
gk (p ; α̂ , β̂ ) ≥ gk (p ) = gk (p ; α̂ , β̂ ) due to the update step of α̂ and β̂ in the power allocation algorithm
shown in Table III and (14) (see [29] and Theorem 1 in [31]). Moreover, we have:
t t t t−1
fk (pt ; α̂t , β̂ ) = minp fk (p; α̂t , β̂ ) ≤ fk (pt−1 ; α̂t , β̂ ) = fk (pt−1 ) ≤ fk (pt−1 ; α̂t−1 , β̂ ).

Therefore, after each iteration t, the objective function value either improves or stays unaltered as the previous itera-
tion (t − 1) value. Thus, The SCA approach will converges to the last feasible solution p∗ obtained since the feasible
region of the power allocation problem (14) is compact.

4. Optimal solution

As the proposed solution of the joint subcarrier and power allocation problem is local, hence, it is necessary

to compare proposed solution to global solution. Since the objective function and the constraints are increasing

function or changeable to difference of two increasing function. We can convert proposed optimization problem to

monotonic form that has a global solution. We design a joint power and subcarrier allocation algorithm for solving

the monotonic optimization problem in (11) based on the outer polyblock approximation approach [23]–[25]. In this

section, we solve problem (11) using the monotonic optimization method that provides algorithms to globally solve

certain classes of non-convex problems. Monotonic optimization has been previously used to globally solve power

control and scheduling problems, and develop rate maximization schemes and beamforming techniques. Recent

surveys of monotonic optimization have applied it to wireless communications [26], [34] and [35].

Definition 4: (Monotonicity) A function g is monotonically increasing if g(a) ≥ g(b) when a  b.

Definition 5: (Hyper-rectangle). If c  d then, the set of all a where c  a  d is a hyper-rectangle in [c, d].

Definition 6: (Normal set and Co-normal set). A set S1 is a normal if ∀ a ∈ S1 , the hyper-rectangle [0, a] ∈ S1 .

A set S2 is a co-normal set in [0, d], if ∀ a ∈ S2 then [a, d] ⊂ S2 .

Definition 7: (Monotonic optimization). A monotonic optimization problem in canonical form is defined as:

max f (a) s.t. a ∈ S1 ∩ S2 , (28)


a
15
Mostafa Alhusseini, Paeiz Azmi, Nader Mokari / Physical Communication 00 (2018) 1–28 16

where f is an increasing function, S1 ⊂ [0, d] is a normal set with non-empty interior and S2 is a closed co-normal

set in [0, d].

The proposed problem contains two parameters: (p and ρ). Firstly, we eliminate parameter ρ, then problem (11)

becomes an optimization problem with only power variables. Secondly, we transform it to the monotonic optimization

problem in a canonical form. For low complexity, on each subcarrier, the maximum of three users can be assigned ,

Umax = 3, where, it can be solved for any number of users. Then the subcarrier allocation constraints (11e) and (11g),

can be changed by:

x
pa,b,c pya,b,c pza,b,c pqa,b,c ≤ 0, ∀a ∈ Na , b ∈ Nb , c ∈ Nc , x ∈ Ka , y ∈ Ka , z ∈ Ka , q ∈ Ka , and x , y , z , q. (29)

Thus, the considered problem (11) can be rewritten as:

Na X
X Ka X
Nb X
Nc
min pka,b,c , (30a)
p
a=1 k=1 b=1 c=1
X Ka X
Na X Nb X
Nc
s.t. pka,b,c ≤ ptot , (30b)
a=1 k=1 b=1 c=1
Ka X
X Nb X
Nc
pka,b,c ≤ pmax
a , ∀a, (30c)
k=1 b=1 c=1
Nb X
X Nc
k
ra,b,c ≥ Rmin , ∀a, k, (30d)
b=1 c=1
Nb X
X K
Pm n m
a,b,c Pa,b,c (N0 Bc |ha,b,c |
2
− N0 Bc |hna,b,c |2 + |hm
a,b,c |
2
|hna,i,c |2 Pva,i,c
i=1,i,b n=1
Nb
X K
X
− |hna,b,c |2 |hm 2 v
a,i,c | Pa,i,c ) ≤ 0, ∀a ∈ Na , b ∈ Nb , c ∈ Nc , m, n ∈ Ka ,
i=1,i,b m=1

|hm 2 n 2
a,b,c | ≤ |ha,b,c | , m , n, (30e)

(29), (30f)

pka,b,c ≥ 0, ∀a, b, c, k. (30g)

Because of the constraints (30d) and (30e), problem (30) is non-monotonic problem. First, we write the problem (30)

as a monotonic optimization problem in a canonical form, then we use the polyblock algorithm to globally solve (30)

[36], [37].
2
k
|hka,b,c | pka,b,c
ra,b,c = log(1 + PNb PK ). (31)
N0 Bc + i=1,i,b |hka,i,c |2 pka,i,c + k=1,|h k 2 n 2 |hka,b,c |2 pna,b,c
a,b,c | ≤|ha,b,c |

16
Mostafa Alhusseini, Paeiz Azmi, Nader Mokari / Physical Communication 00 (2018) 1–28 17

k
Let ra,b,c = qk+ k− k+ k−
a,b,c − qa,b,c ; where qa,b,c and qa,b,c are equivalent to:

Nb
X K
X
qk+ k 2 k
a,b,c = log(|ha,b,c | pa,b,c + N0 Bc + |hka,i,c |2 pka,i,c + |hka,b,c |2 pna,b,c ).
i=1,i,b k=1,|hka,b,c |2 ≤|hna,b,c |2

Nb
X K
X
qk−
a,b,c = log(N0 Bc + |hka,i,c |2 pka,i,c + |hka,b,c |2 pna,b,c ).
i=1,i,b k=1,|hka,b,c |2 ≤|hna,b,c |2

The constraint (30d) can be equivalently rewritten as follows:

q̃k+ k−
a,b,c − q̃a,b,c ≥ 0, (32)

PNb PNc PNb PNc


where, q̃k+
a,b,c = b=1 c=1 qk+ k−
a,b,c and q̃a,b,c = b=1 c=1 qk−
a,b,c + R
min
.

The constraint (32) is a difference of two increasing function. Similarly as constraint (30d), we convert the constraint

(30e) to the following constraint:


k+ k−
T a,b,c − T a,b,c ≥ 0, (33)

P Nb PK
k+
where, T a,b,c = pm n m 2 n 2 m
a,b,c pa,b,c (N0 Bc |ha,b,c | − N0 Bc |ha,b,c | + |ha,b,c |
2
i=1,i,b n=1 |hna,i,c |2 pva,i,c )
k− 2 PNb PK m 2 v
and T a,b,c = pm n n
a,b,c pa,b,c (|ha,b,c | i=1,i,b m=1 |ha,i,c | pa,i,c ).

Define pmask = {pk,mask


a,b,c }, ∀a, b, c, k. Thus we can introduce the auxiliary variables S1 and S2 , and problem (30e)

is reformulated as:

X Ka X
Na X Nb X
Nc
min pka,b,c , (34a)
p,S1 ,S2
a=1 k=1 b=1 c=1

s.t. (30b), (30c), (30 f ), (30g),

0 ≤ S1 ≤ q̃− (pmask ) − q̃− (0), (34b)

q̃− (p) + S1 ≤ q̃− (pmask ), (34c)

q̃+ (p) + S1 ≥ q̃− (pmask ), (34d)

0 ≤ S2 ≤ T − (pmask ) − T − (0), (34e)

T − (p) + S2 ≤ T − (pmask ), (34f)

T + (p) + s2 ≥ T − (pmask ). (34g)

Thus, problem (34) is tronsformed to a monotonic problem in a canonical form because it achieved definition (7).

17
Mostafa Alhusseini, Paeiz Azmi, Nader Mokari / Physical Communication 00 (2018) 1–28 18

Therefore, the polyblock algorithm is used to obtain the global solution for problem (34).

5. Computational Complexity

We discuss the computational complexity of our proposed scheme and the monotonic scheme. The computational

complexity of the global optimal approach using the polyblock algorithm increases quickly depending on the problem

dimensions, and it is exponential in the problem size. Hence, for the problems with large variables, the global optimal

approach can be only a benchmark for the low complexity approaches that converges to a local optima with lower

computational complexity. Computational complexity of the polyblock algorithm heavily depends on the form of the

functions which provides the normal set and number of variables. The polyplock algorithm contains four main steps

as:

1) Finding the best vertex which its projection belongs to the normal set.

2) Finding the projection of selected vertex.

3) Eliminating the improper vertexes.

4) Finding the new vertex set.

If we consider that the dimensions of the optimization problem is D1, overall polyblock algorithm converges after

D2 iterations and the projection of each vertex is given by the bisection algorithm with D3 iterations, a simplified

complexity order can be given by [R8]:

O(D2(D2 × D1 + D3 )), (35)

As mentioned the power and subcarrier allocation algorithm include two stages: (i) determining the subcarrier

allocation from problem (13) which is integer non linear program (INLP) by using the mesh adaptive algorithm search

method, (ii) solving power allocation problem (14) by using the SCA approach. The complexity of the subcarrier

allocation problem depends on the number of variables and constraints where the number of variables and constraints

are (Na × Ka × Nb × Nc ) and (1 + Na + Na × Nb × Ka + Na × Nb × Nc + 2(Na × Nb × Ka × Nc )), respectively [R9, R10].


log((1+Na +Na ×Nb ×Ka +Na ×Nb ×Nc +2(Na ×Nb ×Ka ×Nc )))/t0 η
Hence, The number of required iterations for subcarrier allocation is log(ξ) , ξ

Table 5: Computational complexity

Approach Proposed approach Monotonic approach


0
Subcarrier allocation Csc = O( log((1+Na +Na ×Nb ×Ka +Na ×N b ×Nc +2(Na ×Nb ×Ka ×Nc )))/t η
log(ξ) )
Power allocation C p = O(Na × Ka × Nc )(Ka + Na + (Nc × Ka )) O(D2(D2 × D1 + D3 ))
Total Ct = C s c + C p O(D2(D2 × D1 + D3 ))
18
Mostafa Alhusseini, Paeiz Azmi, Nader Mokari / Physical Communication 00 (2018) 1–28 19

Monotonic Algorithm Na=6(MISO)


590 Monotonic Algorithm Na=3(MISO)
Monotonic Algorithm Na=1(SISO)
Proposed Algorithm Na=6(MISO)
585 Proposed Algorithm Na=3(MISO)
Proposed Algorithm Na=1(SISO)
580

Total transmit power (Watts)


575

570

565

560

555

550
10 20 30 40 50 60 70 80 90
Number of beams

Figure 2: Total transmit Power vs. number of beams for proposed and monotonic algorithm.

is used for the accuracy updating of interior point method (IPM), η is the stopping criterion for IPM and t0 is initial

point for approximated the accuracy of IPM [29, 30]. When the SCALE algorithm is applied, the power allocation is

determined by (20) and (22) with O(Na × Ka × Nc ) computational complexity. While the computational complexity for

the dual variables, where the sub-gradient method is applied, is equal to O(Ka +Na +(Nc ×Ka )) [30]. The computational

complexity of the proposed scheme and those of conventional scheme are indicated in Table 5.

6. Results and discussion

The objectives of the simulation are: Firstly, to evaluate the performance of our proposed system model scenario,

joint subcarrier and power allocation for MISO multi-spot-beam satellite system. Secondly, to compare our novel

system design with the conventional design. The following numerical simulations are purposed to explain the behavior

of the proposed approach and to compare it with the optimal approach (monotonic approach) from Section IV. The

simulations are executed by using the alternating algorithm according to P1 and P2, that jointly allocates the subcarrier

and transmit power to each user. The parameters are chosen to be the same as those employed by the Satellite Task

Force of DVB-NGH [38]. The multi-spot-beam antenna has equipped with (Na × Nb ) beams. The basic parameters

required for the link budget are tabulated in Table 6.

The channel model is based on two scenarios: a) clear sky, and b) rainy sky. a) clear sky: In Fig. (2-4) we
k(−τ)
suppose that the channel model is Rayleigh fading, where hka,b,c = ωka,b,c da,b,c ; τ = 2 is the path loss exponent,

ωka,b,c is an exponential random variable (i.e. representing the Rayleigh fading), and da,b,c
k
is the distance between

user k and beam b from antenna a. b) While in Fig. 5, we suppose that the model channel is lognormal, where
q
hka,b,c = G sGe PLa,b,c
k k
ϑka,b,c ; PLa,b,c is the Free Space Lose coefficient of user k in beam b from antenna a depends on

wavelength and the distance between the satellite and the user. The power attenuation factor ϑka,b,c due to rain fading, it
19
Mostafa Alhusseini, Paeiz Azmi, Nader Mokari / Physical Communication 00 (2018) 1–28 20

Table 6: Simulation Parameters

Parameter value
Orbit GEO
Frequency Band Ka (20GHz)
Satellite EIRP 26 dBW
Gs(0): Satellite peak antenna gain 47.2 dBi
Ge(0): Earth station peak antenna gain 25 dBi
Ds.: Onboard satellite antenna diameter 1.5 m
De. : Gateway antenna diameter 9.1 m
B(tot) : Total Bandwidth 500 MHz
B(c) : Carrier bandwidth 100 MHz
P(tot) : Total transmit power 500 Watts
Pmax
a : maximum transmit power for each antenna a ∈ Na 100 Watts
N0 : noise power spectral density −102 dBm

Monotonic Algorithm Na=6 (MISO)


550 Monotonic Algorithm Na=3 (MISO)
Monotonic Algorithm Na=1 (SISO)
Proposed Algorithm Na=6 (MISO)
Proposed Algorithm Na=3 (MISO)
545
Proposed Algorithm Na=1 (SISO)
Total transmit power (Watts)

540

535

530

525

520
10 15 20 25 30 35 40 45 50
Number of Users

Figure 3: Total transmit Power vs. number of users for proposed and monotonic algorithm.

can be seen from the model of rain fading that the model of channel gain obeys independent and identically lognormal

distribution [39].

Fig. 2 describes the total transmit power on each beam for the proposed method and monotonic approach by

considering clear sky. We can see that, the MISO and PD-NOMA techniques reduce the total transmit power of the

MISO multi-spot-beam satellite system by approximately 20% with respect to the SISO and OMA techniques. And

when Na = 6 and Nb = 50, more than 20 Watts power gain can be reduced by the monotonic approach, then we

have 20 Watts approximately distance gap between the proposed method and the monotonic approach. Fig. 3 shows

that the total transmit power evolution for the number of users Ka considering clear sky for the proposed Algorithm

and Monotonic algorithm. From Fig. 3 can be observed that the total transmit power increases with increasing the
20
Mostafa Alhusseini, Paeiz Azmi, Nader Mokari / Physical Communication 00 (2018) 1–28 21

580
Monotonic Algorithm Na=6 (MISO)
Monotonic Algorithm Na=3 (MISO)
Monotonic Algorithm Na=1 (SISO)
570 Proposed Algorithm Na=6 (MISO)
Proposed Algorithm Na=3 (MISO)

Total transmit power (Watts)


Proposed Algorithm Na=1 (SISO)
560

550

540

530

520
5 10 15 20 25
Minimum Rate requirement (Mbps)

Figure 4: Total transmit Power vs. min rate requirement Rmin for proposed and monotonic algorithm.

605
Proposed Algorithm fo NOMA
Proposed Algorithm for OMA
600
Total transmit power (Watts)

595

590

585

580

575

570
10 20 30 40 50 60 70 80 90
Number of beams

Figure 5: Total transmit Power vs. number of beams for proposed algorithm.

number of users Ka , but the increasing is lower than as compared to before case in Fig.2, Therefore, the distance gap

between proposed method and monotonic approach is decreased. Fig. 4 shows that the total transmit power evolution

for the minimum requirements rate, The total transmit power also depends on the minimum requirements rate Rmin

for both the proposed and Monotonic approach. In Fig. (2, 3, 4), simulation results show that the total transmit power

for multi-spot-beam satellite system by applying the proposed approach is close to the total transmit power by apply-

ing the monotonic approach and the MISO and PD-NOMA techniques reduce the total transmit power of the MISO

multi-spot-beam satellite system with respect to the SISO and OMA techniques.

In Fig. (5, 6, 7), a comparison between NOMA and OMA (Umax = 1) systems is evaluated. We show that the total

transmit power for NOMA multi-spot-beam satellite system is below of OMA system. Thus, the total transmit power

is decreased by applying NOMA technique for NOMA multi-spot-beam satellite system.

21
Mostafa Alhusseini, Paeiz Azmi, Nader Mokari / Physical Communication 00 (2018) 1–28 22

555
Proposed Algorithm fo NOMA
Proposed Algorithm for OMA

550

Total transmit power (Watts)


545

540

535

530
10 15 20 25 30 35 40 45 50
Number of Users

Figure 6: Total transmit Power vs. number of users for proposed algorithm.

560
Proposed Algorithm fo NOMA
Proposed Algorithm for OMA
555
Total transmit power (Watts)

550

545

540

535

530

525
5 10 15 20 25
Minimum Rate requirement (Mbps)

Figure 7: Total transmit Power vs. min rate requirement Rmin for proposed algorithm.

The computational time in seconds for the proposed approaches versus the number of beams is shown in Fig.

(10). We can see that the proposed algorithm for NOMA (U(max) = 5) computational time is larger than both NOMA

(U(max = 3) and OMA (U(max) = 1) where the computational time is obtained by core i7 CPU and 8.00 GB RAM.

Fig. 9 illustrates the variation of the total transmit power versus number of iterations under the proposed algorithm

in Table 3, for the alternating method. This figure shows that both the MISO-NOMA and SISO-NOMA actually

converge. Moreover, the overall algorithm converges faster for proposed algorithm when Na = 6 with respect to

Na = 3 and Na = 1.

The computational time in seconds for the proposed approaches versus the number of beams is shown in Fig.

(10). We can see that the proposed algorithm for NOMA (U(max) = 5) computational time is larger than both NOMA

(U(max = 3) and OMA (U(max) = 1) where the computational time is obtained by core i7 CPU and 8.00 GB RAM.

Fig. 11 describes the total transmit power on each beam for the proposed method by considering rainy sky. It

22
Mostafa Alhusseini, Paeiz Azmi, Nader Mokari / Physical Communication 00 (2018) 1–28 23

Figure 8: Computational time for proposed and existing NOMA/OMA schemes.

580
Proposed Algorithm Na=6 (MISO)
Proposed Algorithm Na=3 (MISO)
Proposed Algorithm Na=1 (SISO)
570
Total transmit power (Watts)

560

550

540

530

520
20 40 60 80 100 120 140 160 180 200
Number of iterations

Figure 9: Convergence of the proposed algorithm in Table 3, for the alternating method.

can be observed that changing from clear sky to rainy sky results in a dramatic increase of transmitted power. The

effect of rain attenuation parameter is directly related to the rain attenuation that results in the increase of transmitted

power of rainy beams. Also the effectiveness of the proposed scenario and methods is considered in clear and rainy

sky situations.

23
Mostafa Alhusseini, Paeiz Azmi, Nader Mokari / Physical Communication 00 (2018) 1–28 24

Figure 10: Computational time for proposed and existing NOMA/OMA schemes.

Proposed Algorithm Na=6(MISO) with rainy sky


650 Proposed Algorithm Na=3(MISO) with rainy sky
Proposed Algorithm Na=1(SISO) with rainy sky
640 Proposed Algorithm Na=6(MISO)
Proposed Algorithm Na=3(MISO)
Proposed Algorithm Na=1(SISO)
630
Total transmit power (Watts)

620

610

600

590

580

570

560
10 20 30 40 50 60 70 80 90
Number of beams

Figure 11: Total transmit Power vs. Number of beams for proposed algorithm.

7. Conclusion

We proposed a novel joint subcarrier and power allocation problem for MISO multi-spot-beam satellite system

design based on PD-NOMA. The considered problem minimizes the total transmit power subject to subcarrier and

transmit power constraints. To solve this problem, an alternating method was used. The subcarrier allocation was

updated by using the INLP and the transmit power values were updated by using the SCA approach. Moreover, the

proposed problem was solved by using the Monotonic Optimization method in order to find the optimal solution.

Simulation results showed that the total transmit power by applying the proposed approach was close to the total

transmit power by applying the monotonic approach. In other words, the proposed low complexity local solution

is near optimal. The results demonstrated that the PD-NOMA system for MISO multi-spot-beam satellite system

reduced the total transmit power. Furthermore, comparison with previous studies proved that the proposed method

can be achieved with better coverage, low complexity, and low cost for the satellite systems. As a future work, we
24
Mostafa Alhusseini, Paeiz Azmi, Nader Mokari / Physical Communication 00 (2018) 1–28 25

study the performance of the proposed system model with imperfect channel state information. Another direction of

future research would be to study joint uplink and downlink transmission by considering both transmitter and receiver

power consumption.
Acknowledgment The first author would like to thank Mr. Moslem Forouzesh for their remarks and helps.
References
[1] F.R. Durand, T. Abrão, Power allocation in multibeam satellites based on particle swarm optimization, AEU-International Journal of Electron-
ics and Communications. 78 (2017) 124–133. https://doi.org/10.1016/j.aeue.2017.05.012.
[2] G. Maral, M. Bousquet, Z. Sun, Communications Systems: Systems, Techniques and Technology, John Wiley and Sons, 5th edition, New
York, NY, USA, 2009.
[3] Z. Katona, F. Clazzer, K. Shortt, S. Watts, H.P. Lexow, R. Winduratna, Performance, cost analysis, and ground segment design of ultra high
throughput multispot beam satellite networks applying different capacity enhancing techniques, International Journal of Satellite Communica-
tions and Networking. 34 (2016) 547–573. https://doi.org/10.1002/sat.1160.
[4] Y. Vasavada, R. Gopal, C. Ravishankar, G. Zakaria, N. BenAmmar, Architectures for next generation high throughput satellite systems,
International Journal of Satellite Communications and Networking. 34 (2016) 523–546. https://doi.org/10.1002/sat.1175.
[5] O. Vidal, G. Verelst, J. Lacan, E. Alberty, J. Radzik, M. Bousquet, Next generation high throughput satellite system, in: Satellite Telecommu-
nications (ESTEL), IEEE First AESS European Conference, IEEE, 2012: pp. 1–7.
[6] T. Yunzheng, L. Long, L. Shang, Z. Zhi, A survey: Several technologies of non-orthogonal transmission for 5G, China Communications. 12
(2015) 1–15. https://doi.org/10.1109/CC.2015.7315054.
[7] S. Chen, J. Zhao, The requirements, challenges, and technologies for 5G of terrestrial mobile telecommunication, IEEE Communications
Magazine. 52 (2014) 36–43.
[8] M. Caus, M.Á. Vázquez, A. Pérez-Neira, NOMA and interference limited satellite scenarios, in: Signals, Systems and Computers, 50th
Asilomar Conference, IEEE, 2016: pp. 497–501.
[9] Y. Saito, Y. Kishiyama, A. Benjebbour, T. Nakamura, A. Li, K. Higuchi, Non-orthogonal multiple access (NOMA) for cellular future radio
access, in: Vehicular Technology Conference (VTC Spring), IEEE 77th, IEEE, 2013: pp. 1–5.
[10] P.D. Arapoglou, K. Liolis, M. Bertinelli, A. Panagopoulos, P. Cottis, R. De Gaudenzi, MIMO over satellite: A review, IEEE Communications
Surveys and Tutorials. 13 (2011) 27–51. https://doi.org/10.1109/SURV.2011.033110.00072.
[11] J. Kyröläinen, A. Hulkkonen, J. Ylitalo, A. Byman, B. Shankar, P. Arapoglou, J. Grotz, Applicability of MIMO to satellite communications,
International Journal of Satellite Communications and Networking. 32 (2014) 343–357. https://doi.org/10.1002/sat.1040.
[12] A.I. Aravanis, B.S. MR, P.-D. Arapoglou, G. Danoy, P.G. Cottis, B. Ottersten, Power allocation in multibeam satel-
lite systems: A two-stage multi-objective optimization, IEEE Transactions on Wireless Communications. 14 (2015) 3171–3182.
https://doi.org/10.1109/TWC.2015.2402682.
[13] A.D. Panagopoulos, P.D.M. Arapoglou, P.G. Cottis, Satellite communications at Ku, Ka, and V bands: Propagation impairments and mitiga-
tion techniques, IEEE Communications Surveys and Tutorials. 6 (2004) 2–14. https://doi.org/10.1109/COMST.2004.5342290.
[14] H. Wang, A. Liu, X. Pan, J. Li, Optimization of power allocation for a multibeam satellite communication system with interbeam interference,
Journal of Applied Mathematics. 2014 (2014) 1–8. https://doi.org/10.1155/2014/469437.
[15] F. Qi, L. Guangxia, F. Shaodong, G. Qian, Optimum power allocation based on traffic demand for multi-beam satellite communication
systems, in: Communication Technology (ICCT), IEEE 13th International Conference, IEEE, 2011: pp. 873–876.
[16] M. Moltafet, P. Azmi, N. Mokari, Power minimization in 5G heterogeneous cellular networks, in: Electrical Engineering (ICEE), 24th Iranian

25
Mostafa Alhusseini, Paeiz Azmi, Nader Mokari / Physical Communication 00 (2018) 1–28 26

Conference, IEEE, 2016: pp. 234–238.


[17] A. Byman, A. Hulkkonen, P.D. Arapoglou, M. Bertinelli, R. De Gaudenzi, MIMO for mobile satellite digital broadcasting: From theory to
practice, IEEE Transactions on Vehicular Technology. 65 (2016) 4839–4853. https://doi.org/10.1109/TVT.2015.2462757.
[18] H. Li, J. Li, Y. Cheng, J. Wu, Capacity Analysis and Optimization of Satellite MIMO System, International Journal of Communications,
Network and System Sciences. 10 (2017) 116–126. https://doi.org/10.4236/ijcns.2017.105B011.
[19] X. Xiao, X. Tao, J. Lu, QoS-guaranteed energy-efficient power allocation in downlink multi-user MIMO-OFDM systems, in: Communica-
tions (ICC), IEEE International Conference, IEEE, 2014: pp. 3945–3950.
[20] S. Zhang, B. Di, L. Song, Y. Li, Radio resource allocation for non-orthogonal multiple access (NOMA) relay network using matching game,
in: Communications (ICC), IEEE International Conference, IEEE, 2016: pp. 1–6.
[21] M. Caus, M.Á. Vázquez, A. Pérez-Neira, NOMA and interference limited satellite scenarios, in: Signals, Systems and Computers, 50th
Asilomar Conference, IEEE, 2016: pp. 497–501.
[22] U. Park, H.W. Kim, D.S. Oh, B. Ku, A dynamic bandwidth allocation scheme for a multi spotbeam satellite system, Etri Journal. 34 (2012)
613–616. https://doi.org/10.4218/etrij.12.0211.0437.
[23] F. Qi, L. Guangxia, F. Shaodong, G. Qian, Optimum power allocation based on traffic demand for multi-beam satellite communication
systems, in: Communication Technology (ICCT), IEEE 13th International Conference, IEEE, 2011: pp. 873–876.
[24] S. Shi, G. Li, Z. Li, H. Zhu, B. Gao, Joint power and bandwidth allocation for beam-hopping user downlinks in smart gateway multibeam
satellite systems, International Journal of Distributed Sensor Networks. 13 (2017) 1–11. https://doi.org/10.1177/1550147717709461.
[25] A. Mokdad, P. Azmi, N. Mokari, Radio resource allocation for heterogeneous traffic in GFDM-NOMA heterogeneous cellular networks, IET
Communications. 10 (2016) 1444–1455. https://doi.org/10.1049/iet-com.2016.0011.
[26] Y. Sun, D. Ng, Z. Ding, and R. Schober, ”Optimal joint power and subcarrier allocation for full-duplex multicarrier non-orthogonal multiple
access systems”, IEEE Transactions on Communications, 65(3), pp.1077-1091, 2017.
[27] T. Wang, L. Vandendorpe, Iterative resource allocation for maximizing weighted sum min-rate in downlink cellular OFDMA systems, IEEE
Transactions on Signal Processing. 59 (2011) 223–234. https://doi.org/10.1109/TSP.2010.2078811.
[28] L. Venturino, N. Prasad, X. Wang, Coordinated scheduling and power allocation in downlink multicell OFDMA networks, IEEE Transactions
on Vehicular Technology. 58 (2009) 2835–2848. https://doi.org/10.1109/TVT.2009.2013233.
[29] D.T. Ngo, S. Khakurel, T. Le-Ngoc, Joint subchannel assignment and power allocation for OFDMA femtocell networks, IEEE Transactions
on Wireless Communications. 13 (2014) 342–355. https://doi.org/10.1109/TWC.2013.111313.130645.
[30] T. Wang and L. Vandendorpe, ”Iterative resource allocation for maximizing weighted sum min-rate in downlink cellular OFDMA systems”,
IEEE Transactions Signal Processing, 59(1), pp.223–234, 2011.
[31] J. Papandriopoulos, J.S. Evans, SCALE: A low-complexity distributed protocol for spectrum balancing in multiuser DSL networks, IEEE
Transactions on Information Theory. 55 (2009) 3711–3724. https://doi.org/10.1109/TIT.2009.2023751.
[32] N. Mokari, F. Alavi, S. Parsaeefard, T. Le-Ngoc, Limited-Feedback Resource Allocation in Heterogeneous Cellular Networks. IEEE Trans-
actions on Vehicular Technology. 65 (2016) 2509–2521. https://doi.org/10.1109/TVT.2015.2428997.
[33] S. Boyd, L. Vandenberghe, Convex optimization, Cambridge university press, 2004.
[34] A. Zappone, E. Björnson, L. Sanguinetti, E. Jorswieck, A framework for globally optimal energy-efficient resource allocation in wireless
networks, in: Acoustics, Speech and Signal Processing (ICASSP), IEEE International Conference, IEEE, 2016: pp. 3616–3620.
[35] M. Moltafet, P. Azmi, N. Mokari, M.R. Javan, A. Mokdad, Optimal and Fair Energy Efficient Resource Allocation for
Energy Harvesting-Enabled-PD-NOMA-Based HetNets, IEEE Transactions on Wireless Communications. 17 (2018) 2054–2067.
https://doi.org/10.1109/TWC.2017.2788406.

26
Mostafa Alhusseini, Paeiz Azmi, Nader Mokari / Physical Communication 00 (2018) 1–28 27

[36] H. Tuy, Monotonic optimization: Problems and solution approaches, SIAM Journal on Optimization. 11 (2000) 464–494.
https://doi.org/10.1137/S1052623499359828.
[37] H. Tuy, ”Monotonic optimization: Problems and solution approaches”, SIAM Journal on Optimization, 11(2), pp.464-494, 2000.
[38] U. Reimers, DVB: the family of international standards for digital video broadcasting, Springer, 2013.
[39] T. Qi, Y. Wang, Energy-efficient power allocation over multibeam satellite downlinks with imperfect CSI, in: Wireless Communications and
Signal Processing (WCSP), International Conference, IEEE, 2015: pp. 1–5.
[40] http://www.i2c2.aut.ac.nz/Wiki/OPTI/index.php/ Solvers/NOMAD.
[41] S. Bayat, Y. Li, L. Song, Z. Han, Matching Theory: Applications in wireless communications, IEEE Signal Processing Magazine, 33(2016),
103-122.

Mostafa Alhusseini was born in Baalback, Lebanon, on February 17, 1984. He received the Ph.D. degree in

electrical engineering from Tarbiat Modares University, Tehran, Iran, in 2018. His current research interests include

satellite communications, wireless communications, digital signal processing, SDR, resource allocation.

Paeiz Azmi (M’05-SM’10) was born in Tehran, Iran, on April 17, 1974. He received the B.Sc., M.Sc., and Ph.D.

degrees in electrical engineering from Sharif University of Technology, Tehran, Iran, in 1996, 1998, and 2002, re-

spectively. Since September 2002, he has been with the Electrical and Computer Engineering Department, Tarbiat

Modares University, Tehran, Iran, where he became an Associate Professor on January 2006 and he is a Full Professor

now. From 1999 to 2001, he was with the Advanced Communication Science Research Laboratory, Iran Telecom-

munication Research Center (ITRC), Tehran, Iran. From 2002 to 2005, he was with the Signal Processing Research

Group, ITRC. His current research interests include modulation and coding techniques, digital signal processing,

wireless communications, resource allocation, molecular communications, and estimation and detection.

27
Mostafa Alhusseini, Paeiz Azmi, Nader Mokari / Physical Communication 00 (2018) 1–28 28

Nader Mokari received the Ph.D. degree in electrical engineering from Tarbiat Modares University, Tehran, Iran,

in 2014. He joined the Department of Electrical and Computer Engineering, Tarbiat Modares University, as an Assis-

tant Professor, in 2015. He has been involved in a number of large scale network design and consulting projects in the

telecom industry. His research interests include design, analysis, and optimization of communication networks.

28

You might also like