ProSky NEAT Meets NOMA-mmWave in the Sky of 6G Ahmed Benfaid Nadia Adem and Abdurrahman Elmaghbub

2025-05-02 0 0 1.37MB 7 页 10玖币

侵权投诉

ProSky: NEAT Meets NOMA-mmWave

in the Sky of 6G

Ahmed Benfaid∗, Nadia Adem∗, and Abdurrahman Elmaghbub∗∗

∗University of Tripoli, Tripoli, Libya, E-mail: {a.benfaid,n.adem}@uot.edu.ly

∗∗ Oregon State University, OR, USA, Email: elmaghba@oregonstate.edu

Abstract—Rendering to their abilities to provide ubiquitous

connectivity, ﬂexibly and cost effectively, unmanned aerial vehi-

cles (UAVs) have been getting more and more research attention.

To take the UAVs' performance to the next level, however,

they need to be merged with some other technologies like

non-orthogonal multiple access (NOMA) and millimeter wave

(mmWave), which both promise high spectral efﬁciency (SE). As

managing UAVs efﬁciently may not be possible using model-based

techniques, another key innovative technology that UAVs will

inevitably need to leverage is artiﬁcial intelligence (AI). Designing

an AI-based technique that adaptively allocates radio resources

and places UAVs in 3D space to meet certain communication

objectives, however, is a tough row to hoe. In this paper,

we propose a neuroevolution of augmenting topologies NEAT

framework, referred to as ProSky, to manage NOMA-mmWave-

UAV networks. ProSky exhibits a remarkable performance im-

provement over a model-based method. Moreover, ProSky learns

5.3times faster than and outperforms, in both SE and energy

efﬁciency EE while being reasonably fair, a deep reinforcement

learning DRL based scheme. The ProSky source code is accessible

to use here: https://github.com/Fouzibenfaid/ProSky

Index Terms—Deep reinforcement learning (DRL), millime-

ter wave (mmWave), neuroevolution of augmenting topologies

(NEAT), non-orthogonal multiple access (NOMA), unmanned

aerial vehicle (UAV).

I. INTRODUCTION

Despite the unprecedented advancements in telecommu-

nication technologies in recent years, around half of the

world's population, mostly living in rural and developing

areas, still has limited or no access to cellular communication

services [1]. Providing connectivity to those underprivileged

areas could unequivocally enhance the quality of their lives.

One of the envisioned, must-be-met, requirements of 6G is

enabling ubiquitous geographical coverage anywhere, anytime.

Due to the lack of essential cellular infrastructures in rural

areas and the high cost of establishing them, along with the

incapability of terrestrial base stations (BSs) to cover hotspot

areas during special events or disaster scenarios, unmanned

aerial vehicles (UAVs), thanks to their ﬂexible 3D mobil-

ity and ease of deployment, are envisioned to be a major

part of the 6G wireless networks, acting as ﬂying BSs [2].

Allowing sharing same spectrum resources among multiple

wireless nodes simultaneously, non-orthogonal multiple access

(NOMA) is emerging as another 6G enabler offering massive

device connectivity without exhausting spectrum resources [2].

The emergence of NOMA-aided UAV networks necessitates

a careful investigation of optimal UAV 3D deployment, and

power allocation (PA) management [2]. Due to the high

complexity of such an optimization problem, authors in [3],

for example, approached the UAV placement and PA problems

disjointly, whereas the number of served users was limited to

two in [4], [5], and the UAV mobility was restricted to a 2D

plane in [3]–[5]. There have been some attempts to address

the UAV 3D placement problem, but these have primarily

been accomplished by disjoining the UAV placement and PA

problems, as done in [6].

Furthermore, due to the high probability of line-of-sight

(LoS) links UAVs offer, the vast millimeter wave (mmWave)

and terahertz spectrum can be utilized to satisfy the ma-

jor 6G data rate enhancement requirement [2]. Nevertheless,

managing NOMA-mmWave-UAV network resources to satisfy

dynamic, heterogeneous, and massive needs adaptively yet

fairly and efﬁciently is a complicated challenge, which con-

ventional and even machine learning methods fail to handle.

The good news, however, is that deep learning and, more

generally, artiﬁcial intelligence (AI) technologies have the

strong potential to handle multi-state network statuses and

demands. After proving their effectiveness in solving problems

with a large degree of freedom in various ﬁelds, AI techniques

are proposed to be a key enabler for self-organizing, self-

optimized networks in the 6G era [2].

Inspired by the remarkable successes of incorporating deep

reinforcement learning (DRL) into different ﬁelds, a DRL

model has been used to solve the placement problem of

UAVs that ﬂy at a ﬁxed height [7]. More interestingly, our

previous work, AdaptSky [8], unlike any other work, jointly

solved the non-convex optimization problem of the 3D de-

ployment and the PA of a NOMA-equipped UAV BS in the

mmWave spectrum. Another AI tool that has recently shown

outstanding performance in a variety of applications, including

robotics and gaming [9], is neuroevolution of augmenting

topologies (NEAT) [10]. Furthermore, although fairly limited,

NEAT has been recently used in the area of communications,

for example, in [11] to improve beam management in vehicle-

to-vehicle communications. To the best of our knowledge,

no work has studied the use of NEAT to manage UAV-

based communications. In this work, we propose ProSky,

a novel NEAT-based framework that jointly optimizes 3D

deployment and PA for NOMA-aided UAV BSs operating in

arXiv:2210.11406v1 [eess.SP] 13 Oct 2022

and advancements over [8], are summarized as follows:

(i) ProSky incorporates and demonstrates the effectiveness

of NEAT, for the ﬁrst time, to manage NOMA-mmWave-

UAV networks. We set the NEAT environment that leads

to the optimal UAV placement and NOMA PA such that,

without sacriﬁcing fairness, the total network data rate is

maximized.

(ii) Although DRL based algorithms show tremendous im-

provements over a state-of-art in managing 3D networks

while exhibiting high generalization capabilities [8], they

have two drawbacks: i) specifying their neural network

(NN) structure beforehand and tuning it using trials and

errors does not only lead to degradation in efﬁciency but

also some times in performance, ii) they train, relatively,

slow and run into local minima issues. As NEAT op-

timally determines the NN structure, ProSky, however,

demonstrates, while being reasonably fair, 5.8% improve-

ments in SE and 21.9% in EE over a DRL based bench-

mark. More interestingly, as NEAT simultaneously trains

multiple NNs to evolve based on a genetic algorithm

(GA), ProSky exhibits more than 400% improvement in

learning rate.

II. SYSTEM MODEL

A. Network Model

Embracing the 3D coverage capability of UAVs, and similar

to our work in [8], we consider a 3D downlink cellular

network that covers an area Aof L×Lunits in which

a UAV serves a total of 2N, for some integer number N,

uniformly distributed ground users, NUE. The UAV and ground

users are assumed to be equipped with NUAV and NUE

antennas, respectively. Throughout the paper, user iis denoted

by UEiwhere i∈ {1,2, .., NUE}. We assume that the users are

grouped into clusters, as depicted in Fig. 1, in such a way that

users UEiand UEi+1 for i∈ {1,3, .., NUE −1}are associated

with the same cluster and UEihas a stronger channel gain than

UEi+1, following the distance-based pairing strategy discussed

in [12]. The assumption of having 2Nusers is set only for

convenience and should not affect the model's generality. In

case there is an odd number of users, a cluster will encompass a

single user and every thing else is still valid. The UAV serves

each cluster over an orthogonal power resource with a total

power PT, distributed between the two corresponding users

based on their channel conditions. The received power at the

UEiat a given time step τcan be expressed as

Pi,τ =PTgMIMO

i,τ (di,τ )αi,τ ,(1)

where gMIMO

i,τ (di,τ )is the gain between the UAV and UEi

separated by a 3D distance of di,τ , and αi,τ is the percentage of

the PTassigned to UEi. If we consider only a large scale fading

and assume a slight difference between antenna pairs, the

gain can be approximated as gMIMO

i,τ (di,τ ) = Ggi,τ (di,τ ), where

G, equals NUAV × NUE, is the gain resulting from applying

UE

UE

UE

UE

Cluster 1

Cluster 2

(xUAV,τ , yUAV,τ , hUAV,τ)

(x3 , y3 )

d3,τ Frequency

Power

α1,τ PT

α2,τ PT

Fig. 1: System model.

multiple-input multiple-output (MIMO) antenna conﬁgurations

at the UAV and UEi.gi,τ (di,τ )is the channel gain.

Based on the principle of NOMA, the superposition coding

(SC) is used at the UAV to transmit signals to users located

in the same cluster. SC encodes different signals into a single

signal while assigning them different power values. The suc-

cessive interference cancellation (SIC) is used at the receiver

side for signal detection. The received signal to interference

plus noise ratio at τfor UEi, SINRi,τ , is expressed as

SINRi,τ =PTGgi,τ (di,τ )αi,τ

PTGgi,τ (di,τ )βi,τ +σ2(2)

where βi,τ =αi−1,τ if iis even and zero otherwise. σ2is

the noise power. The ﬁrst term in the denominator of (2)

represents the interference from the user with the stronger

channel gain on the other user in the same cluster. Using the

SIC technique, however, the interference from the user with

the stronger channel gain gets removed. The data rate of UEi

at τis given by

Ri,τ =Wlog2(1 + SIN Ri,τ ),(3)

where Wis the communication bandwidth. The channel gain,

gi,τ (di,τ ), between the UAV and UEiat the mmWave spectrum,

in the presence of the LoS link [13], is expressed as [14]

gi,τ (di,τ ) = Cd−a

i,τ ,(4)

where aand Care the path loss exponent and intercept,

respectively. We assume that the UAV collects the channel

state information at the beginning of each time step [15], in

that a user can send a pilot signal prior to transmission to allow

the UAV to estimate gi,τ .

B. UAV Mobility Model

The UAV is assumed to be located at

(xUAV,τ , yU AV,τ , hUAV,τ )at time step τ, and

it is able to move, in the next time step, to

(xUAV,τ +dxδx, yU AV,τ +dyδy, hUAV,τ +dhδh), where

dx,dy, and dh∈ {1,−1}.δx,δy, and δhare the magnitude

of change in the x,y, and z axes, respectively. hUAV,τ is

assumed to have a minimum value of h0.

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

ProSky:NEATMeetsNOMA-mmWaveintheSkyof6GAhmedBenfaid,NadiaAdem,andAbdurrahmanElmaghbubUniversityofTripoli,Tripoli,Libya,E-mail:fa.benfaid,n.ademg@uot.edu.lyOregonStateUniversity,OR,USA,Email:elmaghba@oregonstate.eduAbstractRenderingtotheirabilitiestoprovideubiquitousconnectivity,exiblyandcos...

展开>> 收起<<

ProSky NEAT Meets NOMA-mmWave in the Sky of 6G Ahmed Benfaid Nadia Adem and Abdurrahman Elmaghbub.pdf

共7页,预览2页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

ProSky NEAT Meets NOMA-mmWave in the Sky of 6G Ahmed Benfaid Nadia Adem and Abdurrahman Elmaghbub

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: