Artificial Intelligence Applied to NOMA-UAV Networks for IoT Communication Systems

Abstract

The fast proliferation of Internet of Things (IoT) devices and the increasing de mand for reliable and energy-efficient wireless communication have motivated the explo ration of advanced access technologies and intelligent control mechanisms. Unmanned Aerial Vehicles (UAVs), with their flexible deployment and line-of-sight connectivity, have emerged as a key enabler to extend network coverage in hard-to-reach areas. In this context, this thesis examines how the combination of Non-Orthogonal Multiple Ac cess (NOMA), Deep Reinforcement Learning (DRL), and Simultaneously Transmit and Reflect Reconfigurable Intelligent Surfaces (STAR-RIS) can improve the effectiveness of UAV-assisted IoT communication networks. The main objective is to design an AI-based framework that jointly optimizes the UAV trajectory, user scheduling, and resource allocation to maximize energy efficiency and ensure fair resource distribution. A DRL-based algorithm is proposed to enable autonomous decision-making in dynamic and complex environments, using a weighted reward function that balances energy efficiency and fairness and makes it easier to pri oritize energy when needed or fairness when the situation requires it. Furthermore, a STAR-RIS-assisted architecture is introduced to enhance signal quality and EE by intelligently manipulating the wireless propagation environment. Simulation results demonstrate that the proposed approaches significantly outper form baseline methods in terms of energy efficiency, spectral utilization, and fairness. The proposed DRL-based solution, which optimizes all optimization factors simultane ously, achieves higher EE when compared to benchmark methods of around 60% and 17.3% for EE with different fairness weights prioritization and around 85% in fairness index.