A DRL-Based Design for IRS-assisted NOMA Networks

In this paper, the performance of IRS assisted NOMA networks is investigated using Deep Reinforcement Learning (DRL) based approach. It considers the heterogeneous nature of users in dense urban environments, including the impact of imperfect Successive Interference Cancellation (SIC) from users. Additionally, it addresses the placement of IRS units within the environment, accounting for potential interference between IRS units and the line of sight from the base station. The paper aims to maximize the downlink sum rate of communication for all users while satisfying specific Quality of Service (QoS) parameters. The problem involves optimizing the power allocation of signals as per the NOMA scheme and phase shifts of IRS elements. Considering low latency, we have used a DRL-based approach to obtain an efficient solution compared to traditional methods. The paper further utilizes the design to determine optimum deployment strategies for each IRS unit to maximize the sum rate.