Deep RL-based Resource Allocation for User Fairness in STAR-RIS–assisted NOMA-enabled B5G Networks

Simultaneous transmitting and reflecting reconfigurable intelligent surfaces (STAR-RIS) have come forward as propitious auxiliary devices to enhance the connectivity of the wireless network by providing 360° coverage. While STAR-RISs improve energy efficiency and coverage of wireless networks, the spectrum can be a bottleneck for dense urban deployments. Non-orthogonal multiple access (NOMA) is a modern technique which serves multiple users using shared spectrum resources, hence improving the spectrum efficiency. This work utilizes deep reinforcement learning (DRL) to investigate user fairness in a multi-user (MU) STAR-RIS–aided NOMA-enabled downlink communication system under imperfect channel state information (CSI). The DRL-based allocation architecture jointly optimizes the reflection coefficients of STAR-RIS and power allocation coefficients of NOMA clusters. Simulation results show that the proposed solution provides promising performance gains in terms of sum rate, spectral efficiency, and fairness among users, and is a viable resource allocation scheme for ubiquitous services, particularly in Internet of Things (IoT) networks and dense urban environments, where the combination of limited spectrum and high user density necessitates efficient resource allocation to ensure fairness and maintain quality of service (QoS).