DRL-Based Approach for RIS-Aided NOMA System in Short Packet Communications

This paper investigates the power allocation problem in a reconfigurable intelligent surface (RIS)-aided non-orthogonal multiple access (NOMA) system tailored for short packet communications (SPC) in beyond 5G (B5G) and 6G networks. We propose a deep reinforcement learning (DRL)-based approach utilizing the proximal policy optimization (PPO) algorithm to maximize the sum achievable data rate by optimizing transmission power and power allocation coefficients, while adhering to power constraints and ensuring low-latency requirements of SPC. The RIS enhances signal quality by dynamically adjusting the wireless propagation environment, while NOMA enables efficient spectrum sharing for multiple users. However, the nonconvex nature of the optimization problem, coupled with the dynamic channel conditions of SPC, poses significant challenges for traditional methods. The proposed DRL-based solution leverages PPO's sample efficiency and stability to adaptively handle these dynamics, offering a computationally efficient framework that reduces training data and time requirements compared to conventional optimization techniques.