Sum-Rate Maximization of RSMA-Based STAR-RIS-Mounted UAV Communications With Proximal Policy Optimization

Abstract

This paper investigates joint sum-rate maximization in a downlink multi-user multiple-input single-output (MISO) system in which rate-splitting multiple access (RSMA) transmission is assisted by a simultaneously transmitting and reflecting reconfigurable intelligent surface (STAR-RIS) mounted on an unmanned aerial vehicle (UAV). The UAV position, STAR-RIS transmission and reflection coefficients, phase configurations, and beamforming vectors are strongly coupled, making the resulting problem non-convex, high-dimensional, and time varying. Conventional optimization methods may therefore struggle to provide computationally efficient decisions in real time. To address this challenge, the original problem is reformulated as a Markov decision process (MDP) and solved using a deep reinforcement learning (DRL) framework based on proximal policy optimization (PPO). By interacting with the environment, the agent learns a control policy that adapts the UAV trajectory, STAR-RIS configuration, RSMA power allocation, and common-rate allocation to the instantaneous channel state and user positions. A structured action-space design further reduces the training burden: maximum-ratio transmission (MRT) determines the common-stream beam direction, regularized zero forcing (RZF) determines the private-stream beam directions, and a hierarchical power-mapping strategy enforces the transmit-power constraint. Simulation results demonstrate that the proposed framework improves the system sum rate and remains robust under highly correlated channels and stringent quality-of-service (QoS) constraints.