Resource optimization in semantic and bit user coexistence networks with STAR-RIS assistance

In this paper, we investigate an uplink rate-splitting multiple access (RSMA) system assisted by a simultaneously transmitting and reflecting reconfigurable intelligent surface (STAR-RIS), where semantic users coexist with conventional bit users. In the considered scenario, the direct links between the users and a single-antenna access point (AP) are often obstructed. By exploiting the transmission and reflection capabilities of the STAR-RIS, additional non-line-of-sight paths are established, which provide strong penetration through obstacles and restore reliable connectivity for both semantic and bit users. On this basis, we formulate a joint optimization problem to enhance the system performance, incorporating the users’ transmit powers, the bandwidth allocation between semantic and bit users, and the configuration of the STAR-RIS elements. To tackle this challenging optimization task, we develop a deep reinforcement learning approach based on the proximal policy optimization (PPO) algorithm. Simulation results demonstrate that the proposed PPO-based algorithm exhibits stable convergence and achieves approximately 63.83% and 62.31% performance gains over the DDPG-based benchmark in two representative scenarios. Furthermore, they show that employing rate-splitting multiple access (RSMA) instead of non-orthogonal multiple access (NOMA), as well as deploying a simultaneous transmitting and reflecting reconfigurable intelligent surface (STAR-RIS) rather than a conventional RIS, yields a higher sum rate for the served users.