Long-Term Fairness Scheduling in Downlink RSMA Systems with SAC-Based Beamforming Design

In this paper, we consider the long-term user fairness of rate-splitting multiple access (RSMA)-assisted downlink multiple-input single-output (MISO) system. The aim is to maximize the system’s proportional fairness scheduling (PFS) utility over a long period by optimizing user scheduling, beamforming, and public rate splitting. To effectively solve this challenging time-varying mixed-integer non-convex optimization problem, we first transform the long-term overload system into a series of short-term underload systems by user clustering and time-division scheduling. Then, to deal with strong interference between partial users within short-term, we superpose public stream beams for these users. Finally, we design beamforming within multiple time slots by utilizing maximum entropy reinforcement learning-based soft actor-critic (SAC) algorithm. Simulation results show that the proposed scheme achieves better performance in comparison with other user scheduling schemes.