Research on Optimization of RIS-Assisted Air-Ground Communication System Based on Reinforcement Learning

In urban emergency communication scenarios, building obstructions can reduce the performance of base station (BS) communication networks. To address such issues, this paper proposes an air-ground wireless network enabled by an unmanned aerial vehicle (UAV) and assisted by reconfigurable intelligent surfaces (RIS). This system enhances the efficacy of UAV-enabled MISO networks. Treating the UAV as an intelligent agent moving in 3D space, sensing changes in the channel environment, and adopting zero-forcing (ZF) precoding to eliminate interference from ground users. Meanwhile, joint design is performed for UAV movement, RIS phase shifts, and power allocation for users. We propose two deep reinforcement learning (DRL) algorithms, which are termed D3QN-WF and DDQN-WF, respectively. Simulation results indicate that D3QN-WF achieves a 15.9% higher sum rate and 50.1% greater throughput than the DDQN-WF baseline, while also demonstrating significantly faster convergence.