Resource optimization for RIS-UAV integrated air-ground cooperative semantic communication driven by deep reinforcement learning

Abstract

To meet the 6G semantic transmission requirements, this paper proposes an air-ground dual-mode semantic transmission architecture for Reconfigurable Intelligent Surface (RIS)-aided Unmanned Aerial Vehicle (UAV) networks. By dynamically assigning RIS-reflected links to users with long packets and direct UAV line-of-sight (LOS) links to users with short packets, we jointly optimize semantic symbols, RIS phase shifts, UAV three-dimensional (3D) trajectory, and transmit power to maximize the system-level semantic Quality of Experience (QoE). To efficiently solve the formulated non-convex mixed-integer optimization problem, we present an Action Space Decoupled Hybrid Deep Reinforcement Learning (ASD-HDRL) framework, which employs the Dueling Double Deep Q-Network (D3QN) for discrete semantic resource decision-making and Deep Deterministic Policy Gradient (DDPG) for continuous control optimization. Simulation results verify that the proposed ASD-HDRL achieves superior system-level semantic QoE performance.