Joint Resource and Trajectory Optimization for UAV-Assisted Semantic Communication
The high flexibility of Unmanned Aerial Vehicles (UAVs) enables effective network coverage in remote areas where traditional wireless networks are unavailable. This article proposes a UAV-assisted semantic communication (SC) framework where UAVs act as aerial relays to facilitate efficient information collection from ground devices to remote centers. Unlike conventional approaches, we introduce a comprehensive semantic transmission efficiency metric that integrates semantic similarity with the semantic age of information (AoI), capturing both the quality and timeliness of transmitted information under varying channel conditions. We formulate a joint optimization problem to maximize semantic transmission efficiency by optimizing device association, UAV trajectory, and power allocation while accounting for channel signal-to-noise ratio (SNR) constraints. To solve this problem, we propose a hierarchical solution framework, where device association is optimized at the upper level, followed by joint UAV trajectory planning and power allocation at the lower level. For device association, we develop a cluster-match algorithm that employs a $K$ -means constrained clustering approach followed by greedy matching. For trajectory and power optimization, we propose a diffusion-twin delayed deep deterministic policy gradient (Diffusion-TD3) algorithm, which leverages a diffusion model as the actor network to generate control policies. Simulation results show that the proposed cluster-match algorithm substantially improves UAV flight stability and service reliability compared to conventional greedy matching schemes. Moreover, the Diffusion-TD3 algorithm exhibits superior exploration performance and convergence characteristics relative to traditional deep neural network methods.