Predictive and Adaptive Semantic Communication for QoE Optimization in Multi-UAV Networks

Abstract

Deep learning-based semantic communication has demonstrated superior efficiency in wireless image transmission. However, traditional reactive schemes often suffer from outdated Channel State Information (CSI) in highly dynamic multi-UAV environments, leading to severe latency and utility degradation. To address this challenge, we propose the Predictive and Adaptive Semantic Communication (PASC) framework. PASC integrates a GRU-based predictor to anticipate channel evolution, enabling the proactive adjustment of compression rates by dynamically calibrating attention-weight thresholds. Furthermore, a deadline-aware deep reinforcement learning (DRL) algorithm is proposed to jointly assign sub-channels, allocate bandwidth, and adjust power based on predictive states, thereby preventing resource monopolization. Simulation results confirm that PASC achieves a 58.8% improvement in average Quality of Experience (QoE) compared to non-predictive baselines in low-SNR regimes. Crucially, the framework demonstrates formidable robustness to imperfect CSI, strictly bounding end-to-end latency and maintaining high semantic fidelity even in the presence of extreme prediction noise.