Optimal Joint Trajectory and Caching with Graph-based Federated Knowledge Distillation in UAV-assisted Networks

The use of unmanned aerial vehicles (UAVs) as mobile cache nodes can enhance data traffic throughput and reduce user latency in urban areas. However, jointly optimizing UAV trajectories for efficient content delivery and onboard caching strategies presents significant challenges stemming from the tight coupling between these two problems, as well as issues related to the non-independent and non-identically distributed (non-IID) nature of real-world content popularity and wireless channel dynamics. To address these challenges, we propose KDLMGCN, a deep learning framework that employs graph convolutional networks with LSTM to capture intricate relationships between UAV locations and ground user content demand patterns. Through empirical evaluation, we demonstrate that KDLMGCN surpasses benchmark methods in terms of convergence, cache hit rates, and average transmission, with real-world trajectory dataset and Zipf-based content popularity data.