Federated DDQN-Based Frequency Hopping Strategy for FANETs in Complex Jamming Environments

In the realm of flying ad-hoc network(FANETs), unmanned aerial vehicles (UAVs) frequently encounter challenges related to adaptability in complex jamming environments. To address this issue, our research introduces an innovative frequency hopping(FH) strategy that combines federated learning(FL) with double deep Q-Networks (DDQN). The UAVs adaptively select frequency hopping patterns by a locally tuned DDQN model, which they then share with a multi-access edge computing (MEC) server to facilitate global aggregation optimization. The study also introduces an difficulty-adaptive training mechanism to optimize the allocation of computational resources. Experimental validations of our method confirm its efficacy in significantly improving both the adaptability and overall performance of FANETs in challenging jamming conditions, demonstrating robust convergence and high accuracy.