Adaptive Jamming Decision-Making Against FHSS Communications via Inexpert Demonstrations Assisted Meta Reinforcement Learning

Abstract

Reinforcement learning (RL)’s powerful optimization capabilities have been extensively applied in the field of wireless communication jamming decision-making. However, the generalization of jamming policies has rarely been explored, and most existing studies rely on task-customized reward functions, which are often intractable to design. To address these issues, we propose a meta RL method for frequency-hopping spread spectrum (FHSS) jamming decision-making, aided by inexpert demonstrations. Firstly, the policy network is meta-trained with multiple diverse tasks to obtain initial network parameters with good generalization. Subsequently, we combine RL and behavioral cloning (BC) to extract useful information from demonstrations, along with learning rate adaptation to achieve efficient policy exploration without the task-customized jamming reward. Simulations confirm that our proposed method not only adapts to unseen jamming tasks with just a few fine-tuning steps under general binary rewards condition, but also achieves higher accumulated jamming rewards and results in lower normalized throughput for users, outperforming state-of-the-art methods.