Reinforcement Learning for Reactive Jamming Mitigation

In this paper, we propose a strategy to avoid or mitigate reactive forms of jamming using a reinforcement learning approach. The mitigation strategy focuses on finding an effective channel hopping and idling pattern to maximize link throughput. Thus, the strategy is well-suited for frequency-hopping spread spectrum systems, and best performs in tandem with a channel selection algorithm. By using a learning approach, there is no need to pre-program a radio with specific anti-jam strategies and the problem of having to classify jammers is avoided. Instead the specific anti-jam strategy is learned in real time and in the presence of the jammer.