Optimal Jamming using Delayed Learning

Recent advances in cognitive radios for electronic warfare create the potential for dynamic environmental conditions, which makes it difficult to rely upon predict-then-adapt approaches in unfamiliar environments. It is thus imperative that radios have increasingly intelligent capabilities in order to be effective in harsh unknown surroundings. In this paper, we explore whether an intelligent jammer can learn and adapt to its surroundings in an electronic warfare-type scenario. We address this problem from a reinforcement learning perspective where the jammer has delayed information regarding the packets exchanged between a victim transmitter and the receiver. This is different from the traditional assumption that feedback is available instantaneously in reinforcement learning-based algorithms. A new framework, to enable delayed learning in scenarios where rewards are associated with state transitions rather than the states themselves is developed. The benefits of such a framework are shown by studying the optimal jamming strategies against an 802:11-type wireless network that uses the RTS-CTS protocol to communicate and deliver information.