Reinforcement Learning for Jamming Games Over AWGN Channels With Mobile Players

We study a jamming problem in a wireless scenario, where a legitimate receiver and a jammer compete in a zero-sum game with the value to maximize/minimize being the channel capacity at the receiver's side. Classic approaches consider stationary nodes. Instead, we investigate what happens when they can move along a linear geometry, which requires expanding the game into a dynamic setup. Even though the strategy space becomes huge, we can (i) provide some analytical criteria to identify good strategies, and (ii) explore efficient strategies leading to equilibrium outcomes using reinforcement learning techniques. We analyze different scenarios of sequential and simultaneous dynamic moves, as well as perfect versus imperfect information about the position of the adversary. Our numerical evaluations show the consistency of our findings in all the considered scenarios.