A Frequency-Hopping Game Between a Smart Jammer and a Cognitive Radio

A smart jammer can determine the frequency-hopping patterns of a cognitive radio using machine learning. This paper develops a two-state stochastic game where the receiver and jammer choose at each stage whether to stay on their current frequency or hop to a new one. The cost of hopping to a new frequency is also modeled with a discounted payoff. Machine learning algorithms are implemented to update the policies for both players in order to converge to a stationary Nash equilibrium. The game is simulated using Nash Q-learning and WoLF -PHC, as well as the computed single-stage Nash equilibrium for each state. Of the strategy options included for both players, there are ideal strategy-updating algorithm pairs depending on the values used for the discounted payoffs.