Deep Counterfactual Regret Minimization Algorithm with Regret Discount in Radar Anti-Jamming Game.

Aiming at the countermeasures between radar and intelligent jammer, in this paper, an optimal strategy for radar anti-jamming game is achieved by deep counterfactual regret minimization (CFR) with regret discount, where the frequency agility radar can change its carrier frequency based on its strategies to counterwork the specific interference. At first, the non-cooperative game between radar and jammer is modeled in the form of game tree in the frame of extensive-form game, which expects to find Nash equilibrium strategy through multiple rounds of interaction. Then, to reduces the amount of calculation from the increasing of radar pulses, redar pulses, deep CFR is proposed to utilize the deep neural network to approximate the traditional CFR. Finally, through reducing the impact of past decisions by discounting regret values, deep CFR with regret discount is further considered to accelerate the convergence speed of the algorithm. Experimental results verify the effectiveness of the proposed algorithm in finding the optimal strategy and approximated Nash equilibrium.