Intelligent Suppression of Interferences Based on Retroactive-DQN

This paper aims to investigate intelligent interference suppression strategies through reinforcement learning. Particularly, the Retroactive-Deep Q Network (R-DQN) is devised as the principle of updating suppression strategies by introducing a backtracking Q value, which links the evaluations in each step of training. At the analysis stage, three interferences, including the Interrupted Sampling Repeater Jamming (ISRJ), the SMeared SPectrugram (SMSP) jamming and repeater jamming, and five suppression methods are considered. Moreover, the selection probabilities of the implementation sequence for interference suppression between the R-DQN and the conventional DQN are compared. Numerical results corroborate the effectiveness and robustness of the considered suppression strategies.