Deep Q-Network based Anti-Jamming Strategy Design for Frequency Agile Radar

摘要

In this paper, a deep Q-network (DQN) based strategy design method for frequency agile (FA) radar is proposed, in which the FA radar is regarded as the agent in reinforcement learning (RL) and learns how to take actions in the presence of a spot jammer. Due to the existence of the spot jammer, the agent must alter its carrier frequency frequently to avoid being jammed. To measure the performance of the agent with varied carrier frequencies in a coherent processing interval (CPI), the detection probability is derived and regarded as the reward signal in RL. By applying a DQN algorithm, an optimal strategy can be learned guiding the agent how to choose the carrier frequency at every pulse. The learned strategy enables the agent not only to avoid being jammed but also to have high detection probability. Simulation results illustrate the effectiveness of the proposed method.