Design of Anti-Jamming Waveforms for Cognitive Radar Based on Deep Reinforcement Learning

In the face of various complex interferences encountered by radar systems in modern electronic warfare, the performance of radar is greatly affected. Therefore, intelligent and adaptive radar anti-jamming waveform design methods have become crucial. This paper proposes a training algorithm for anti-jamming strategies under multiple types of interference based on deep reinforcement learning. Considering that radar jamming resistance possesses Markovian properties, meaning that the current anti-jamming strategy of the radar is only related to the current momentary state and independent of previous states, a reinforcement learning algorithm is introduced to construct the Markov Decision Process (MDP) framework. Due to the complex and continuous nature of the state-action space, neural networks are introduced. By quantifying anti-jamming decisions through the construction of state sets, action sets, and reward functions, and modeling the anti-jamming process as a Markov decision process, we propose a radar anti-multiple interference algorithm based on the Soft Actor-Critic (SAC) algorithm, called RAMI-SAC, to optimize strategies. Specifically, considering the synchronization between radar signals and interference signals, we utilize the characteristics of radar signals and interference signals from the previous time step to construct the state space. We use the RAMI-SAC algorithm to generate anti-jamming strategies, including the masking pulse width, frequency band, and signal power of the current radar sequence. Through multiple sets of simulations, the convergence of the algorithm is demonstrated, resulting in anti-jamming strategies that can withstand attacks from multiple types of interference, maintaining a certain level of detection capability even in the presence of multiple types of interference.