Research On Deep Reinforcement Learning Based Cognitive Radar Waveform Selection

Cognitive radar can adapt environment better than traditional radar.Optimizing the transmitted waveform in real time by sensing the environment is an important way to realize adaption.By using more and better waveform optimization methods,radar can get better detection effect.This paper mainly studies the combination of deep reinforcement learning and cognitive radar waveform selection and applies the algorithm of deep reinforcement learning(DRL)to the detection of cognitive radar,which promotes the intelligence of cognitive radar waveform selection.This paper designs a DRL-based cognitive radar waveform selection(optimization)framework.The traditional reinforcement learning algorithm cannot be well applied to the perception and representation of the environment.By using DRL to adapt to the constraints of the complex and changeable electromagnetic and geographical environment of radar tasks,this paper proposes two waveform selection algorithms based on deep Q network(DQN)and double deep Q network(DDQN)in target tracking task.First,we give the specific elements needed by DRL algorithms,use entropy state to represent the environment and target state,establish an optional waveform library,and determine the reward and punishment function according to the effect of tracking.Then,we build the depth neural network according to the algorithm and train the network by repeating tracking process in which we can get the corresponding detection results and reward and punishment.Finally,the end-to-end waveform adaptive selection can be realized by applying the trained network to the tracking process.Simulation results show the effectiveness and feasibility of the proposed framework and algorithm.Furthermore,this paper proposes a multistatic cognitive radar waveform selection(optimization)method framework based on multi-agent depth reinforcement learning(MDRL).The cooperative waveform selection and optimization of multistatic radar is realized by using MDRL which integrates multi-agent reinforcement learning(MARL)and DRL.The multistatic cognitive radar waveform selection method based on independent deep Q network(IDQN)is implemented in the stealth target tracking task.By given the same reward and punishment,the independent neural network can learn to cooperate in the training process,and finally realize the joint waveform selection and optimization for multistatic cognitive radar.The simulation results show that the proposed method has better tracking performance compared with the multistatic cognitive radar with fixed waveform,which proves the feasibility of the proposed method.