Radar Waveform Parameter Optimization with Offline Reinforcement Learning Strategy

designing the radar waveform parameter is crucial for target detection. However, these parameters are usually adjusted manfully, and thus, while adjusting these parameters, not all factors are taken into consideration. Consequently, the optimality cannot be ensured. In this work, we proposed a CMRDRL (convolutional mixed reward deep reinforcement learning) model to optimize the parameters, which can optimize the strategy by interacting with the environment and taking many environmental factors into consideration. Firstly, a convolutional neural network model is designed to provide an initial strategy model, which can ensure a basic parameter setting performance by considering numerous environmental variables. Then considering the training cost, a mixed reward and an offline reinforcement learning model is designed to optimize the parameter setting strategy by interact with the historical data. Experiment shows that the proposed strategy can provide an effective way to set the waveform parameters during the radar operation.