Optimum Medium PRF Selection Using Reinforcement Learning

Determining the best medium pulse repetition frequency (PRF) set for radar has long been a problem. Because it cannot be represented in analytical form, it is not possible to solve it using conventional optimization methods. Although heuristic methods can provide a good PRF set, determining the best PRF set remains an open question. In this study, we propose a new method to find the best PRF set that can provide the smallest blind region in the range-Doppler domain. We applied a reinforcement learning algorithm to determine the PRF set. By applying the Deep Q-network (DQN) algorithm with hybrid rewards, we can find the PRF sets that yield the minimum blind region (maximum visible region). The proposed method can find the global optimum PRF sets with a high probability even if it begins from initial values close to the local optima.