WEAK TARGET DETECTION IN MASSIVE MIMO RADAR VIA AN IMPROVED REINFORCEMENT LEARNING APPROACH

Massive multi-input-multi-output (MMIMO) cognitive radar can enhance the target detection ability in a dynamic environment via a continuous ?perception-action? cycle. In our previous work, we proposed a reinforcement learning (RL) based approach for multi-target detection in MMIMO. However, this method shows poor detection performance for weak targets attributed to its imperfect action and reward mechanisms. In this paper, we propose an improved RL based method to enhance the detection probability of weak targets. In the action stage, the transmit power is divided into omni-directional and directional components, the former significantly reduces the missed detection probability of weak targets and the latter improves the detection probability by focusing more power on weak targets. Moreover, the reward mechanism of RL is modified to further improve the detection performance. In addition, the transmit weight matrix is designed by an optimum combination of the beampatterns of all unit orthogonal transmit waveforms, thus greatly reducing the computational complexity. Simulation results are provided to demonstrate the effectiveness of the improved RL based method for weak target detection.