Weak Target Detection with Integrated Multi-user Communications via Reinforcement Learning

Although being investigated extensively in the literature, integrated sensing and communications (ISAC) still encounters challenges in simultaneous weak target detection and multi-user communications, particularly in harsh environments with strong clutter. To address this issue, we introduce a reinforcement learning (RL) based ISAC design, which applies a fully data-driven “SARSA” training framework to adaptively optimize the beam steering. The RL algorithm facilitates a balanced adjustment of beamforming and power allocation, enabling a swift adaptation to the unknown dynamic environment. Additionally, we introduce several communication embedding strategies using beampattern modulation to enable the parallel multi-user communications.