Reinforcement Learning for Cognitive Integrated Communication and Sensing Systems

In this paper, we propose a cognitive Massive MIMO integrated communication and sensing (ICAS) system that integrates both functionalities, enabling efficient use of the congested spectrum. To achieve this, we introduce a reinforcement learning (RL) approach that involves adaptability and that is able to optimize a joint waveform for the aforementioned system to achieve multiple objectives. We demonstrate that cognitive RL can improve state-of-the-art techniques that aims at designing the joint waveform from the ground-up achieving sensing and communication trade-off. Our results show that cognitive RL can greatly enhance sensing performance without compromising the communication performance. In contrast to previous works, we assume no prior information on the sensed scene such as the number of targets or the statistics of the disturbance.