Using Deep Reinforcement Learning to Enhance Channel Sampling Patterns in Integrated Sensing and Communication

In Integrated Sensing and Communication (ISAC) systems, estimating the micro-Doppler (mD) spectrogram of a target requires combining channel estimates retrieved from communication with ad-hoc sensing packets, which cope with the sparsity of the communication traffic. Hence, the mD quality depends on the transmission strategy of the sensing packets, which is still a challenging problem with no known solutions. In this letter, we design a deep Reinforcement Learning (RL) framework that fragments such a problem into a sequence of simpler decisions and takes advantage of the mD temporal evolution for maximizing the reconstruction performance. Our method is the first that learns sampling patterns to directly optimize the mD quality, enabling the adaptation of ISAC systems to variable communication traffic. We validate the proposed approach on a dataset of real channel measurements, reaching up to 40% higher mD reconstruction accuracy and several times lower computational complexity than state-of-the-art methods.