Optimizing Radio Resources for Radar Services in ISAC Systems by Deep Reinforcement Learning

In integrated sensing and communication (ISAC) systems, the radar and communication functionality share the same infrastructure and radio resources. The flexible access scheme in mobile communication systems allows for variable and efficient radio resource allocation for multiple users and services. In this paper, we present a resource allocation strategy for Orthogonal Frequency Division Multiple (OFDM) based radar sensing. Furthermore, we propose a Deep Reinforcement Learning (DRL) method combined with classic OFDM radar signal processing techniques to optimize the radio resources for radar sensing. Two agents are trained with the DRL method on simulated data to predict the radio resources for the subsequent signal, aiming to achieve the required radar performance while improving the resource efficiency. One agent optimizes the transmission power, while the other optimizes the signal bandwidth/duration. The investigated scenario is a highway with traffic. We evaluate the performance of the agents based on detection loss and radio resource efficiency the metrics. Further, we compare the results against random and maximum resource-selecting agents.