Joint Antenna Selection and Beamforming in Integrated Automotive Radar Sensing-Communications with Quantized Double Phase Shifters

We consider an integrated sensing-communication system operating in a dynamic environment, such as an autonomous vehicle scenario. We propose a novel, low-cost, low power consumption and low-computation approach for designing a beam that can simultaneously reach the radar target of interest and the desired communication destination. The transmitter is a uniform linear array, equipped with quantized double phase shifters, which enables a flexible beam design while using analog only processing. Only a small number of antennas are selected to transmit in each channel use, in order to save system power and reduce antenna coupling. We propose a deep reinforcement learning approach to adaptively adjust the double phase shifters and select the active antennas in order to optimize the transmit beamforming, through a transmission and feedback trail. The actor-critic network strategy together with the Wolpertinger policy is adopted to obtain the optimal solutions efficiently and effectively. Numerical results demonstrate the feasibility of the proposed method.