Deep Learning Based Hybrid Beamforming for mmWave Dual-Functional Radar-Communication

We propose a novel deep learning (DL) based HBF design for the dual-functional radar-communication (DFRC) system with the millimeter wave (mmWave) massive multiple-in-multiple-output (MIMO) architecture, in which the HBF is formulated as a non-convex optimization problem. First, the DL-based HBF is designed to minimize the sum-MSE of downlink communications while carrying out necessary radar sensing concurrently. Then the synchronization noise is attached to the input channel data to enhance the robustness of the CNN. After that, an attention mechanism is added into the prediction stage to improve the prediction without affecting the accuracy of the prediction results. Finally, the numerical simulation results show significant tradeoff performance improvements between communication and radar sensing can be obtained over existing HBF designs.