Self-Supervised Learning for Enhancing Angular Resolution in Automotive MIMO Radars

A novel framework to enhance the angular resolution of automotive radars is proposed. An approach enlarge antenna aperture using artificial neural networks developed a self-supervised learning scheme. Data from high radar, i.e., radar with large aperture, used train deep network extrapolate element's response. Afterward, trained compact, low-cost radars. One million scenarios are simulated in Monte-Carlo fashion, varying number targets, their Radar Cross Section (RCS), and location evaluate method's performance. Finally, method tested real data collected outdoors commercial system. significant increase ability resolve targets demonstrated, which can translate more accurate faster responses planning decision making system vehicle.