Enabling Super-Resolution for Automotive Imaging Radars in the Presence of Antenna Calibration Errors

Abstract

Radar is an essential technology for Advanced Driving Assistance Systems (ADAS), used to accurately localize objects even in unfavorable weather conditions. Most radar systems that are now being produced for ADAS provide either 3D or 4D point clouds, containing range, Doppler, azimuth, and elevation information for every detected point target. Out of all dimensions, the azimuth and elevation are estimated using more advanced algorithms than the ones generally used for range and Doppler. This is due to the restricted size of the aperture that can be safely mounted on a vehicle, hence the resolution must be enhanced digitally. When using advanced algorithms challenges such as precise antenna manufacturing are of significant importance, to avoid phase and gain mismatch between the antenna elements along with their inherent coupling. These negative effects lead to a significant degradation in the Direction of Arrival estimation. Super-resolution techniques such as MUSIC and CAPON are widely referenced, however their performance throughout prior work is evaluated in ideal environments and generally with multiple available data acquisition snapshots. In this paper we address the issues faced when applying such algorithms in a radar application and offer a solution based on linear prediction and spatial smoothing to enhance the performance of such algorithms.