Spectrum-based Single-Snapshot Super-Resolution Direction-of-Arrival Estimation using Deep Learning

A method for multi-target direction-of-arrival estimation for commercial FMCW radar systems in the automotive domain is presented. The proposed approach realizes single-snapshot direction-of-arrival estimation utilizing an artificial neural network. The network is trained on synthetic data to predict the spatial spectrum directly from the received signal vector. Traditional peak detection is then used to extract the specific target angles as well as the number of targets from the spatial spectrum. The model is validated on real-world measurement data and its performance is compared to traditional spatial-spectrum based direction-of-arrival estimators. Our findings indicate super-resolution like performance while significantly reducing and simultaneously limiting computation time compared to a maximum-likelihood search.