MSEDOA: Enhancing DOA Estimation with Multiscale Squeeze-and-Excitation Networks for Automotive Millimeter-Wave Radar

Direction-of-Arrival (DoA) estimation is a critical task in automotive millimeter-wave radar systems, enabling the localization of targets and the implementation of essential safety functions. Traditional DoA methods struggle under low SNR and limited snapshot conditions. To address these challenges, we introduce MSEDOA, a novel deep learning method that processes raw IQ signals with a neural network featuring multiscale feature extraction, channel attention mechanisms, and Squeeze-and-Excitation Residual Networks. Experimental results demonstrate that the proposed model outperforms traditional DoA estimation methods, providing precise estimates in challenging environments.