Efficient Gridless DOA Estimation for Nonuniformly Spaced Linear Arrays in Automotive Radar Sensors

Automotive millimeter-wave (MMW) radar sensor stands out for its weather-independent capability, serving as an indispensable tool in advanced driver assistance systems (ADAS). Accurate and efficient direction-of-arrival (DOA) estimation is crucial for automotive tasks such as object detection, tracking, and collision avoidance. While recent popular single snapshot DOA estimation algorithms based on compressed sensing (CS) offer super-resolution capabilities, they are highly sensitive to mismatches between assumed and actual sparsity bases, known as the off-grid effect. To address this issue, atomic norm minimization (ANM), a promising gridless sparse recovery algorithm under the Toeplitz model, has gained prominence in DOA estimation for uniform and sparse linear arrays (SLAs). However, cost and hardware limitations of automotive radar sensors often result in nonuniform array structures, obstructing the application of ANM-based gridless DOA estimation methods. This article presents a novel gridless DOA estimation method for nonuniform linear arrays (NLAs) based on ANM with efficiency. It achieves this by transforming NLA manifolds into Vandermonde vectors corresponding to virtual uniformly spaced arrays, and employing accelerated proximal gradient (APG) techniques while leveraging the Vandermonde structure and low-rank property. This approach ensures the applicability of ANM-based algorithms to NLA without sacrificing efficiency. Simulation and measurement experiments on automotive radar sensors demonstrate its superiority.