Real-time DoA Estimation for Automotive Radar

For automotive radar, direction-of-arrival (DoA) estimation is the most challenging component in the target detection problem. To cope highly dynamic driving conditions and to achieve full autonomy, there are stringent requirements on the processing time and DoA estimation resolution. None of the state-of-the-art methods can accomplish both at the same time: FFT-based algorithm is computationally fast but cannot provide high resolution, while subspace-based algorithms such as MUSIC and ESPRIT can achieve super-resolution but cannot meet the timing requirement. In this paper, we present MARS - a real-time super-resolution algorithm based on maximum likelihood (ML) estimation. In contrast to traditional ML estimation, MARS exploits the intrinsic correlation between the input data of adjacent time slots to reduce the search space. To further reduce computation time, MARS decomposes the problems in each step into independent sub-problems that can be efficiently executed on GPU parallel computing platform. Simulation experiments show that MARS can achieve 1° super-resolution in DoA estimation under 1 ms.