A Real-Time Super-Resolution DoA Estimation Algorithm for Automotive Radar Sensor

Direction-of-arrival (DoA) estimation is critical to obtaining precise information of targets for automotive radar sensors. To obtain fine radar imaging, super-resolution DoA estimation is needed to distinguish adjacent targets. Moreover, to adapt to the rapidly changing driving dynamics, the processing time for DoA estimation must meet very stringent timing requirement. Unfortunately, none of the existing methods can achieve both super-resolution and real-time requirement at the same time. In this article, we present the maximum-likelihood-based real-time super-resolution (MARS)-a real-time super-resolution DoA estimation algorithm. The main idea in MARS is to use maximum likelihood estimation (MLE) as the objective function. Unlike traditional MLE, MARS exploits the intrinsic correlation between the input data of adjacent time slots to substantially reduce the search space. Moreover, instead of using an exhaustive search to find the DoA solution, we employ the compressed sensing algorithm orthogonal matching pursuit (OMP) to efficiently discover an optimal or near-optimal solution for the MLE objective function based on the reduced search space. To further accelerate computation time, MARS decomposes the problems in each step into independent subproblems that can be efficiently executed on a GPU parallel computing platform. Simulation experiments show that MARS can achieve super-resolution in DoA estimation under 1 ms. Compared to state-of-the-art algorithms such as multiple signal classification (MUSIC) and estimation of signal parameters via rotational invariance techniques (ESPRIT), MARS outperforms both of them in DoA estimation while being the only known algorithm that can meet the stringent real-time requirement. Hardware experiments further illustrate that MARS outperforms state-of-the-art algorithms in target detection by achieving superior resolution.