Compressed sensing based high-resolution DoA estimation for automotive radar with real measurements

Abstract

Radar sensors are one of the key elements in autonomous driving systems. The performance of radar sensors is robust independently of weather and visibility conditions, and can accurately detect targets positioned over a hundred meters away from the sensor. In vehicles, radar sensors are often required to be able to position the detected targets in a 2D grid, by measuring the distance and the angular position of the object with respect to the sensor, as well as the radial speed of the targets. While the accurate measuring of range and radial speed of the targets can easily be achieved with the commonly used modulations and frequency bands assigned to automotive radar, the angular positioning of the targets, or direction of arrival (DOA) estimation, still requires improvement for complex autonomous driving systems. The goals of this project are two. First, we aim to improve the performance in DOA estimation by applying compressed sensing (CS) based algorithms instead of the traditional FFT approach. The FFT approach needs uniform arrays, requiring a high amount of elements as the aperture of the sensor widens. In contrast, the CS based approach allows for the use of non-uniform sparse arrays, allowing for the design of arrays with a big aperture but a reduced amount of elements. The second goal of this project is to develop a user environment for the operation of our radar testing device, the Radarbook. With this goal in mind, two UIs are created: one enabling the user to take and store data from measurements for future processing, as well as processing of previously acquired data; and a different one for real time measurement and processing, with the capability of applying complex adaptive algorithms for DOA estimation. Finally, both goals merge as the validation of this approach is tackled with real data acquired with the Radarbook via our previously developed user interfaces