Demo Proposal For Fast-Learning Sparse Antenna Array For Automotive Radar

In this demo, we present fast-sparse array hardware built for tracking both slow- and fast-moving targets in an automotive application. Specifically, the direction of arrivals (DOA) of the moving targets are estimated at regular time-intervals by applying sparse arrays that are adaptively configured by using deep learning. High-resolution DOA estimation requires a large number of antenna elements which results in a high cost. To reduce cost and power while obtaining sufficient performance, sparse array structures are used where a subarray of the whole antenna array is utilized. Recently, we proposed a deep learning-based (DL) approach for antenna selection that has low computational complexity and optimum solution [1, 2]. In these works, the data labeling method is based on an exhaustive search where a best sparse array is selected among all possible configurations that minimize a predefined cost function. In addition, the proposed techniques assume that the targets are moving slowly between the scans. In this demo, we address these shortcomings and demonstrate a computationally efficient labeling technique and a modified sparse array configuration strategy that can track both slow- and fast-moving targets. The results are particularly useful for automotive applications, where the radars are required to track slowly moving objects such as pedestrians and fast-moving vehicles for efficient navigation. Our demonstration platform consists of a control unit implementing over-the-air dedicated antenna array of 16 elements capable of receiving signals in the ISM band (2.4 / 5 GHZ). By using a dedicated GUI, our demo will illustrate the DOA with sparse array of the receive antenna. References: [1] A. M. Elbir, K. V. Mishra, Y. C. Eldar, “Cognitive Radar Antenna Selection via Deep Learning”, IET Radar, Sonar & Navigation, vol. 13, issue 6, pp. 871-880, June 2019.