A Multipath Signals Recognition Method based on Deep Neural Network

Millimeter-wave radar is the primary sensor for enabling autonomous driving and ADAS functions. MIMO technology has demonstrated its effectiveness to deliver precise angular estimation of objects using few antennas, making it a popular choice in automotive radar applications. However, a significant challenge for MIMO systems is multipath reflection. This phenomenon not only degrades angular estimation accuracy by disrupting the integrity of the virtual arrays in MIMO radar, but also impairs detection performance by introducing ghost targets. This paper addresses the issue of ghost target identification in multipath scenarios for vehicle-mounted MIMO radar. A method called DNN-IAA is proposed, which combines Iterative Adaptive Algorithm based on separation of sending and receiving (srt-IAA) with the Deep Neural Network (DNN) to tackle multipath signals recognition in complex scenarios. Firstly, srt-IAA is employed to obtain the two-dimensional angle-frequency spectrum of the target’s Direction of Arrival (DOA) and Direction of Departure (DOD). Subsequently, the DNN is used to extract features from the DOA-DOD spectrum for the classification of multipath signals. Finally, a simulation is conducted to validate the proposed solution.