Phased-Array-based Real-Time 3D Radar for AI-Based Event Classification

This presentation first discusses approaches for leveraging mm-wave phased arrays to create 3D radar systems with fast (100s of frames per second) scanning capabilities. Effective system vertical integration from antennas to compute is a key aspect in these systems in order to implement an effective and fast orchestration between beam steering and radar waveform acquisition/processing steps. Examples of such systems will be presented based on 94GHz and 60GHz phased arrays including a real-time imaging demonstration. Next, a novel approach for AI-driven perception is described, where a custom-designed DNN is used to extract temporal and volumetric features from the raw output of the fast-scanning 3D radar system. Moreover, it will be described how such a system is amenable to be augmented with other sensors to enable coherent multi-spectral sensing. Example results will be presented for concealed object detection and gesture classification for augmented reality systems.