High-resolution Imaging with Compact Millimeter Wave Radars

Millimeter wave (mmWave) radars have enabled seeing through occlusions and provide valuable sensing information in scenarios where cameras and lidars fail. This includes applications ranging from cruise control in automobiles driving through fog to airport scanners seeing through clothes. Today, mmWave radars, have been miniaturized to the point where they are integrated in portable consumer electronics like smartphones. For example, mmWave radar in Google Pixel 4 assists in no light, hand gesture recognition. Unfortunately, the physical properties of these radar systems, compact form factors with few number of antennas, limits their sensing resolution. Current techniques for increasing radar resolution rely on large form factor radars or radar motion on motorized motion stages that dramatically increase weight, power, and more generally system practicality.

In this thesis, we propose three resolution enhancing techniques from just a 1.5 cm form factor mmWave radar with upto 8 antenna elements. We design these techniques for applications with dif ferent radar and object-to-be-imaged dynamics. First, we show imaging a structured moving object with spatially coded patterns achieves fine resolution sensing. Second, we show that by encoding the wave front with passive spatial codes, we can achieve general purpose imaging with a static single antenna radar and unconstrained object motion. Third, we design deep learning techniques to upsample directly from low resolution data and output high resolution images of static objects in the scene while the radar moves freely. To further enrich these images, we introduce temporal coded markers that can annotate radar images with object-specific information (akin to the role of QR codes for cameras). Together, we show that these techniques help in pushing the limits of compact radars and empower them with high resolution imaging for emerging applications.