High-Resolution Imaging from a Compact mmWave Radar
Abstract
High-Resolution Imaging from a Compact mmWave RadarSwarun Kumar, Anthony Rowe[swarunk, agr]@andrew.cmu.eduCarnegie Mellon UniversityApproved for Public ReleaseProject AbstractThis proposal explores 3-D lidar-like high-resolution imaging of a highly occluded scene(e.g., dueto smoke or fog) solely from a compact (few centimeter sized) mmWave radar. Despite extensivework on radar imaging, including at mmWave # state-of-the-art solutions strike a trade-off betweensize and resolution. High-resolution radar imaging systems that offer one degree-scale resolutionare several meters in length to achieve the desired radar aperture. In contrast, compact mmWaveradars such as those found for automotive collision sensing or smartphone-based gesture sensors aretypically a few centimeters across,yet offer extremely coarse resolution (several tens of degrees).Having a radar imaging solution that is both compact and high-resolution can enable transformativenaval applications. Consider for example, tiny single-chip embedded radar platforms mounted atopUnmanned Aerial Vehicles for imaging, sensing and navigation through a foggy environment forsearch-and-rescue. Or consider a body-worn orground-robot enabled radar system imaging andproviding situational awareness through dust or smoke in a warzone or for disaster recovery.This proposal presents mmSense, a compact mmWave radar system that obtains 3-D lidar-likepoint clouds from a few-centimeter across mmWave radar platform. mmSense proposes to achievethis through a combination of two innovations: programmable smart surfaces and a machine-learningpipeline. First, mmSense explores designing a form of structured signal multipath thataims to spatially modulatesignals from the radar platform, thereby enhancing overall aperturewithout requiring a large number of antenna elements or physicalmovement of the radar. Second,mmSense investigates a deep neural network architecture that designs a data-driven pipeline to allowradar to emulate lidar, improve resolution and eliminate clutter that is specifically common inradar images. mmSense will be implemented and evaluated in smoky and foggy environments inheterogeneous demonstrations including ground robotic, aerial and body-worn use cases.mmSense directly addresses multiple naval operational contexts including search-and-rescue,disaster recovery and intelligence gathering. The result of this research will enable a new sensingmodality for situational awareness in size and weight-constraint contexts across varied operationaldomains including on-body, aerial, and ground platforms. This project will engage Navy personneland ROTC Midshipmen in the Pittsburgh region and leverage their rich experience and knowledgeof naval platforms and use cases. The project will employ the Hacking for Defense methodologyto investigate the high-level Concept of Operations that demonstrate the capabilities of mmSense.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Nov 08, 2024
- Source ID
- N000142412410
Entities
People
- Swarun Kumar
Organizations
- Carnegie Mellon University
- Office of Naval Research
- United States Navy