MM-2FSK: Multimodal Frequency Shift Keying for Ultra-Efficient and Robust High-Resolution MIMO Radar Imaging

Abstract

Accurate reconstruction of dynamic targets demands three-dimensional imaging at high temporal resolution. Radar is a particularly attractive sensing modality for this purpose due to its fast capture capabilities and robustness under varying lighting and weather conditions. Consequently, it has been successfully employed in a variety of dynamic applications, such as automotive and robotic perception. Nevertheless, dynamic sensing remains comparatively unexplored in operating regimes requiring near-field, high-fidelity surface reconstruction, as commonly targeted by large-aperture mmWave MIMO radars. Here, achieving high frame rates is challenged by the substantial computational cost of reconstruction, imposing stringent constraints on bandwidth, compute resources, and ultimately sensor capture rate. To enable fast captures at reduced computation cost, prior work explored few-tone imaging, reducing the number of transmitted frequencies at the expense of increased range-phase ambiguity; however, this requires coarse knowledge of the target position and is primarily designed to reconstruct flat surface geometries. We present MM-2FSK, which extends two-tone stepped frequency imaging with an external modality-agnostic per-point depth prior for precise near-field 3D imaging of geometrically complex surfaces. To address range-phase ambiguities, the prior is processed to mitigate inter-modality data mismatch, transformed into a radar-compatible representation, and integrated into signal processing as a phase constraint. In experiments with a mmWave FSCW MIMO radar and with representative priors obtained from a co-located optical sensor, we demonstrate superior depth accuracy of our method over comparable near-field few-tone radar imaging approaches in static and dynamic scenes, competitive with time- and resource-intensive measurements of many-tone backprojection.