High-resolution mmWave Imaging using Metasurface and Diffusion

Yida Wang, Yu Lu, Yuxuan Zhou, Yifei Shen, Lili Qiu, Zeyuan Lai, Yi-Chao Chen, Hao Pan, Juntao Zhou, Dian Ding, Mei Wang, Guangtao Xue, Qian Zhang

Abstract: In this paper, we propose a novel high-resolution mmWave imaging technique that operates with a small, off-the-shelf mmWave module and eliminates the need for any mechanical movement, offering a streamlined, portable solution. Our approach tackles two primary challenges: 1) mmWave commodity hardware is constrained by a limited number of antennas, limiting imaging resolution, and 2) most wireless imaging algorithms rely on compressive sensing to overcome the physical constraints, which assumes sparsity - a condition that may not always apply. To address these challenges, we first design an optimized mmWave metasurface specifically tailored for high-resolution imaging. This involves deriving a unit cell pattern that achieves high signal penetration and near-2π phase control, followed by joint optimization of both the metasurface and the codebook to further refine the signal quality and imaging resolution. We further propose a diffusion-based neural network model that transforms mmWave signals into high-quality images by directly exploiting the inherent features of target images, providing a robust alternative to conventional compressive sensing approaches. Our method encodes mmWave signals into physical representations and employs conditional generation through stable diffusion, effectively enhancing image quality. Through comprehensive implementation and rigorous testbed experiments, we demonstrate the feasibility and effectiveness of our approach.

External IDs:doi:10.1145/3711875.3729162