An Amplitude-Phase Informed Diffusion Model for Near-Range SAR Imaging with Single Frequency and Sparse Arrays

Reconstructing millimeter-wave (MMW) images from single-frequency, spatially sparse sampling measurements reduces system complexity and facilitates real-time sensing. However, due to the limited information in sparse measurements, existing reconstruction methods often struggle to recover high-quality images with precise edges, fine structural details, and effective artifact suppression. An amplitude-phase informed diffusion model (APIDiff) for sparse reconstruction is proposed in this paper, where the amplitude information is exploited to characterize the object's 2-D planar shape and scattering intensity, while the phase information is leveraged to delineate object contours, textures, and 3-D structural characteristics. A spatial-wavelet (SW) encoder is introduced to extract amplitude and phase features from both spatial and frequency domains. A subsequent multimodal fusion module effectively integrates these features to enhance feature representation, providing conditional inputs to APIDiff. Additionally, an adaptive weighting strategy is integrated to accelerate model convergence. Extensive experiments on both simulated and real-world datasets demonstrate that APIDiff consistently outperforms existing methods in imaging quality across various spatial sampling rates and signal-to-noise ratios.