PSF Distortion Correction for Multistatic FMCW SAR Imaging Using Geometry-Aware Network

This paper proposes a geometry-aware point spread function (PSF) regularization network (GPR-Net) to correct geometry-induced distortions in cross-track multistatic frequency modulated continuous wave synthetic aperture radar (FMCW SAR) imaging. Bistatic configurations suffer from anisotropic PSFs that vary nonlinearly with the transmitter–receiver baseline, bistatic angle, and squint angle. GPR-Net incorporates these geometric parameters as conditional inputs and employs a feature-wise linear modulation (FiLM)-based U-Net coupled with a dynamic kernel prediction module to learn geometry-specific PSF corrections. A GPU-based FMCW SAR simulator was implemented to generate an ideal single-point target dataset that reproduced the PSF distortions under varying geometric conditions. The experimental results show that GPR-Net reduced the anisotropy by approximately 23 % and the orientation error by over 70%, thereby achieving morphologically stable PSFs across diverse geometries. These findings demonstrate that geometry-aware regularization can achieve resolution homogenization and PSF-level response consistency in multistatic SAR imaging simulations, and suggest its potential extension to actual SAR data and more complex multistatic environments.