Autofocus Algorithm for Real-Time Correction of Residual RCM Based on ANCPS

• Published in: IEEE transactions on geoscience and remote sensing Vol. 63; pp. 1 - 16
• Main Authors: Xie, Yi, Chen, Longyong, Zhang, Fubo, Xu, Yihao, Yang, Ling
• Format: Journal Article
• Language: English
• Published: New York IEEE 2025; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Apertures; Autocorrelated normalized cross-power spectrum (ANCPS); Azimuth; Cell migration; correction of range cell migration (RCM); Field programmable gate arrays; Fragments; Graphics processing units; High resolution; Image resolution; Image segmentation; Imaging; Lightweight; Low cost; parallel computing; Radar imaging; Real time; Real-time systems; Robustness; SAR (radar); Synthetic aperture radar; synthetic aperture radar (SAR) real-time autofocus; Trajectory; Weight reduction
• ISSN: 0196-2892; 1558-0644
• DOI: 10.1109/TGRS.2025.3537865

With synthetic aperture radar (SAR) systems being developed with miniaturization, lightweight, low-cost, high-resolution, and real-time capabilities, the scope of their application has expanded to some lightweight platforms that cannot carry a high-precision position and orientation system (POS) and whose trajectories are susceptible to interference. This often results in residual range cell migration (RCM) exceeding several or even dozens of range resolution cells, which notably degrades the performance of traditional autofocus algorithms and seriously affects the quality of real-time imaging. Considering this, we propose a real-time residual RCM correction scheme based on the autocorrelated normalized cross-power spectrum (ANCPS). First, the range-compressed image is segmented into blocks based on the azimuth and range. The ANCPS is then used to estimate the optimal residual RCM for each subblock. Second, the optimal segment is selected from the residual RCM curve fragments estimated by various subblocks. Finally, the residual RCM fragments are spliced and filtered to correct the complete data. When the algorithm is deployed on a GPU platform, it only takes 1.39 s to process an <inline-formula> <tex-math notation="LaTeX">8\times 40 </tex-math></inline-formula> K high-resolution original SAR image, and the calculation time is reduced by 94.6% compared with only using a CPU for calculation. The experimental results demonstrate that the proposed residual RCM correction scheme has good convergence, can achieve subpixel residual RCM compensation without iteration and interpolation, has a simple calculation process, is efficient, and has strong parallelism. It is thus suitable for deployment in GPUs and is conducive to realizing real-time SAR autofocus at a low cost and with a high resolution.