RFI Mitigation on Raw SAR Data Using U-Net for Enhanced Signal Reconstruction

Abstract

Synthetic Aperture Radar (SAR) systems are increasingly at risk of Radio Frequency Interference (RFI), driving the need for effective mitigation techniques to maintain data quality in challenging operational settings. This paper introduces a lightweight and effective framework designed to mitigate the impact of RFI directly on raw Radio Frequency (RF) SAR data. Given the scarcity of publicly labeled raw SAR data with interference events, a synthetic RFI dataset is created using a realistic RF-level SAR emulation tailored for spaceborne scenarios. An optimized U-Net architecture is developed to mitigate RFI in affected segments, inferring clean SAR data from contaminated RFI inputs. The framework’s mitigation performance is quantitatively validated on unseen synthetic data. Furthermore, qualitative assessments are performed on real raw Sentinel-1 data with RFI events, showcasing the framework’s ability to generalize effectively to real-world scenarios, even in the absence of ground truth data. This approach has the potential to enhance the quality of SAR imagery impacted by RFI.