RFInject: Injection of Simulated Radio Frequency Interference in Sentinel-1 Level-0 Data

Radio Frequency Interference (RFI) remains a major source of performance degradation in modern Synthetic Aperture Radar (SAR) missions, including the Copernicus Sentinel-1 constellation. Despite extensive research on RFI detection and mitigation, progress has been constrained by the lack of standardized, large-scale, and reproducible labelled datasets suitable for training and benchmarking data-driven methods. This work introduces RFInject, the first open framework for controlled injection of narrowband terrestrial interference—such as continuous-wave tones, pulsed radar signals, and frequency-hopping emitters—into clean Sentinel-1 raw bursts. The framework enables systematic generation of large, metadata-rich datasets that combine physical realism with full reproducibility. RFInject employs a parametric signal model that superimposes modulated chirp trains and tone-like components onto authentic radar echoes while preserving the spectral and statistical characteristics of operational systems. This design allows scalable control over waveform diversity, temporal behaviour, spatial extent, and interference power, facilitating reproducible evaluation and training of advanced detection and mitigation pipelines. The dataset covers 145 globally distributed Sentinel-1 Interferometric Wide (IW) products (2019–2025), each decoded at burst level and augmented with 10 independent RFI realizations per burst. To ensure compactness and transparency, the repository preserves both the clean echoes and the complete RFI parameter vectors, enabling on-demand regeneration of contaminated observations. Together, these design choices establish a physically consistent, metadata-rich benchmark for developing, validating, and comparing algorithms in RFI-resilient SAR processing.