Synthetic Aperture Radar (SAR) systems are critical for Earth observation, providing high-resolution imaging capabilities under all weather conditions and independent of daylight. These systems are widely utilized in environmental monitoring, disaster response, and security operations. However, SAR imaging is susceptible to anomalies caused by unintentional electromagnetic emissions, intentional jamming, multipath propagation, terrestrial radars, and system-induced artifacts, which can degrade the quality and reliability of SAR data. The growing use of the Radio Frequency (RF) spectrum has exacerbated the issue of Radio Frequency Interference (RFI), posing a significant threat to SAR systems. RFI can obscure features, introduce artifacts in raw data and focused images, and complicate data interpretation, potentially undermining SAR’s effectiveness in critical applications. Terrestrial RFI, particularly Narrow-Band Interference (NBI), is a dominant source of anomalies, and its mitigation is challenging due to the complexity of SAR data and the need for real-time processing in operational scenarios.

This research explores the challenges of RFI in spaceborne SAR systems by investigating both traditional and Machine Learning (ML)-based detection and mitigation methods. Traditional approaches, such as notch filtering, adaptive filtering, and signal decomposition, offer established methodologies for addressing interference under certain conditions. At the same time, ML and Deep Learning (DL) approaches present promising capabilities in handling complex, non-Gaussian interference environments by leveraging their ability to model and isolate interference patterns. This study assesses and quantifies the potential and limitations of each approach to identify their suitability for various RFI scenarios. Akey contribution of this work is SEMUS (SAREMUlator for Spaceborne Applications), an open-source emulator designed to simulate RFI effects on SAR data at the RF level. SEMUS generates Phase History Data (PHD) and employs the Range-Doppler Algorithm (RDA) to produce focused images. It incorporates an empirical azimuth matched filter and injects various types of interference, including NBI, to evaluate their impact on SAR data. This enables comprehensive testing of RFI detection and mitigation methods in a controlled environment, advancing the understanding of radar signatures and interference suppression in spaceborne SAR systems. Two innovative approaches are employed to detect and mitigate RFI. The first approach leverages advanced signal processing techniques, including a two-dimensionalVariableAttenuation Space-Frequency Filter (VASFF), which exploits the time-frequency characteristics of RFI signals.

This method has been validated on real TerraSAR-X raw data in collaboration with the German Aerospace Center (DLR). The second approach apply DL, featuring a U-Net-based segmentation model, along with a mitigation model, developed in collaboration with the European Space Agency (ESA). These models are trained and tested on synthetic datasets emulating Sentinel-1 SAR data with radar interference and quantitatively verified on the synthetic data and qualitatively validated on real Sentinel-1 data, demonstrating its effectiveness in detecting and mitigating NBI. As an extension of radar signature recognition beyond SAR, this work explores air-writing recognition. The proposed approach utilizes a single ultrawideband (UWB) radar, leveraging five ML models and multiple data representations to achieve robust and practical performance. With a simplified design requiring minimal preprocessing, the system adapts effectively to various handwriting styles, hand orientations, and writing speeds. Experimental results validate its effectiveness in realistic scenarios, showcasing its potential to advance human-computer interaction and gesture-based control applications.

The findings of this research offer significant contributions to both signal processing and deep learning approaches, setting a new benchmark for enhancing SAR performance in the presence of RFI. These innovative methodologies not only demonstrate feasibility for practical implementation but also pave the way for real-time processing across diverse Earth observation applications. By addressing critical global monitoring and management challenges, this work establishes a robust foundation for the operational deployment of high-quality SAR systems, reinforcing their pivotal role in tackling pressing environmental and societal needs. Additionally, the exploration of air-writing recognition using UWB radar extends the scope of radar applications, showcasing its potential to revolutionize human-computer interaction and gesture-based technologies through robust and adaptable machine learning frameworks.