Simulating radio frequency interference in Sentinel-1 at raw data level

The radio-frequency spectrum has become increasingly congested due to the proliferation of communication and sensing technologies serving the needs of an expanding society. This growing occupancy has led to the coexistence of multiple services within the same or nearby frequency bands, raising the likelihood of Radio Frequency Interference (RFI). On top of that, the escalation of military conflicts worldwide adds further complexity to the use and management of the spectrum.

Synthetic Aperture Radar (SAR), despite being an active microwave imaging sensor, is not immune to RFI, which can significantly compromise the integrity and reliability of acquired data. At the same time, the increasing frequency and severity of natural disasters driven by climate change reinforce the need for continuous, accurate monitoring of the Earth’s dynamic environment. SAR plays a key role here thanks to its ability to operate regardless of weather or lighting conditions. 

In this context, the detection, characterisation, and mitigation of RFI have become critical to ensuring reliable SAR data for disaster prediction, monitoring, and damage assessment.

The approach: a modular RFI simulator

To address this need, we developed a modular software framework that simulates multiple types of RFI at raw data level and injects them into clean Sentinel-1 acquisitions. The simulator serves two purposes.

First, it provides a controlled environment to evaluate the performance of conventional detection and mitigation techniques, identify their limitations, and explore potential improvements.

Second, and equally important, it enables the generation of a synthetically contaminated and accurately labelled dataset, which is essential for training Deep Learning models. Reliable labels are notoriously hard to obtain when working with real-world RFI, and a high-fidelity simulator helps achieve good generalisation while reducing the risk of domain shift when these models are later applied to operational data.

What the project delivered

The thesis combines three contributions in a single framework: a modular implementation of multiple RFI types at raw processing level, a set of detection and mitigation algorithms integrated under a common evaluation framework that enables fair comparison, and a labelled dataset designed to train Deep Learning models capable of detecting RFI before SAR focusing, where mitigation is most effective.

Making it possible at Zenithal Blue

The purpose of a master’s thesis is to push into the borders of knowledge in a specific field. Thanks to Zenithal Blue, I was able to do exactly that: explore new directions for the company while laying the foundation of my own career, at least for the coming years through a PhD. The trust the company placed in me was the main reason I stayed motivated and felt capable of taking the project on.

For a master’s student taking the first steps into the space sector, collaborating with highly skilled professionals is an invaluable experience. Being part of a company involved in top-tier projects and feeling like a real contributor rather than an observer is something not every student gets to experience, and it has made all the difference.