In a world where more and more systems transmit simultaneously (5G/6G, IoT, satellites, drones), the electromagnetic spectrum is becoming increasingly crowded, and interference is no longer an exception but a frequent and natural condition. The challenge in this project is to separate signals that arrive simultaneously, within the same frequency band, often without an accurate model of the source signal or the interference, and to do so in real time. To address this, we leverage powerful deep learning tools capable of learning complex statistical structures. Yet a central and non-trivial question remains: how can we design solutions that are reliable, stable, interpretable, and capable of generalizing to new scenarios?

In this project, we combine approaches from statistical signal processing and machine learning. Our goal is to develop models that, on the one hand, respect well-established mathematical formulations, and on the other hand, learn complex physical phenomena that are difficult to describe precisely through analytical models alone.