“AI Models for Telecommunication Interferences Between GSO and NGSO Satellites: The model Generalization Problem”

Over the past decade, the deployment of multiple satellite telecommunication constellations in Low Earth Orbit (LEO) has posed a significant challenge to Geostationary Orbit (GSO) systems. As more satellites are launched, telecommunications interference has increased, primarily because both
GSO and Non-Geostationary Orbit (NGSO) satellites operate within the same frequency bands.
This interference may degrade GSO satellite services, impacting their overall quality. To maintain reliable service delivery, it is crucial to develop methodologies and solutions for detecting and managing telecommunication interference. The present study focuses on the generalization of an AI model developed by the Signal Processing and Communications (SIGCOM) research group at the University of Luxembourg for interference detection in satellite communications.[1][2] It examines the current state of the model and explores strategies to enhance its applicability across diverse scenarios. The model performance is evaluated with different interfered signals changing from the initial conditions of frequency and Signal to Noise Ratio (SNR) to different modulations which lead to multiple scenarios. The study demonstrated that the model works better detecting the interference in the frequency domain than in the time domain. Also, the changes of SNR have a greater impact in the behavior of the model in the time domain. In general, the model struggles with the detection of no-interference in the signal, confusing it with the interference. The study demonstrated that the model could generalize in frequency and can be trained with all the different scenarios to enhance its applicability.