Détection d’anomalies dans les signaux radio par apprentissage non supervisé

Abstract

This master's thesis focuses on anomaly detection in radio signals using Machine Learning techniques. The notion of anomaly is defined based on the Signal-to-Noise Ratio (SNR), which is an essential indicator of radio signal quality. The experimental study uses the RadioML2016.10a dataset to compare three anomaly detection methods: Autoencoder, One-Class Support Vector Machine (OCSVM), and Isolation Forest, by exploiting the In-phase and Quadrature (I/Q) components of radio signals. The results show that the three methods can effectively detect anomalies. However, the Autoencoder achieved the best performance due to its ability to learn deep and complex representations of the data. These findings confirm the potential of Machine Learning techniques to improve the reliability and security of wireless communication systems.