Machine Learning Radio-Frequency-Based Anomaly Detection for Ground Station and Satellite Telecommunication

Machine Learning Radio-Frequency-Based Anomaly Detection for Ground Station and Satellite Telecommunication

Utah State University, Logan, UT

Satellite-to-ground station telecommunication is a crucial aspect of satellite missions, representing a single point of failure of the entire space system.

Each failed contact is an issue for all satellite missions, leading to a potential data loss. The detection and forecasting of data transfer failures are critical challenges in satellite operations, given the unpredictability and variety of potential causes for such anomalies.

Considering the spectral waterfall plot the most appropriate tool to describe the anatomy of satellite contacts, an automatic waterfall analysis could help satellite mission operators, by promptly discovering potential data transmission failures between satellites and ground stations, and by forecasting anomaly behaviors.

The work reported in this paper exploits machine-learning models, trained with spectrogram waterfall diagrams to provide real-time and automatic anomaly detection of data transmission failures. Long-Short Term Memory and Deep learning models have been trained and validated, for anomaly detection and forecasting of contacts failures, with a dataset encompassing a semester’s worth of satellite contacts in both S-band and X-band.

With examples to identify the most appropriate model, this research will present practical outcomes and data-informed best practices in support of mission operators.