Excess Attenuation Detection in Satellite Communication Channel Measurements with Deep Learning Architectures

In this paper a robust methodology for detecting excess attenuation events in satellite communication links is thoroughly investigated and described. The scope is to distinguish the existence of large excess attenuation values in the next future time-steps based on data from previous time-steps. Deep learning architectures, such as encoder-decoder and traditional paradigms are investigated and tested using experimental data from AL-PHASAT campaign in Athens and Lavrion, in Greece, at Ka and Q frequency bands. Specifically, artificial neural networks with convolutional and gated recurrent units are examined. Results of the deep learning techniques for the in excess attenuation’s event detection in the four datasets are very promising.