AI Based Short Term Prediction of KA Band Rain Attenuation: A Comparative Evaluation

The ever-growing need for high throughput broadband services continues to drive the exploitation of EHF bands for satellite communication services. But the impact of rain attenuation remains a challenge for communication at these bands. To this end, several FMTs have been developed to make signals more robust against attenuation. For accurate operation of these FMTs, there is the need for real time knowledge of the link condition to adjust the configurations and maintain quality of service (QoS). This study presents a comparative evaluation of three artificial intelligence (AI) algorithms, namely Multilayer Perceptron (MLP), Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN), for the short-term prediction of rain attenuation at Ka band, using both beacon and meteorological data obtained over a three-month period at Chilbolton observatory in Southern England. The prediction term length (i.e., how far into the future a prediction is made) was varied from 1 s up to 6 s, and the prediction was done using past measured beacon attenuation data only as well as using both past measured beacon attenuation and meteorological data.