Deep Learning based Scintillation Prediction for Satellite Link using Measured Data

The satellite communication link is affected by the inhomogenities of the refractive index over the atmospheric channel that causes rapid variations in the amplitude of the received signal at the ground station i.e. scintillation fading. Such scintillation fading show significant enhanced activity depending on certain time during a typical day and seasonal variation that could act as a potential degrading source especially for a low-margin satellite link under clear weather condition. In this work, we apply a recurrent neural network (RNN) for the prediction of scintillation fading at a future observation time that is 240 milliseconds ahead (around the round trip delay of a Geostationary satellite). We show that the performance metric of the predicted scintillation- root mean square error and mean absolute error are observed to be very low. The error histogram reveals that the error mostly lies within the range of 0.02 dB on a typical clear weather day and thus making it extremely useful for adapting to fade mitigation techniques in real time for a satellite link.