Rain Attenuation Prediction Modeling for Earth-Space Links using Artificial Neural Networks

Abstract

Predicting rain attenuation statistics of radio links is important for satellite communications. Currently, most existing rain attenuation prediction models are semi-empirical, with limited accuracy and anomalous behaviors, especially for links different from those used to develop these model. This is due to the limited human capability and the relatively small database for model development. In this work, rain attenuation prediction models were developed using an artificial neural network (ANN) with a single hidden layer. Despite of their simplicity, these ANN-based models, with sufficiently large training databases, can provide better performances than most existing models. Although, like most existing models, the overfitting is unavoidable for small training databases, it could be mitigated using the proposed method of database expansion. An ANN-based model developed using this expanded database could feasibly satisfy the physical consistency criteria with reasonably good performances.