Neural Network for Forecasting Parameters of Broadband Transionospheric Radio Channels

Methods of synthesis of a special recurrent neural network for the problem of predicting the coherence band of transionospheric radio channels are investigated. An algorithm for data preparation and neural network training has been developed. Validation of the developed neural network was carried out on the data of full-scale experiments conducted with the help of passive radio sensors of global navigation satellite systems using phase and code measurements. The achieved high accuracy of the forecast is experimentally shown using the metrics of the coefficient of determination, MAE and MAPE.