Neural network prediction of HF ionospheric propagation loss

A generalised regression neural network is used to predict losses inherent in ionospheric radiowave propagation. Network inputs consist of sun declina tion, time of day, radio flux, geomagnetic A-index and X-ray flux. Simulati ons for a 400km path demonstrate a 2.5dB error between network predictions and actual measured values, representing a 46% reduction in errors compared to the linear regression method.