Real-time estimation of excess atmospheric attenuation using an artificial neural network with a two-frequency input

A simple artificial neural network is considered for real-time estimation of excess atmospheric attenuation on a satellite communication link with known attenuation at two frequencies. All atmospheric contributors to attenuation are considered except for gases. The network has a two-layer feed-forward structure with 32 neurons in the hidden layer. Its performance is evaluated by computer simulation using 447 hours of measured attenuation data at 20, 40, and 50 GHz. Estimated attenuation tracks well the measured attenuation at 50 GHz. Estimation error standard deviation is 0.36 dB. RMS error is a function of attenuation: it increases slowly with attenuation, but the ratio of error to attenuation decreases with increasing attenuation. This approach accurately estimates excess attenuation without requiring assumptions, but required training data. (4 pages)