A New Image Based Regression CNN for Predicting V-band Power Attenuation

A novel application of machine learning and image processing for estimating mm-wave attenuation due to varying weather conditions is developed and tested. A deep learning approach is taken to analyze doppler weather radar images obtained from the NEXRAD radar system. Pairing the radar images with RF power data obtained from a 24 km, 72 GHz terrestrial link, the received power is estimated through a Convolutional Neural Network, tuned for regression, and the attenuating effects of clear skies and cloudy days are compared to experimentally measured results.