Predicting V-Band Attenuation Through a Multiple Input Modality CNN/MLP Hybrid Model

A hybrid input deep-learning model utilizing both radar images obtained from the NEXRAD platform along with tabular meteorological measurements was developed to predict V-band attenuation due to the atmosphere and precipitation events. Pairing radar images and tabular data with RF power measurements obtained from a 24- km, 72 GHz terrestrial link, the received power along the link was estimated and compared to an image-only CNN model. The hybrid model demonstrated better performance relative to the CNN model both in terms of overall error and in mean absolute percentage error.