Atmospheric Attenuation Model Using Gaussian Process in Sub-THz Terrestrial Wireless Communications

Abstract

Atmospheric elements affect signal propagation and cause significant signal loss called atmospheric attenuation, especially in high-frequency range including sub-THz frequencies. Atmospheric attenuation mainly consists of gaseous, fog, and rain attenuation and the traditional approach considers them separately to develop good prediction models. However, the traditional models are not accurate and even unavailable for the high-frequency range, such as sub-THz frequencies. In this letter, we propose a new attenuation model to predict atmospheric attenuation for terrestrial line-of-sight propagation at high frequencies, such as sub-THz. The proposed model is based on Gaussian process regression (GPR). To train our model with a big measurement dataset, we use a scalable variant of GPR called blockbox matrix-matrix Gaussian process. We validate our model with a dataset obtained from a long-term measurement campaign at sub-THz frequencies. Our experiments show that our model significantly outperforms the existing model. We also show that our model provides reliable prediction intervals of atmospheric attenuation.