Stochastic Characterization of outdoor Terahertz Channels Through Mixture Gaussian Processes

This contribution aims at experimentally validating the suitability of Gaussian mixture (GM) distributions to capture the stochastic characteristics of outdoor terahertz (THz) wireless channels. In this direction, we employ a machine learning enabled approach, based on the expectation maximization algorithm, in order to identify the suitable number of Gaussian distributions as well as their corresponding parameters that result to an acceptable fit. The fitting accuracy of the GMs to the empirical distributions is evaluated by means of the Kolmogorov-Smirnov (KS), Kullback-Leibler (KL), root-mean-square-error (RMSE) and R-2 fitting accuracy tests. These tests verify the suitability of GMs to model the small-scale fading channel amplitude of outdoor THz wireless links. In addition, the fitting accuracy results indicate that as the number of mixtures increases the resulting GMs achieve a better fit to the empirical data.