An Introduction to Dimensionality Reduction for Propagation Measurements

This paper presents the dimensionality reduction of a propagation channel dataset created through a measurement campaign. The scenario knowledge is essential for the signal processing prior to the transmission and the application of these techniques allows a quick scenario differentiation. The measurement campaign, which has been performed in the SWAT research group's facilities at the University of Granada, includes propagation channels from three different scenarios: anechoic, reverberation and indoor. t-SNE, a Machine Learning technique that performs a non-linear dimensionality reduction, has been applied to the propagation channels. The results show the separability of the three scenarios in a low-dimensional space for an optimal election of the t-SNE hyperparameters. The most relevant hyperparameters in the performance of the technique are the distance metric and the perplexity.