UC Irvine

Channel charting, is a novel unsupervised machine learning method to get the representation of multiple transmit locations to reflect the local ``radio geometry'' by processing the channel state information with a dimensionality reduction method. In this thesis, channel charting results in three different channel charting algorithms, channel state information (CSI) feature extraction method, efficient distance measure method, and correlation matrix distance (CMD) method, together with three dimensionality reduction methods, principal component analysis (PCA), sammon's Mapping (SM), and Isomap, under two Quadriga channel models, line-of-sight (Q-LoS) model and a non-line-of-sight (Q-NLoS) model, are presented. In order to evaluate the quality of the performance of these nine alternative methods, three performance measures, continuity, trustworthiness. and Kruskal’s stress, are considered. The simulations are set under both 2-dimension and 3-dimension scenario, and also under single subcarrier and multi-subcarriers. Several observations are made from these simulation results.