Channel Charting with Angle-Delay-Power-Profile Features and Earth-Mover Distance

Abstract

We are interested in deducing whether two user equipments (UEs) in a cellular system are at nearby physical locations from measuring similarity of their channel state information (CSI). This becomes essential for fingerprinting localization as well as for channel charting. A channel chart is a low dimensional (e.g., 2-dimensional) radio map based on CSI measurements only, which is created using self-supervised machine learning techniques. Analyzing CSI in terms of the angle-delay power profile (ADPP) takes advantage of the uniqueness of the multipath channel between the base station and the UE over the geographical region of interest. We consider super-resolution features in the angle and delay domains in massive multiple-input multiple-output (MIMO) systems and consider the earth-mover distance (EMD) to measure the distance between two features. Simulation results based on the DeepMIMO data set show that the super-resolution ADPP features with EMD leads to a better quality channel chart as compared to other CSI features and distances from the literature.