Super-Resolution Features and Distances for CSI Similarity Estimation

We consider super-resolution channel state information (CSI) representations in massive multiple-input multipleoutput wireless systems, and corresponding CSI distances. Taking advantage of the uniqueness of the multipath channel between the base station and the user equipment over a geographical region of interest, these features and feature distances can be used with, e.g., fingerprint localization and channel charting applications. We analyze CSI in terms of the angle-delay-power profile. The angle, delay, and power variables, having different units, can be combined in many ways to provide an interpretation of superresolution features in terms of point clouds in Euclidean geometry, corresponding to the set of multipath components. Point cloud techniques can then be used to compare channels described in terms of super-resolution features to each other. We evaluate feature and distance performance in terms of preservation of local topology and global geometry between the distances in the representation domain and distances in the spatial geometry.