Kolmogorov Model for Large Millimeter-Wave Antenna Arrays: Learning-based Beam-Alignment

A new approach is presented to the problem of beam alignment for large-dimensional millimeter-wave antenna systems, with a single radio-frequency chain, based on the application of the Kolmogorov Model (KM) framework to enable learning-based beam alignment. Unlike the conventional exhaustive search-based approach, the proposed KM does not require the entire beam space search, i.e., the number of beam soundings can be extremely smaller than the conventional approach, which is achieved by exploiting the predictive power of KM. We show, across several metrics, that by just sounding 25% of the beams, the proposed method approaches the performance of the exhaustive search method. Simulation results that validate the training and test performance of KM and illustrate the new method with significantly reduced overhead are presented.