Spectrum Hole Prediction in LTE-Based Cognitive Radio System Using Kolmogorov-Arnold Network

Abstract

This paper examines the problem of channel resource occupancy prediction using Radio Environment Maps (REM) based on Long Term Evolution (LTE) for a cognitive communication system. REM is a spatiotemporal database in the form of a resource grid with passing traffic in cells. For prediction, the Kolmogorov-Arnold network (KAN) architecture is used. A model structure has been developed that collects data, trains and tests KAN. The predictive model control algorithm is implemented in Python. MatLab was used to prepare input data and implement the LTE simulation model. Experiments are carried out and results of free channel resource prediction based on KAN and long-term memory are presented.