Q-learning Based Radio Channels Utility Evaluation Algorithm for the Local Dynamic Spectrum Management in Mobile Ad-hoc Networks

Abstract

This paper presents advantages of the machine learning used for estimation of specific radio channel usefulness, necessary for dynamic spectrum access. This method enables more efficient use of spectral resources, that are temporarily not used by licensed users. It indicates which channels are the most useful, i.e. give the highest probability of successful transmission and avoidance of interferences. Profile of Q-learning algorithm operation may be controlled by adaptation of the learning rate and greedy parameter.