Cognitive Radio Networks Channel State Estimation Using Machine Learning Techniques

Abstract

In interweave Cognitive Radio Networks (CRNs), monitoring the spectrum to detect unused portions (holes) is done by the spectrum sensing function however, it consumes both time and energy. So, some protocols use prediction to estimate the channel availability. One of these protocols use Hidden Markov Model (HMM) but in a very simple way. So, it does not perform well in several cases. In this paper, we propose two new protocols for cognitive radio channel availability prediction. Both protocols use HMM but in a more advanced way. They divide the data into two sets, thus create two HMM models. The first protocol uses Bayes theorem together with these two models, while the second one uses Support Vector Machine (SVM) with the two models HMM parameters. Evaluation of the two protocols proves that both protocols perform better than the old one that uses HMM in a classical way. It also proves that using SVM with HMM parameters is better than using HMM only. This is because dividing the data into two sets for training the protocols with, gives more flexibility to both protocols.