Detection of spectrum hole from n ‐number of primary users using machine learning algorithms

Abstract

A method for detecting spectrum holes based on the n ‐number of primary users (PU's) in a cognitive radio environment, using a cooperative spectrum sensing model is proposed in this study. The fusion centre, senses the n ‐number of PUs. When the number of PUs is >200, the probability of detection decreases, while the probability of a false alarm increases. The authors use the random forest (RF) algorithm to classify a customised dataset of 600 training samples. Further, they compare the RF algorithm and the k ‐means clustering algorithm, using test datasets with a minimum of ten PUs and a maximum of 500 PUs. Five different signal features are considered as the attributes in the proposed model. The maximum probability of detection is achieved using the k ‐means clustering algorithm in the case of 200 PUs and is 99.17%, while the false alarm probability is 0.8%. The receiver operating characteristic curves indicated that probability of detecting a spectrum hole in the case of the dataset with 500 PUs is 97.67% with the signal to noise ratio ranging from 10 to −12 dB. The accuracy can be increased if the number of clusters formed is increased, depending on the number of test samples.