Machine Learning based Spectrum Prediction in Cognitive Radio Networks

Abstract

According to the Cisco’s white paper for the year 2018-2023, machine-to-machine (M2M) connections are mentioned as the first fastest growing connections, with a 2.4 fold increase between 2018 and 2023. This will possibly lead to an increase in radio spectrum utilization. The spectrum will be congested due to its limited availability, and interruption of services also occurs in high-traffic scenarios. To overcome this drawback, Cognitive Radio (CR) acts as a promising and intelligent technology that facilitates the unlicensed users (Secondary Users) to efficiently utilize the spectrum allotted to the licensed users (Primary Users) without imposing any interference to them. In order to increase the coexistence of devices without modifying anything in terms of hardware, CR has the feasibility of providing solutions to spectrum prediction for end users. Further, to improve spectrum prediction, machine learning algorithms greatly help the cognitive radio to select the appropriate spectrum based on the requirements of secondary users. In this paper, machine learning algorithms like Random forest classifier, Logistic Regression, KNN classifier, Decision Tree classifier, Artificial Neural Network (ANN), Support Vector Machine (SVM) are used to demonstrate how the proposed model can be used for making spectrum prediction based on the dataset applied to the network and predicting whether the spectrum is used for voice or data communication. The selected machine learning algorithms are implemented, and their performances are compared against a given data set consisting of transmission power, frequency, and duty cycle. The proposed model will have the capability of selecting the best suitable algorithm for the given data set. Further, the processed information can be used in cognitive radio networks for the effective utilization of channels. From simulations, it is clear that, by using appropriate ML technique, it will most probably increase the spectral prediction with the highest accuracy of 85%.