A Supervised Machine Learning Model based Spectrum Sensing using NI USRP-2922 SDR

Abstract

Spectrum scarcity is a major problem in this millennial communication engineering. Machine learning based spectrum sensing approaches are getting more attention among research community. The spectrum sensing techniques detects the licensed and unlicensed bands and supports spectrum management. In this work, we have proposed the spectrum detection as two level classification problem and solved using a supervised machine learning model based support vector machine(SVM) algorithm. The data samples are collected in real campus environment ranging from line of sight to non-line of sight using the Labview enabled NI USRP-2922 software defined radio platform for 815 MHz. Correlation and moving average metrics are two features used for classifcation. The effectiveness of the classfication based on the feature vectors are observed through confusion matrix, prediction estimation and detection probability.