Extreme Learning Machine based Spectrum Sensing in Coloured Noise with RTL-SDR

Abstract

The availability of inexpensive software defined radios (SDR) has enabled the deployment of cognitive radio (CR) features in large-scale networks such as internet-of-things (IoT). However, such radio receivers are limited by their non-ideal characteristics like coloured noise, IQ imbalance, phase noise etc. Performance of existing spectrum sensing algorithm degrade in coloured noise due to swelling effect of received signal covariance matrix. To overcome this limitation, we propose a novel spectrum sensing technique based on extreme learning machine (ELM) which uses eigenvalue and log determinant (LogDet) of covariance matrix features. Experimental results show the effectiveness of the proposed technique over existing algorithms in literature.