Spectrum Sensing using USRP SDRs and Convolutional Neural Networks

Current allocation of the spectral bands to only licensed users leads to inefficiencies in spectrum utilization. Designing realistic cognitive radios that are capable of sensing spectrum holes is the key to solve this problem and to increase the capacity of next-generation wireless networks. In this paper, we propose convolutional neural networks for predicting the spectrum holes from a data set obtained via USRP software-defined radios [1]. Preliminary results show performance improvements over the previously proposed methods.