Hardware implementation for 802.11b/g/n signal classification

Abstract

The number of devices which can communicate wirelessly are increasing with the technological improvements day by day. This situation causes congestion on electromagnetic RF spectrum. RF spectrum should be utilized e ciently due to the fact that it is highly limited. Cognitive Radio (CR) detects primary user signals in spectrum and lets secondary users transmit their signals when spectrum is free. CR achieves detection of primary user signals by spectrum sensing techniques such as energy detection, cyclostationary, matching lter, radio identi cation and classi cation based methods. Radio identi cation provides more information than the others by detecting transmission technology of signal. In last years, machine learning and deep learning based methods are also used in spectrum sensing tasks. Success possibility of deep learning extremely depends on data. Therefore, deep learning shows strong performance on radio identi cation applications due to obtaining huge amount of data from Software De ned Radios (SDR). In this work, Convolutional Neural Network (CNN) based radio signal classi er hardware is designed using Universal Software Radio Peripheral (USRP) E310 embedded SDR. Most of the works in literature use SDRs as slave devices that run connected to computer. However, embedded hardware is small sized and brings opportunity of eld deployments by running standalone. Designed hardware can classify 802.11b/g/n signals blindly. Classi cation process starts with scanning spectrum and obtaining spectrogram image which is time-frequency transformation of spectrum. Generated spectrograms are fed into the CNN based classi er and prediction of input is achieved. CNN based classi er is trained previously by collecting 802.11b/g/n signals from wireless modem. Our work can classify LTE, GSM or other signals due to adaptivity of deep learning by training CNN with those signal samples.