A Deep Learning Model for LoRa Signals Classification Using Cyclostationay Features

Abstract

With the witnessed exponential growth of Internet of Things (IoT) nodes deployment following the emerging applications, multiple variants of technologies have been proposed to handle the IoT requirements. Among the proposed technologies, LoRa stands as a promising solution thanks to its tiny footprint in terms of cost and power consumption. Since the ISM band is usually used for such applications and multiple different systems are allocated in this band, a smart spectrum management and awareness is highly required. In this paper, we propose a convolutional neural network (CNN)-based classifier to identify LoRa spreading factors (SF) and the inter-SF interference. Specifically, in the proposed model LoRa signals are pre-processed using spectral correlation function (SCF) and fast Fourier transform (FFT). We show that using the SCF pre-processed signals for training can attain a better performance as compared to those with FFT pre-processed training data in terms of classification accuracy at a very low signal-to-noise ratio. Furthermore, the proposed model outperforms the related model in literature in terms of accuracy for the FFT and SCF pre-processed signals.