Emitter detection for the Internet of Things based on transient data using convolutional neural networks and the general linear chirplet transform with optimization

Abstract

In this letter, the General Linear Chirplet Transform (GLCT), a time-frequency representation recently introduced in the literature, is probed for its potential use in conjunction with Using a Convolutional Neural Network (CNN) to recognize IoT wireless devices.
During transmission, radio frequency emissions from the IoT devices are analyzed to determine their identities. By presenting an optimization approach for GLCT tailored to emitter identification, we use the innovative combination of CNN and GLCT (CNN-GLCT) to the transient sections of the radio frequency emissions. We demonstrate empirically that this combination outperforms previous Deep CNN techniques and shallow machine learning methods like SVM and KNN, notably in low Signal-to-Noise Ratio (SNR) and fading environments, where these other methods struggle.