Autenticação de dispositivos de comunicação sem fio utilizando aprendizagem profunda

Abstract

Electronic wireless devices cause unique physics variations in the emitting signals that can be used as a signature or a fingerprint. These variations can be interpreted as noise introduced by the electronic components of the transmitter into the signal. Due to subtle variations in the manufacturing process, no component is identical to another. To conduct the study, signals emitted by low-cost wireless communication devices were collected using a Software-Defined Radio (SDR) equipment. These signals were processed to create a database that was used for training and validation of a Convolutional Neural Network (CNN). The developed CNN demonstrated a 98% accuracy in device identification using only the signals emitted by them. The proposed technology operates passively and can be implemented in the control and gateway elements of a data network for device authentication. There is no need to modify the programming of the other network edge elements, which often suffer from severe hardware limitations. Convolutional neural networks have proven to be a powerful and promising tool with the ability to identify unique patterns in radiofrequency signals and unequivocally differentiate one device from another.