Analysis of RF Device Fingerprinting using Convolutional Neural Network

Abstract

The authentication of devices and identification of unauthorised signals has been difficult in many scenarios. In this paper, a method for identifying a specific device among a range of similar devices is discussed using the techniques of convolutional neural network. The approach is to use radio fingerprints as samples for the identification process. The variability caused in the manufacture of these wireless transmitters serves as a unique feature, creating repeatable signatures for a particular device. These signatures are used as fingerprints for identification and verification. A framework is then developed for training the convolutional neural network using complex baseband error signals in time domain for varying signal-to-noise ratios (SNRs).