Imaging time series for internet of things radio frequency fingerprinting

Abstract

The concept of Radio Frequency (RF) fingerprinting is that electronic devices can be identified and authenticated through their radio frequency emissions, which contain intrinsic features of the device itself. RF fingerprinting can be used to enhance the security of wireless networks because the fingerprints provide a form of authentication. In previous research papers, the RF fingerprints have typically been obtained by extracting statistical features from the time series generated by the analog-to-digital conversion of the RF emissions. In this paper, we investigate a novel approach to the RF fingerprinting of Internet of Things (IOT) devices, where the time series are converted into images, out of which image processing features are extracted. The performance of this approach is experimentally evaluated by applying different machine learning algorithms on different types of conversions of time series to images. Our analysis shows that the proposed approach provides a better identification accuracy as compared to the accuracy achieved by conventional sets of statistical features used in the literature. Even if relatively small (around 1%), this accuracy improvement is statistically significant when classification is repeated over different folds of the training and test data. Yet, this enhanced accuracy is obtained at the cost of the longer time taken to process the images.