Specific Emitter Identification with Principal Component Analysis (PCA) Dimensionality Reduction and Convolutional Neural Network

Abstract

The capability to identify and authenticate electronic devices through their physical intrinsic features has received considerable attention by the research community in recent years. This capability has been called in literature Specific emitter identification, radiometric identification or RF-DNA as a conceptual link to the physical identification of human being through the DNA. As in other fields of wireless communication, deep learning has been applied to implement such capability, where it has proven an excellent performance at the cost of the need of significant computing resources and time. To address the latter problem, this paper proposes a set of pre-processing steps based on time series segmentation, feature extraction and Principal Component Analysis (PCA) to reduce the data dimension given as an input to a Convolutional Neural Network (CNN) for classification. The proposed approach is able not only to reduce significantly the overall classification time, but it is also able to improve the classification accuracy in presence of noise in comparison to the baseline case where all the data is used.