Embrace Imperfect Datasets: New Time Representation for RFF Identification

Abstract

As the inherent attribute of equipment circuit hardware, Radio Frequency Fingerprints (RFFs) is hardly-forged and has become one of the most powerful guarantees of physical layer security. Most existing RFF-based methods ignore the temporal relation and are designed under an ideal dataset with a large number of samples and complete signal records, thus they tend to be less versatile in real-world scenarios. To address this problem, we propose a novel time representation method for wireless signal pictorialization called modified gramian angular fields (MGAF), which depicts the characteristics of the signal along the time axis through the transformation of coordinate system and a representation of trigonometric difference. After that, a channel-selectable convolution neural network (CNN) is used to extract high-dimensional feature vectors as the RFFs for further identification. The entire experiments are conducted with purposely poorly designed datasets. The results shows the accuracy can reach at 94.82% with only three half-sine waves and 99.26% with a quarter of the preamble at the SNR level of 30 dB.