Deep Learning Based Radio Frequency Fingerprint Identification by Exploiting Spatial Stereoscopic Features

The unique radio frequency fingerprint (RFF) resulting from hardware imperfections can be used to identify wireless devices to resist impersonating or spoofing attacks. The existing methods for RFF identification typically rely on the transient or steady-state features of RF signals. However, the limited number of extracted features affects the performance of radio device identification, which is not satisfactory. This paper proposes a spatial stereoscopic feature extraction method that transforms the three statistical features based on radio signal frequency envelopes into three-dimensional images to address this issue. By increasing the dimensionality, it can improve the quantity of features. Additionally, we propose a multi-channel parallel convolutional neural network (MPCNN) for classifying devices based on the spatial stereoscopic features. Experimental results demonstrate that our proposed method outperforms benchmark approaches in terms of identification accuracy. Specifically, it achieves mobile device identification with an accuracy higher than 99% even with a small sample size.