Message Structure Aided Attentional Convolution Network for RF Device Fingerprinting

RF device fingerprinting has become an emerging technology which identifies the device-specific fingerprint based on inherent defects in the hardware. However, existing methods pay little attention to the potential improvement of rough priori information such as message structure on the identification performance. In this paper, we propose a message structure aided attentional convolution network (MSACN) for RF device fingerprinting. Portions with different pulse waveform distribution are separated and fed into the identification network. The network extracts and merges the feature map contained in multiple data blocks, which is helpful to explore the internal relation of data. Furthermore, we design a spatial attention mechanism for low-dimensional discrete signals to pursue more efficient feature fusion. Experimental results on the dataset of real-world ADS-B transmissions show that MSACN can achieve 98.20% identification accuracy outperforming previous works.