Exposing the Fingerprint: Dissecting the Impact of the Wireless Channel on Radio Fingerprinting

Research on emitter individual identification technology based on Automatic Dependent Surveillance–Broadcast signal

“…The features we extracted ensure that what our machine learning method has learned is the result of the combination of the transmitter's hardware damage and channel characteristics. This is the biggest difference between our research and the research by Al-Shawabka et al 10 Ying et al 11 study a two-stage spoofing detector based on deep neural network (DNN), and perform individual identification of ADS-B signals in the second stage of detection. Ying et al 11 use phase information independent of International Civil Aviation Organization (ICAO) address as detection feature in the second stage.…”

“…Al-Shawabka et al 10 study the influence of wireless channels on the classification accuracy of convolutional neural network (CNN)-based radio fingerprint recognition algorithms, and have obtained convincing experimental results. The results of literature 10 show that only in the two experimental settings of ''setup C'' and ''setup D'' is the hardware damage of the transmitter learned by CNN. Under the two experimental settings of ''setup A'' and ''setup B,'' what is learned is the channel characteristics.…”

“…Under the two experimental settings of ''setup A'' and ''setup B,'' what is learned is the channel characteristics. The ADS-B signal used in literature 10 was provided by DARPA and was used to test the impact of changes in the number of equipment and signal-to-noise ratio on the recognition accuracy of the CNN recognition algorithm. This article does not clearly explain how the ADS-B signal is collected.…”

“…The summary of the differences between the literature [10][11][12][13][14] and the study of this article is shown in Table 1.…”

“…Difference(s) from this study Al-Shawabka et al 10 The features we extract ensure that the transmitter's hardware damage can be learned, while the CNN in the literature 10 may only learn the characteristics of the channel. Ying et al 11 Compared with the recognition algorithm in the second stage of SODA in the literature, 11 our algorithm is more suitable for real battlefield signal recognition.…”