A study on the performance evaluation of wavelet decomposition in transient‐based radio frequency fingerprinting of Bluetooth devices

Abstract Radio frequency fingerprinting (RFF) is used as a physical‐layer security method to provide security in wireless networks. Basically, it exploits the distinctive features (fingerprints) extracted from the physical waveforms emitted from radio devices in the network. One of the major challenges in RFF is to create robust features forming the fingerprints of radio devices. Here, dual‐tree complex wavelet transform (DT‐CWT) provides an accurate way of extracting those robust features. However, its performance on the RFF of Bluetooth transients which fall into narrowband signaling has not been reported yet. Therefore, this study examines the performance of DT‐CWT features on the use of transient‐based RFF of Bluetooth devices. Initially, experimentally collected Bluetooth transients from different smartphones are decomposed by DT‐CWT. Then, the characteristics and statistics of the wavelet domain signal are exploited to create robust features. Next, the support vector machine (SVM) is used to classify the smartphones. The classification accuracy is demonstrated by varying channel signal‐to‐noise ratio (SNR) and the size of transient duration. Results show that reasonable accuracy can be achieved (lower bound of 88%) even with short transient duration (1024 samples) at low SNRs (0–5 dB).