Employing Pretrained Networks for Reducing the Latency and the Complexity of a Radio Frequency Fingerprinting System

This study proposes an approach employing a number of trained networks for radio frequency (RF) fingerprinting to reduce the latency and the complexity of the identification system. The networks are trained using identification signals with different durations which are determined based on their SNR levels. RF fingerprints are obtained from these signals by extracting features. In the testing stage of RF fingerprinting, for each test signal, one of the pretrained networks is selected according to the estimated SNR value of the received signal and classification is carried out by this network. We utilize an experimental dataset comprising IEEE 802.11n signals captured from 20 Wi-Fi transmitters to evaluate the performance of the proposed approach. The experimental results demonstrate that, instead of relying on a long fixed-duration, we can utilize pretrained networks to achieve lower latency and reduced computational complexity while maintaining high classification accuracy.