Learning-Based RF Fingerprinting for Device Identification using Amplitude-Phase Spectrograms

Radio frequency fingerprinting (RFF), a technique based on specific transmitter hardware impairments, has emerged as an effective solution for wireless device identification. In this paper, we present a flexible deep CNN-LSTM for RF feature extraction capable of handling inputs with varying lengths. We construct a channel-independent spectrogram by exploiting the amplitude and phase information of the received RF signals, ensuring the extractor’s resilience to channel variations. To evaluate the performance of the proposed approach, we utilize the open-source LoRa dataset consisting of 60 commercial off-the-shelf LoRa devices and a USRP N210 software-defined radio platform. The experimental results show that classifiers perform better when trained with RF templates generated from amplitude-phase spectrogram than amplitude-only spectrogram. This is due to the additional information present in the amplitude-phase channel-independent spectrogram.