Radio Frequency Fingerprinting Method Based on Squeeze-and-Excitation Attention Multisource Domain Adversarial Network

The radio frequency (RF) fingerprint uses the hardware features of the RF front-end as identifiers to identify the device. To solve the problem of existing RF fingerprinting methods ignoring the interference of receiver hardware features, which results in poor generalization of the model to different receiver devices, an RF fingerprinting method based on squeeze-and-excitation (SE) attention multisource domain adversarial network is proposed. In this method, adversarial training is carried out using multiple source domain labeled data and a small amount of target domain labeled data to extract features that are not related to the receiver domain. The SE attention mechanism is integrated to enhance the model's learning ability for the transmitter's RF fingerprint features. Combined with a very small amount of labeled data in the target domain, the model parameters are fine-tuned to further improve the recognition performance of the transmitter. Experimental results on the WiSig public dataset show that the proposed method can effectively identify the sender device in the cross-receiver scenario, with an average accuracy of 83.3%. The average accuracy can be further improved to 93.4% after fine-tuning with a small amount of labeled data.