RF Fingerprinting Based on Contrastive Learning and Convolutional Neural Networks

Hardware differences create a unique fingerprint of the radiation source and are attached to the radio signal, and this unique property of the radiation source can be used for RF fingerprinting. The RF fingerprinting method based on expert experience has excessive prior knowledge and poor robustness in different environments. The RF fingerprint recognition method based on deep learning, especially the method that can directly process Raw I/Q shows great potential, but most current deep learning-based RF fingerprint recognition methods require manual annotation of I/Q data. In this paper, the SimSiam model is used to process the data in the form of self-supervised comparative learning, which greatly reduces the labor cost while ensuring the accuracy. The backbone network uses an optimized convolutional neural network (CNN) for classification recognition, which saves manpower and time while ensuring recognition accuracy.