Deep learning based RFF recognition with differential constellation trace figure towards closed and open set
Tianwen Yang, Jianing Zhao, Xin Wang, Fei Xu
Abstract
Although radio frequency fingerprint (RFF) has been widely adopted in the field of wireless communication equipment identification, most researches are focused on closed set recognition. Considering the open set recognition challenge faced by RFF recognition with the rapid increase of wireless devices, this paper proposes a deep learning based RFF recognition framework towards closed and open set. RFF caused by the analog parts of the transmitter is extracted based on differential constellation trace figure (DCTF) algorithm, and then preprocessed to enhance features. By utilizing SoftMax and OpenMax algorithms, a deep learning based RFF recognition network with ResNet framework towards closed and open set is built to recognize the extracted RFF features. Simulation results show that the proposed deep learning based RFF recognition framework achieves an accuracy of more than 99% in closed set when SNR reaches 20dB. In open set simulation experiment, the recognition accuracy of the proposed RFF recognition method with OpenMax is superior to that of the traditional closed set framework by more than 13% when the openness is greater than 0.074.