Radio Frequency Fingerprint Identification Of Mobile Devices Based On Deep Learning

Radiofrequency fingerprint refers to the transient or stable characteristics of the radio frequency signal when the wireless transmitter sends information,also known as the radio frequency DNA or the radio frequency metric(RF metric).Radiofrequency fingerprint is like human fingerprint or human DNA,which can uniquely and reliably distinguish network equipment and make up for the defect of no identification means on the physical layer.According to its unique,difficult to clone and stability characteristics,after the development of data collection tools,the selection and processing of RF fingerprint data and the compilation and improvement of recognition algorithm,this paper relies on the RF fingerprint to identify mobile devices,the main work is as follows:(1)A channel state information data acquisition tool based on 802.11 ax protocol is developed.For the feature of RF fingerprint,corresponding data are needed to carry it.The traditional way of collecting channel state information based on 802.11 n protocol is out of date,and there is no maintainer.The acquisition speed is slow,and it is difficult to meet the demand of large amount of data.The tool developed in this paper ADAPTS to the underlying architecture of the sixth generation of Wi Fi chips and meets the requirements of 802.11 AX protocol.Compared with the original acquisition tool,the data acquisition speed is improved by 20 times and the reliability of the tool is greatly improved.(2)The nonlinear phase offset error caused by IQ imbalance in the channel state information was selected as the RF fingerprint of the mobile device,and the RF fingerprint frame was constructed by frame group.General radiofrequency data,information content is small,different equipment radiofrequency fingerprint data overlap rate is high,not suitable for fingerprint identification.In the algorithm,the RF framing mode is proposed to act as the device's RF fingerprint,which enlarges the identification points of the device's RF fingerprint data and improves the model identification capacity.(3)A RF fingerprint identification method combining voting mechanism and improved residual neural network was designed,and it was specifically applied to the RF fingerprint identification task of mobile devices.Different from the general convolutional neural network used for classification task,the algorithm uses Triplet Loss to make the residual neural network act as the feature extractor and construct the feature space to achieve the RF fingerprint matching,which makes the recognition accuracy of mobile devices reach 94%.