Feature Extraction And Recognition Of Radio Frequency Fingerprint Based On Deep Residual Shrinkage Network

The current development of wireless communication technology and the Internet of things engineering technology in a day,in this under the time background of risks and opportunities coexist,wireless network security hidden danger is increasingly apparent.Radio-frequency fingerprint technology is an important subject in the field of wireless network security.It has been widely used in the military field to track and identify specific transmitters.In the civil field,radio-frequency fingerprint technology can be used to prevent malicious attacks,fault detection,etc.,and is an aid and enhancement to realize wireless network security.Due to the subtle hardware differences in the manufacturing and production process of each transmitter,the RF fingerprints generated by the transmitter are different and unique.Therefore,each transmitting machine has its own unique fingerprint in the field of physical layer,which can be used for the identification of transmitter equipment.Radiofrequency fingerprint technology is to extract transmitter signal from physical layer and realize specific transmitter identification technology.For the research of rf fingerprint feature extraction and recognition technology to enhance the security system of social network and recognition mechanism of optimization,and take advantage of the forerunner in the field of national electronic strategy has important and far-reaching significance,as well as the leading advantage in the field of national electronic strategy.This paper takes wireless network security as the core and radio-frequency fingerprint as the entry point to carry out research based on radio-frequency fingerprint feature extraction and recognition technology.The main innovations include:(1)A radio frequency fingerprint feature extraction and recognition method based on axial integral bispectrum and depth residual shrinkage network is proposed.At present,most RF fingerprinting technologies based on deep learning lack professional preprocessing for data signals,which leads to the low accuracy of neural network classification and recognition.In order to increase the accuracy of the feature extraction and recognition,restrain gaussian noise and increase the pretreatment of the data set,this paper puts forward a kind of based on the axial integral bispectrum features and the depth of the residual shrinkage network integration of specific emitter feature extraction and recognition method.First three points for common double spectrum experiment analyzed respectively,and finally chose the best effect of the axial integral bispectrum.Secondly using axial integral bispectrum preprocessing,in the depth of the residual shrinkage network for soft threshold denoising signal.The experimental results show that the method under the real data sets can effectively improve the classification accuracy of low signal-to-noise ratio,under the condition of low signal-to-noise ratio,maximum distance of 62 ft can realize the classification accuracy of 98.5%.(2)A method of radio frequency fingerprint feature extraction and recognition based on coordinate attention mechanism is proposed.The common radio frequency fingerprinting technology,based on deep learning to extract the characteristic vector of lack of diversified dimension,often fitting problems such as are in the experiments,the results in the decrease of classification effect.To solve above problems,this paper proposes a coordinate attention mechanism based on rf fingerprint feature extraction and recognition method.First RSBU module internal attention mechanism optimized for better attention mechanism,the coordinates of generating new RSBU-CA modules.Then embedding RSBU-CA DRSN generated DRSN-CA network,finally under different number of transmitter,different SNR circumstance,sending and receiving distance under the classification performance of the experiment summary.The experimental results show that,the effect of DRSN-CA is relatively good,and the traditional rf fingerprint feature extraction and recognition method,compared the 0 db under low signal-to-noise ratio,maximum distance of 62 ft can realize the classification accuracy of 92.5%.