Research On PHY Fingerprints Extraction And Identification Of Fast Ethernet Devices

Nowadays,Ethernet communication technology has become an indispensable part of human life,and as more end devices are connected to the network,identity authentication has gradually become an important part of network security.However,the commonly used MAC address or digital certificate based device authentication technology has certain security flaws,and these digital credentials can be easily copied or forged by malicious users to carry out counterfeit attacks.To address the traditional security threat scenarios,this paper conducts research on NIC fingerprint extraction and identification technology.The NIC fingerprint is a hardware characteristic formed by the difference between the electronic components of the NIC itself,which has the same characteristics of non-clonable and uniqueness as the biological fingerprint,and can be used in the NIC authentication technology to obtain higher security.In the LAN communication technology,Ethernet series communication technology occupies most of the market,and the existing NIC fingerprint extraction methods are all oriented to 10 M transmission rate Ethernet NIC.However,the existing fingerprint extraction methods are limited by the signal format,transmission rate,transmission protocol and sampling rate of the signal acquisition device,which makes them not applicable to the fingerprint extraction of NICs in higher speed scenarios.In addition,the test equipment used in the current study is small and the test scenarios are not rich enough,and the use of Ethernet NIC fingerprinting lacks practice.To address the limitations of the current study,this paper conducts further research on Fast Ethernet NIC fingerprint extraction and identification with the following main works.(1)Given the existence of scrambling processing in Fast Ethernet signal protocols,a data-independent adaptive filter-based fingerprint extraction method is proposed for this scenario.The NIC fingerprint is modeled as a filter,the original ideal signal is reconstructed by the actual acquired signal,and finally the fingerprint of each NIC is obtained using the LMS algorithm for filter coefficient estimation.The signal data set containing 25 NICs is acquired with consistent MAC address settings,and the fingerprint extraction method can obtain 96% test accuracy under the MDA/ML classification algorithm,while the traditional fingerprint extraction method based on digital features can only achieve a maximum test accuracy of 66%.(2)Using the condition that there are idle data segments in Fast Ethernet signal frames,this paper proposes a fingerprint extraction method based on periodic data using the periodicity of scrambled code processing,and uses the amplitude spectrum,autocorrelation power spectrum and inter-correlation power spectrum as NIC fingerprints,respectively.In the case of using the same signal dataset,these three spectrum-based fingerprint feature vectors are longer compared to the adaptive filter-based extraction methods,and the test accuracies exceed 98% in all scenarios using MDA/ML classification methods.(3)To address the problem of insufficient experimental scenarios in the current study,the differences of fingerprint extraction under different quantization accuracies,the stability of fingerprints and the influence of line characteristics on fingerprints were tested separately.The experimental results show that the fingerprint recognition accuracy of the network card extracted in this paper does not drop significantly due to the reduction of the quantization accuracy of the acquisition equipment,and has good robustness,with the test accuracy higher than 90%;in the experiment of testing the line characteristics on fingerprints,the fingerprints are affected when the line characteristics change drastically,but the line interference problem can be solved initially by distinguishing the lines subsequently to ensure that the fingerprint recognition rate accuracy can still be higher than 90%.(4)A fast Ethernet NIC identification system based on NIC fingerprint is designed and implemented,and tested in practical application scenarios.The system uses a splitter and the AD9484 evaluation kit to acquire NIC signals.During the identification of multiple identical NIC models from the same manufacturer,the average identification accuracy of NICs under the fingerprint extraction method based on autocorrelation power spectrum and the MDA/ML classification method reaches over 90%,which proves its strong practicality.