Research On Fingerprint Extraction And Recognition Of Wired Network Card

The network has become an indispensable part of our modern life,and the authentication of access device identity is a vital part of network security.MAC address authentication and device digital certificate authentication are the two current mainstream access authentication methods.However,both methods are authenticated at the software level thus being defective.The MAC address is easily forged,which possesses security risk;the digital certificate is relatively hard to falsified,but it needs to be installed with client software or hardware support,which makes its application inconvenient.Therefore,this thesis has studied scheme to generate the fingerprint information and recognize the identification from network identification card(NIC)from the aspect of hardware level.At present,research on wired network access based on hardware features has no precedent in China.Meanwhile,most foreign studies have two problems: one is the small sample size that makes the results not representative enough;another is lack of discrimination of hardware features,which makes the fingerprint recognition rate relatively low.To solve issues aforementioned,this paper presents experiments that increase the number of NICs to 100,which cover most of the current mainstream brands.Meanwhile,the conducted tests are based on a large amount of NICs that bear similar hardware features from the same manufacturer.In addition,three new NIC fingerprint generation methods are proposed,and corresponding experiments have been performed using machine learning and neural network algorithms to train and identify the generated fingerprints.The main work of this paper is presented as follows:1.Studied the generation mechanism of NIC fingerprint and the related data transmission protocol of wired network.An NIC signal acquisition scheme is designed,which is able to effectively collect signals based on 10Base-T protocol in the wired network.2.This work studied the variation characteristics of level trajectory between NIC signal symbols,proposed a new NIC fingerprint generation method named "symbolic difference constellation trajectory map",and applied deep neural network for training.The false recognition rate for the dataset of 20 NICs is 0.02%,whereas the false recognition rate on the dataset of 100 NICs is 0.16%.3.This work studied the statistical characteristics of frequency offset and amplitude of NIC signal transmission,proposed a new NIC fingerprint generation method named "signal statistical feature vector set",and applied the naive Bayesian algorithm for training.The false recognition rate for the dataset of 20 NICs is 0.05%,whereas the false recognition rate on the dataset of 100 NICs is 0.26%.4.Designed specifically for the NIC signal acquisition features at low sampling rates,a new NIC fingerprint generation method named “signal amplitude distribution histogram”.This method does not require signal decoding,which makes the implementation cost relatively low.It can be applied when the dataset size is small.The false recognition rate for the dataset of 10 NICs is 0.97%.5.An access control scheme based on NIC fingerprint is designed.The scheme is composed of a signal acquisition module,a fingerprint extraction module,a fingerprint identification module and an access control module.It can automatically acquire the terminal NIC fingerprint,and serve as a decision basis for accessing the internal network.