Research On Security Authentication Algorithm Based On Physical Features

Wireless network technology evolves to thoroughly integrate with multiple fields,to gradually realize intelligent connection,holographic connection,deep connection,and ubiquitous connection.However,due to the broadcasting and openness of wireless networks,wireless networks are extremely vulnerable to malicious attacks such as eavesdropping and interference.At the same time,the popularity of micro wireless devices makes these micro-terminals very limited in computing and energy.Traditional security mechanisms are difficult to apply to wireless networks with low power consumption,low computing resources,and high security requirements under high throughput.The wireless physical layer security technology covered in this work provides new ideas for solving these security problems.The physical layer security technology doesn't rely on complex encryption calculations at the upper layer,and makes full use of the characteristics of channels and devices,thereby realizing low-complexity,lightweight secure transimission and authentication.Due to the unclonability of the radio signal fingerprint and the uniqueness of channel fingerprints in space-time domain,the device identity authentication mechanism and data packet authentication mechanism can meet the needs of light-weight access authentication and data packets for massive wireless terminals.At present,the physical layer security technology has attracted attentions of academia and industry.It has potentials to be one of the key technologies for effective equipment and data access authentication and to prevent man-in-the-middle attacks and malicious node attacks.High authentication recognition rate and lower computational complexity are key factors that promote the practical application of this technology.This dissertation intends to combine the physical feature extraction and machine learning methods to improve the performance of physical feature security authentication.The specific research content and contribution are summarized as follows:1.To address difficulties reside in the selection of the characteristics of radio signal fingerprints,this dissertation analyzes radio frequency fingerprints of wireless ultra-low power integrated circuits with ultra-low power based on samples in the time and frequency domains,and proposes a variety of fingerprint feature vector construction methods.At the same time,this dissertation proposes a radio frequency fingerprint recognition algorithm based on the feature vector of the quadratic discrete wavelet approximation coefficients combined with machine learning naive Bayes.We compared its performance with the performance of K-nearest neighbor and shallow neural network algorithms.The numerical results show the new algorithm has good noise resistance.For instance,the proposed algorithm can still achieve good authentication results in a low signal-to-noise ratio regime(such as SNR=0).Besides,to facilitate the feature extraction of a radio signal,which is manually selected in existing works,we propose a radio frequency fingerprint identification algorithm based on time-frequency analysisconvolutional neural network to automatically obtain blind features to simplify the complexity of manual feature selection.Finally,the performance of the algorithm against noise is further analyzed through experiments.2.In current works,the statistics of test for channel fingerprint authentication technology cannot be established.In this dissertation,we propose an adaptive enhancement learning method for channel fingerprint data packet authentication.With the aim to improve the accuracy of data packet authentication,the method first combines the constructed inspection statistics and machine learning algorithms to avoid traversal searching for the artificial optimal threshold.The proposed approach can realize the adaptive judgment of legal data packets.Secondly,this dissertation proposes a twodimensional fusion feature channel fingerprint data packet authentication method,which combines two inspection statistics to improve the feature dimension of machine learning and obtain a higher authentication rate.Finally,the validity of the proposed authentication scheme is verified by setting up a wireless communication platform in a real environment to conduct a three-party authentication simulation experiment.The results show that the channel fingerprint authentication method based on adaptive enhancement learning has better performance than the artificial traverse threshold method and has a higher accuracy rate.3.To tackle the limitation of current physical feature methods,which cannot detect the cloning attack and Sybil attack simultaneously,this dissertation proposes a physical layer malicious node detection method based on identity ID-channel fingerprints.This method firstly distinguishes the existence of cloning attack and Sybil attack by judging whether the identity ID is declared by the node conflicts.Furthermore,the space-time uniqueness of the channel fingerprint and the difference caused by the attack characteristics of different malicious nodes are comprehensively used to realize the detection of malicious nodes and legitimate nodes.In order to further improve the detection rate of malicious nodes,this dissertation combines machine learning algorithms to propose a physical layer detection method based on automatic labeling and learning,and solves the problem of lack of labeled learning samples in the offline training process of machine learning algorithms.Through simulation based on the open channel test data set provided by the National Institute of Standards and Technology(NIST),as well a s attack detection experiments in a real industrial environment,the feasibility of the proposed scheme is proved.4.To realize the efficient training of physical feature authentication using machine learning,this dissertation proposes two edge computing collaborative physical feature authentication methods in the typical application scenario of physical feature authentication-terminal physical feature authentication under edge computing.The computing tasks between the edge computing side and the terminal are optimally allocated to achieve less time delay and faster speed in training and computing.In addition,the proposed collaborative identity authentication method based on radio frequency fingerprint identification can coordinate computing tasks to multiple terminal devices and edge terminals to achieve fast training and real-time access authentication according to the theoretical model.The method can realize low-latency data packet authentication through collaborative authentication of a single terminal device and the edge end.In order to verify the accuracy and timeliness of the above two security authentication methods,the dissertation finally carried out corresponding theoretical analysis and modeling,and proved the feasibility and effectiveness of the method using numerical simulation.