| Industry 4.0 and 5G technologies have enabled a large number of smart Io Tterminals to join the network,which brings many security risks while enhancing the convenience and diversity of the network.Access authentication is the first line of defense to ensure network security,however,as smart terminals are mostly miniaturized and low-power,it is difficult to afford the high computational complexity and storage capacity brought by the traditional crypto--based security technology.The authentication method based on physical characteristics is very suitable for edge computing scenarios with a large number of Io Tterminals in close proximity because of its light weight and the burden is concentrated on the authentication side.However,physical feature-based authentication has the following problems due to its use of the randomness and uniqueness of the physical features of the device or channel to authenticate the device: large interference of external noise on the results,high demand for sample size,difficulty in extracting physical features,low authentication accuracy,and low authentication efficiency.To address these problems,the following solutions or improvements are proposed in this dissertation.The optimized coherent integration method can effectively eliminate the random noise in the hardware waveform without additional demand for integration samples,while removing the white Gaussian noise in the channel.Multi-resolution analysis is used to control the computational complexity,the classification accuracy is greatly improved at low SNR.In this dissertation,derivations and experiments are applied to demonstrate a positive correlation between the kernel scale of the optimal Gaussian SVM and the fea-ture dimension of the classification waveform.Given the environment and the equipment used in this experiment,the advantages and disadvantages of the proposed method are compared.The equipment used in the experiments is simple and has a relatively wide applicability.The data augmentation method based on pseudo-random integration borrows from the random integration of DNA in genetic engineering,where data are randomly selected for integration to obtain additional training data.Compared with the traditional method and random integration,pseudo-random integration improves the average classification accuracy and eliminates instability.Hardware simulation experiments validate the per-formance of the method and demonstrate the stable and accurate classification results of the data enhancement method based on pseudo-random integration by comparing the tra-ditional integration method,random integration and pseudo-random integration.In ad-dition,the method can be applied to data enhancement in signal processing and other one-dimensional data processing.The one-dimensional convolution-based signal feature extraction algorithm consid-ers both lightweight and accuracy of terminal authentication in edge computing systems.The waveform variation characteristics of the signal are effectively extracted to improve the authentication accuracy while avoiding the high complexity associated with repeated training optimization of convolutional neural networks.Simulation experiments show that the algorithm improves classification accuracy especailly at low SNR while keeping the training time for machine learning within an acceptable range,thus improving the effi-ciency and security of edge computing access authentication.Cooperative decision for wireless channel access authentication considers dividing a single sample into several groups,assigning weights based on the pre-validation accu-racy of each group,and obtaining the final result by weighted voting.This method can effectively improve the authentication accuracy and successfully get rid of the limitation of requiring multiple receiving devices for voting.Using this method,a single device can complete the entire voting decision.In addition,it is applicable to edge computing sys-tems,which can make full use of edge computing and network resources,while ultimately reducing the time of the access process.The access authentication experiments of mobile scenarios in the factory environment are conducted,and it is proved that the method has significantly improved the authentication accuracy on the basis of the original method.In summary,this dissertation solves a series of problems in the research area of phys-ical layer authentication and optimizes the accuracy and efficiency of authentication.In this dissertation,a large number of simulations and field experiments are designed,and most of the data sets used in the simulations are publicly available,and real scenarios are chosen for the experiments,which are comparable and repeatable.At the same time,the approach in this dissertation is not limited to physical layer authentication,but is also applicable to other fields such as signal processing,and has certain applicability. |
