Research On Physical Layer Authentication Based On Deep Learning

with the rapid development of wireless communication technology,the transmission of a large number of confidential information in wireless channels brings the security requirements to communication systems.Because of the openness of the communication channel and broadcast characteristics,wireless communication technology is facing with many security problems.Before5 G,the traditional security authentication is the main encryption technology built on the physical layer.with the development of network,explosive growth of access devices and increasing demand for computing power,the traditional authentication scheme has appeared many deficiencies.The scheme Physical Layer Authentication(PLA)can make use of many characteristics of wireless channel to achieve authentication with simpler principle and more reliable performance.Meanwhile,machine learning technology has a good effect on the classification and prediction of massive data,and can solve the problems of wireless channel estimation and channel information data processing.Therefore,in order to make full use of the characteristic attributes of wireless channel,this thesis focuses on the physical layer security authentication scheme based on deep learning.The main contents are as follows:(1)for the rician fading channel model,an improved channel estimation method based on DFT is proposed by using various channel estimation algorithms,and the statistical characteristics of channel estimation are derived.The result shows that this method can effectively reduce the error of channel estimation.At the same time in order to make full use of the characteristics of wireless channel,this thesis improved feature extraction technology and abandoned the authentication scheme using a single channel characteristics,explores the channel coherence time,the combination of channel frequency response and multi-dimensional channel characteristics,properties in the channel estimate the presence of small error,the extraction has high robustness of three-dimensional channel characteristic vector.Combination of multi-dimensional channel features.Experimental results show that the method based on the improved DFT channel estimation can effectively reduce the error of channel estimation,and the 3d channel feature vector extracted by LSTM feature extraction technology can effectively improve the accuracy of physical layer authentication.(2)Among numerous machine learning algorithms,this thesis proposes a physical layer security authentication scheme based on convolutional neural network in deep learning algorithm.The convolutional neural network’s convolutional layer,pooling layer and full connection layer are designed in detail.The structure of two convolutional units is adopted,and the convolutional data is evenly pooled.Meanwhile,the momentum gradient descent method is used to update and train the parameters of the network.Finally,other physical layer authentication schemes based on LSTM are compared with,and the authentication performance and efficiency are compared with and analyzed.Experiments show that the physical layer security authentication scheme based on convolutional neural network can make full use of the characteristics of wireless channel and has good authentication performance.(3)Considering that the training of deep convolutional neural network requires a large number of training samples,this thesis studies the physical layer authentication scheme based on data enhancement.Considering that the classification prediction schemes based on machine learning algorithm are often extremely dependent on the quality of training data set,the physical layer authentication scheme based on data enhancement algorithm is researched and improved.By strengthening the correlation between channel information sample sets,a weighted random data enhancement algorithm is proposed to construct new samples by adding random weights to the original samples.Experiments show that physical layer authentication based on data enhancement algorithm enhances the robustness of training data set,speeds up model training,and has high authentication efficiency.