Research On Security Technology Based On End-to-End Learning Communication System

A communication system based on end-to-end learning designs the entire system in a globally optimized manner,thereby enabling the system to obtain a higher performance upper limit.In the communication system based on end-to-end learning,symbols are represented based on channel characteristics,and the parameters of deep learning network are highly complex.These characteristics make it have natural endogenous security characteristics,which can be used for the design of secure transmission and security authentication.The main work of this article includes the following aspects:1.Propose a secure transmission scheme and a corresponding performance evaluation system in an end-to-end learning communication system.For the end-to-end learning communication system,redesign the loss function and training method of the system,so that the training optimized communication system can only work on the predetermined main channel.system.In addition,the security of the system is evaluated by counting the error rate of different coders in different decoding methods under different a priori information.The experimental results show that under a few types of a priori information settings,the error rate is less than 50%.2.Propose security authentication scheme and corresponding performance evaluation system in end-to-end learning communication system.Based on the complexity of the neural network model,the neural network model is extracted with the characteristics,and the output range of the transmitting end based on the end-to-end learning communication system is used as the digital signature of the transmitting end.Then,a neural network authentication model is constructed using the signature pair.The transmitting device is identified.In addition,the randomness and robustness of this digital signature were systematically tested,proving its feasibility as a secure authentication signature.Aiming at the attack of the illegal person that the neural network authentication model may suffer,the cost of the power consumption cost of the illegal person's prior information is not analyzed.3.A model structure and training method based on end-to-end learning communication system under multi-user communication conditions are proposed,and corresponding secure transmission and security authentication schemes are designed.In a multi-user communication scenario,a communication system is established based on a noise reduction autoencoder,and the system is trained and optimized in an end-to-end learning manner.A many-to-one relationship between the encoding method and the decoder is built to enable different users It can maintain the uniqueness of the encoder and the reliability of the decoder,that is,all users can share the same decoder to achieve reliable decoding.Furthermore,for a communication system with end-to-end learning under multi-user conditions,the secure transmission and security authentication schemes under single-user conditions are extended to multi-user communication systems,and the performance limit of system security authentication schemes under extreme conditions is pointed out.