In this paper, we propose an asymmetric encrypted end-to-end communication system based on convolutional neural networks to solve the problem of secure transmission in the end-to-end wireless communication system. The system generates a key generator through a convolutional neural network as a bridge. The private and public keys establish a key pair relationship of arbitrary length sequence information. The transmitter and receiver consist of autoencoders based on convolutional neural networks. For data confidentiality requirements, we design the loss function of the end-to-end communication model based on a convolutional neural network. We also design bugs based on different predictions about the information the system eavesdropper has. Simulation results show that the system performs well on additive Gaussian white noise and Rayleigh fading channels. A legitimate party can establish a secure transmission under a designed communication system; an illegal eavesdropper without a key cannot accurately decode it.

Most of the existing physical layer watermarking authentication schemes are based on a single-input single-output system and require pre-issue of shared keys. To address these problems, in this thesis, a physical layer authentication scheme without the distribution keys is proposed based on the constellation dithering physical layer authentication watermarking mechanism with a multiple-input multiple-output (MIMO) system, and space-time block coding (STBC) is used to improve the robustness of transmission. Specifically, the legitimate node obtains channel state information (CSI) through channel probing and couples CSI with the message signal using a hash function to generate an authentication tag, which is then embedded through constellation dithering. The receiver extracts the tag and authenticates it using hypothesis testing. Performance analysis shows that the scheme is resistant to various attacks such as replay, interference, tampering, and forgery. Simulation results show that the use of MIMO multi-antenna diversity with STBC coding technique reduces the bit error rate (BER) of message signals and tag signals and improves the detection rate of legitimate signals.