DCGAN-based symmetric encryption end-to-end communication systems

In this article, we propose a symmetric encryption end-to-end communication system based on deep con-volutional generative adversarial networks to solve the security problem of transmission in end-to-end learning-based wireless communication systems. The system generates a key by a deep convolutional generative adversarial network and shares it to the transmitter and receiver. Both the transmitter and receiver are represented by convolutional neural networks. We propose to use neural networks as a bridge to make an irreversible mapping relationship between the message and the key. The proposed method achieves more secure message transmission than the symmetric encryption end-to-end communication systems using randomly generated keys. We also propose a concept of the error impact factor to explain the system's characteristics. From the simulation results, the proposed method can achieve encryption transmission of arbitrary length bit messages. The system has excellent performance on additive white Gaussian noise channels and Rayleigh fading channels. The legitimate parties can establish secure transmission in this setting, and the illegal eavesdropper cannot decode accurately without the key.