Deep learning approach for physical-layer security in Gaussian multiple access wiretap channel

Deep learning (DL) has exhibited great potential in communication systems. Recent advances in DL-based physical-layer techniques have shown that the communication system can be modeled as an autoencoder (AE), which performs end-to-end learning tasks. In this article, we investigate an AE-based deep learning framework for physical-layer security where multiple transmitters send their own data to a common receiver under an eavesdropping scenario (i.e., Gaussian multiple access wiretap channel). We have newly designed an integrated loss function with respect to secrecy performance in terms of symbol error rate among multiple users. Further, we verify that our training approach based on the proposed loss function can achieve better secrecy performance compared with the conventional training one. & COPY; 2022 The Authors. Published by Elsevier B.V. on behalf of The Korean Institute of Communications and Information Sciences. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Automated summary

This article investigates the application of an autoencoder-based deep learning framework in physical-layer security and the secrecy performance among multiple users in an eavesdropping scenario. By designing an integrated loss function and training based on this function, it is verified that our proposed training approach achieves better secrecy performance compared to conventional methods.