Secure Communication in Gaussian Multiple Access Wiretap Channels: A Deep Learning and Friendly Jamming Approach

The use of deep learning (DL) in communication systems shows great promise, particularly through DL-based physical-layer techniques with autoencoders (AEs) for end-to-end learning. This paper presents an AE-based DL framework to enhance physical-layer security in scenarios where multiple transmitters communicate with the receiver under eavesdropping threats, specifically within a Gaussian multiple-access wiretap channel. A key feature is a friendly jammer that emits a high-power Gaussian signal to disrupt eavesdroppers. The proposed framework is particularly relevant for security-critical applications such as wireless health monitoring systems, where safeguarding sensitive data is paramount. We assess secrecy performance by analyzing the symbol error rate among users in the presence of both an eavesdropper and a jammer. Simulation results show that our DL-based Gaussian jamming strategy significantly improves secrecy performance, effectively safeguarding communications from eavesdropping. This study highlights the potential of DL techniques to enhance communication security in complex multi-user environments.