Single-channel blind source separation empowered joint transceiver optimization for wireless communications using deep learning

To tackle the physical layer security challenges in wireless communication, this paper introduces a multi-user architecture that leverages single-channel blind source separation, centered around a Multi-source Signal Mixture Separator (MSMS). This architecture consists of a multi-user encoder, a channel layer, and a separation decoder, allowing it to handle multiple functions simultaneously, including encoding, modulation, signal separation, demodulation, and decoding. The MSMS receiver effectively enables the separation of numerous user signals, making it exceedingly difficult for unauthorized eavesdroppers to extract valuable information from the mixed signals, thus significantly enhancing communication security. The MSMS can address the challenges of few-shot sample training and achieve joint optimization during transmission by employing a deep learning-based network design. The design of a single receiver reduces system costs and improves spectrum efficiency. The MSMS outperforms traditional Space-time Block Coding (STBC) strategies regarding separation performance, particularly in Block Error Rate (BLER) metrics. Modulation constellation diagrams further analyze the effectiveness of multi-source signal mixture separation. Moreover, this study extends the MSMS framework from a two-user scenario to a three-user scenario, further demonstrating the flexibility and scalability of the proposed architecture.