End-to-End Learning-Driven Dual Independent Key Generation and Distribution for Optical Network Physical Layer Security
Yanwen Zhu, Yixin Wang, Yuang Li +6 more
To address the urgent demand for secure optical fiber communication, a novel dual independent key generation and distribution scheme for Optical Network Physical Layer Security is proposed for the first time. End-to-end (E2E) learning with a Transformer algorithm featuring adaptive positional encoding is introduced to accurately model coherent fiber transmission systems, achieving an R-squared value exceeding 0.95 for waveform fitting and data distribution. The error vector phase (EVP) and amplitude noise (AN) of the E2E system are utilized to comprehensively characterize channel uniqueness, reducing system reconstruction risks while serving as sources for highly random key sequences for legitimate parties. Key consistency stems from channel reciprocity and precise E2E modeling, with feasibility analyzed across diverse link conditions. Leveraging the fact that eavesdroppers cannot access the system's full lifecycle data, keys are extracted from outputs of transfer learning (TL)-updated E2E models to enhance security. Security analysis under various eavesdropping scenarios demonstrates resistance to attacks. Experimental results show that, with FPGA-implemented post-processing, error-free key distribution is achieved over 120 km SSMF at EVP and AN key rates of 306.74 Mbps and 521.82 Mbps, respectively.
Journal of Lightwave Technology•Jan 13, 2026•0 citations