Multi-Span Long-Haul Fiber Transmission Model Based on Cascaded Neural Networks With Multi-Head Attention Mechanism

• Published in: Journal of lightwave technology Vol. 40; no. 19; pp. 6347 - 6358
• Main Authors: Zang, Yubin, Yu, Zhenming, Xu, Kun, Chen, Minghua, Yang, Sigang, Chen, Hongwei
• Format: Journal Article
• Language: English
• Published: New York IEEE 01-10-2022; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Cascaded neural networks; Data models; long-haul fiber transmission; multi-head attention mechanism; Neural networks; Optical fiber amplifiers; Optical fiber devices; Optical fiber dispersion; Optical fiber networks; Predictive models; Quadrature amplitude modulation; Signal transmission; Training
• ISSN: 0733-8724; 1558-2213
• DOI: 10.1109/JLT.2022.3195949

In this paper, a novelty fiber transmission model consisting of cascaded neural networks with the multi-head attention mechanism has been put forward to solve signal transmission prediction problems in multi-span long-haul fiber link. After appropriately training the model with collected data by the gradient descent method, it can gradually handle the rules of signals' changes over each span and predict the signal transmission results with notably low time cost compared with traditional split-step Fourier method based long-haul model. Through numerical demonstration, this new model can predict 1000 km fiber link with symbol rate up to 40GBaud in the QAM modulation with extremely low predicting error.