Enhancing Performance and Quality of Transmission Through Knowledge-Driven Machine Learning-Based FWM Mitigation

Abstract

This article proposes a knowledge-driven four-wave mixing (FWM) mitigation using supervised learning approaches and a multilevel regression-based dense wavelength division multiplexing (SMR-DWDM) system design. The evolution of 5G and the Internet of Things (IoT) results in an immense data rate consumption and introduces unprecedented dynamic network traffic. DWDM networks effectively accommodate these challenges by being highly responsive and adaptable to changes in traffic impact and network conditions. High-capacity DWDM transmission causes fiber nonlinearities, reducing system performance and effective bandwidth utilization and affecting Quality of Transmission (QoT) by inducing crosstalk, dispersion, and Inter-Symbol Interference (ISI). This work discusses knowledge-driven DWDM design, utilizing machine learning to improve flexibility, identify FWM parameters, and predict transmission quality. Firstly, machine learning optimizes parameters at the transmitter end to identify FWM monitoring factors, predict QoT based on subscriber requirements, and create a comprehensive database for training Machine Learning (ML) models. Then, supervised multilevel regression builds the knowledge-driven QoT Estimator, accurately selecting input parameter combinations for the automatic monitoring controller of the DWDM system. The accuracy of the proposed SMR-DWDM system is confirmed by validating it with various FWM mitigating factors monitored by Optical Spectrum Analyzer (OSA) and Bit Error Rate (BER) analyzers. Through parametric analysis and supervised multilevel regression, the system achieves high precision and accurately predicts QoT by over 80%, and improves 25% of the QoT enhancement compared with traditional methods, proving its effectiveness in managing fiber nonlinearities.

Keywords

• FWM
• machine learning
• knowledge-driven DWDM
• multilevel regression
• quality of transmission