Research On Key Issues Of Optical Performance Monitoring In Elastic Optical Networks Based On Machine Learnin

The landscape of data transmission has changed a lot while the optical communication capacity grows rapidly,driven by emerging data technologies such as Digital Twin,Metaverse,and Industrial Internet.In order to meet the complex and diverse network requirements,the next generation optical communication system needs to sense the transmission quality of link signals in real time and monitor the physical state of transmission devices,and dynamically adjust the transmission rate,modulation format,transmission power and other physical parameters of transmission signals according to the channel status and service bandwidth,so as to establish optical connections with high spectral efficiency and high transmission performance.However,the dynamic optical fiber communication link greatly increases the complexity of optical network and affects the accuracy of physical layer parameter monitoring.Focusing on optical performance monitoring techniques in elastic optical networks,this paper aims to reduce the computational complexity of optical performance monitoring schemes and improve the accuracy of multi-parameter monitoring based on machine learning.A Stokes sectional images-based modulation format recognition method with a generative adversarial network and an optical performance monitoring scheme based on confrontational knowledge distillation with lifelong learning in seven-core fiber are proposed innovatively.The main innovative work of this paper is as follows:1)In order to improve the performance of the modulation format recognition algorithm of the DSP module in the coherent communication system,this paper proposes a modulation format recognition scheme based on Stokes sectional image with a generative adversarial network and establishes a relevant experimental system for verification.This scheme avoids transmission link interference by generating Stokes sectional feature images.Furthermore,the number of training samples required for modulation format recognition is also significantly reduced via the feature extraction and data generation capabilities of the generative adversarial network in this paper.The proposed scheme could recognize five signals’ modulation format with an input power of-3 dB and transmission rate of 12.5 Gb/s on a coherent transmission platform based on a 25 km single-mode fiber.Compared to traditional convolutional neural networks,the training dataset is reduced by 33 percent.2)For sake of realizing the lightweight optical performance monitoring of multi-channel in the elastic optical network based on spatial division multiplexing,this paper designs an optical performance monitoring scheme based on lifelong learning with confrontational knowledge distillation in 7-core fiber and establishes a relevant experimental platform for verification.In this scheme,the transverse and vertical sectional images of the Poincaré sphere are chosen as the recognition features and the complex network model is trained by lifelong learning to ensure accuracy for all channel optical performance monitoring tasks.Knowledge in the complex network model is transferred to the lightweight network via confrontational knowledge distillation technology,greatly reducing the computing time.Finally,one lightweight neural network is used to perform the optical performance monitoring tasks for all channels of the 7-core fiber.The proposed scheme is able to identify the five modulation formats successfully for each channel in a coherent transmission system based on a 2 km weakly coupled 7-core fiber with a root-mean-square error of less than 0.1 for the optical signal-tonoise ratio.