Intelligent Monitoring And Analysis Of Constellation Diagram Based On Deep Learning In Optical Fiber Communication

The trade-off between greater user bandwidth and more Qo S requirements brings unprecedented challenges to next-generation optical networks.Therefore,Optical Performance Monitoring(OPM)technology has become a common requirement for next-generation autonomous optical networks.The traditional optical performance monitoring technology is more intrusive,lacks the self-extraction ability of core data features,complex monitoring models,limited monitoring accuracy and low monitoring efficiency,complex equipment required,high labor costs,and it is difficult to adapt to the next generation of autonomous light development of the network.Deep learning has strong self-extraction capabilities of core data features,and the monitoring model is clear and simple.Using deep learning technology in the field of optical performance monitoring can help optical networks achieve excellent results in reliability,transmission efficiency,and many other aspects.Therefore,this paper proposes two intelligent constellation monitoring and analysis methods based on deep learning in optical fiber communication,including convolutional neural network and few-shot learning network.When using convolutional neural network for constellation monitoring,different from the previous research on only a few modulation format signals,this paper conducts performance monitoring and modulation format identification for multiple modulation format signals.Another focus of this paper is to propose a network intelligent constellation monitoring and analysis method based on small sample learning.In this method,the network adopts the multi-task learning method of the episode training mechanism,and the training and optimization of the monitoring model can be completed with less data.Quickly adapt to new tasks;and has good mobility,which can be migrated and reused to other monitoring tasks,saving network parameter optimization time.In the previous constellation map detection,conventional large-sample data sets are usually used for model detection and training.However,constellation maps are highly symmetrical information pictures,and it is easier to extract features in the pictures.This feature also makes the use of small sample environments.The training and optimization of the network model becomes possible.Compared with the constellation map recognition and detection under the conventional training model,a lot of time is saved.At present,there are few research works on the recognition of constellation diagrams using small-sample learning algorithms.In this paper,we use the small-sample learning network to conduct in-depth research on the identification and monitoring of constellation diagrams.The research content and results of this paper are as follows:1.A coherent optical communication system with QPSK,8QAM,8PSK,16 QAM and 16 PSK modulation formats with a symbol rate of 25 Gbaud is built for the convolutional neural network analysis model.A coherent optical communication system with BPSK,QPSK,8QAM,8PSK,16 QAM,16PSK,32 QAM,and 64 QAM modulation formats with a symbol rate of 25 Gbaud is built for the small sample learning network analysis model.These two networks are used to realize modulation format identification of various signals,Optical Signal to Noise Ratio(OSNR)estimation,and fiber link transmission distance estimation.The fiber link transmission distance estimation is a rough estimate.The estimated accuracy is 100 km.2.The monitoring and analysis of intelligent constellation diagram based on convolutional neural network is carried out.The modulation format recognition can reach 100% accuracy.When performing OSNR estimation,it can achieve 1 d B estimation effect with 100% accuracy for the above five modulation formats,which is better than K-Nearest Neighbors(KNN)and Artificial Neural Network(ANN)algorithms.When estimating the distance of the optical fiber link,the above five modulation formats can be estimated with at least 94% accuracy,and the estimated distance interval is 100 km.3.The monitoring and analysis of intelligent constellation diagram based on small sample learning network is carried out.When performing modulation format recognition,three OSNR models are built,and the modulation format recognition rate can reach 99.08% in the model with the best effect.When OSNR estimation is performed,2d B estimation effect with 100% accuracy is achieved for the above 8modulation formats.When estimating the distance of the optical fiber link,the three OSNR models built in the modulation format identification are used to estimate the transmission interval of 100 km,and the estimation accuracy can reach more than 90%for the modulation formats except 32 QAM and 64 QAM.