Machine Learning Techniques for Optical Performance Monitoring From Directly Detected PDM-QAM Signals

Abstract

Linear signal processing algorithms are effective in dealing with linear transmission channel and linear signal detection, while the nonlinear signal processing algorithms, from the machine learning community, are effective in dealing with nonlinear transmission channel and nonlinear signal detection. In this paper, a brief overview of the various machine learning methods and their application in optical communication is presented and discussed. Moreover, supervised machine learning methods, such as neural networks and support vector machine, are experimentally demonstrated for in-band optical signal to noise ratio (OSNR) estimation and modulation format classification, respectively. The proposed methods accurately evaluate optical signals employing up to 64 quadrature amplitude modulation (QAM), at 32 Gbaud, using only directly-detected data.

An Overview on Application of Machine Learning Techniques in Optical Networks

“…Additionally, a data-driven approach using a machine learning technique, Gaussian processes nonlinear regression (GPR), is proposed and experimentally demonstrated for performance prediction of WDM optical communication systems [49]. The core of the proposed approach (and indeed of any ML technique) is generalization: first the model is learned from the measured data acquired under one set of system configurations, and then the inferred model is applied to perform predictions for a new set of system configurations.…”

“…Artificial neural networks are well suited machine learning tools to perform optical performance monitoring as they can be used to learn the complex mapping between samples or extracted features from the symbols and optical fiber channel parameters, such as OSNR, PMD, Polarization-dependent loss (PDL), baud rate and CD. The features that are fed into the neural network can be derived using different approaches relying on feature extraction from: 1) the power eye diagrams (e.g., Q-factor, closure, variance, root-meansquare jitter and crossing amplitude, as in [49]- [53], [69]); 2) the two-dimensional eye-diagram and phase portrait [54]; 3) asynchronous constellation diagrams (i.e., vector diagrams also including transitions between symbols [51]); and 4) histograms of the asynchronously sampled signal amplitudes [52], [53]. The advantage of manually providing the features to the algorithm is that the NN can be relatively simple, e.g., consisting of one hidden layer and up to 10 hidden units and does not require large amount of data to be trained.…”

Artificial intelligence (AI) methods in optical networks: A comprehensive survey

“…Deep neural networks (DNN) Neural networks [52]: uses DNN, trained with asynchronously sampled raw data, for OSNR monitoring. [53]: uses neural networks based nonlinear regression for OSNR estimation.…”

“…Principal component analysis Support vector machines (SVM) Clustering k-means [41]: identifies modulation formats/bit rates from a known set. [53]: classifies modulation formats using the variance of eye opening width. [40]: identifies modulation formats based on the number of levels and clusters in constellation diagram.…”

“…Yet, this DNN needs to be configured with at least 5 layers and needs to be trained with 400,000 samples to achieve accurate results, requiring long training time. Alternately, Thrane et al [53] propose an OSNR estimator and a modulation format classifier for systems employing advanced modulation formats (up to 64 QAM) and direct detection. The OSNR estimator employs a neural network, while the modulation format classifier uses a support vector machine (SVM), both in order to learn a continuous mapping function between input features extracted from the power eye-diagram after the photodetector and the reference OSNR and modulation format, respectively.…”