Field Trial of Machine-Learning-Assisted and SDN-Based Optical Network Management

Abstract

In this paper, we reported machine-learning based network dynamic abstraction over a field-trial testbed. The implemented network-scale NCMDB allows the ML-based quality-of-transmission predictor abstract dynamic link parameters for further network planning.

ANN-Based Multi-Channel QoT-Prediction Over a 563.4-km Field-Trial Testbed

“…The ANN-based OSNR prediction were evaluated over a field-trial testbed [18]. The results showed that ANN-based regression model achieved a high accuracy that the mean square error (MSE) is less than 1 dB.…”

“…Currently, most of the works mentioned above focused on QoT estimation of single channel, the relevant algorithms have achieved high accuracy for both classification and regression. For example, most of ML-based regression models for OSNR or Q-factor predictions in [18][19][20][21] have achieved the high accuracy for a single channel within at most 1 dB and down to 0.02 dB error, however, the real scenarios of optical systems are generally multi-channel transmission systems. The physical parameters of optical opponents have not been considered in most cases, these physical parameters like EDFA parameters have deep influence on the performance of transmissions.…”

“…ML-based QoT classification was proposed to predict whether the QoTs of channels exceed the pre-defined QoT thresholds without providing real values for predictions; MLbased QoT regression was investigated to predict the exact value of Optical Signal Noise Ratio (OSNR) or Q-factor, which can provide more details than the classification [13]. Some progress has been made in ML-based QoT estimations, various Machine Learning (ML) algorithms such as Support Vector Machine, Ensemble Learning methods, and artificial neural network (ANN) have been applied in QoT prediction [14][15][16][17][18][19][20][21][22][23][24][25][26][27][28][29]. However, there are several issues that prohibit the deployment of ML-based models for practical networks.…”

Dynamic Abstraction of Optical Networks with Machine Learning Technologies

“…However, failure to handle dynamics in optical device will lead to inaccuracy of QoT estimations in long term. In [15,16,21], we built a cloud network configuration and optical performance monitoring database to collect the physical parameters over the whole network. ANN-based QoT prediction was developed to predict a single channel.…”

An overview of ML-based applications for next generation optical networks

“…In [21], Gaussian process regression is used to estimate the optical signal-noise-ration (OSNR) of the link, and the mean-square error (MSE) is only 0.7 dB. In [22], an artificial neural network (ANN) is utilized to predict the OSNR. As a novel planning framework, this ML-based prediction tool and the software-defined network (SDN) controller are cooperated to adapt the actual network states.…”