Machine Learning In Optical Networks Research Articles

A survey on QoT prediction using machine learning in optical networks

In optical networks, a connection (e.g., light-path and light-tree) is set up to carry data from its source to destination(s). When the optical signal transmits through the fiber links and optical devices, the quality of transmission (QoT) degrades due to various physical layer impairments (PLIs), including linear and nonlinear impairments. QoT is an important metric that determines the availability of a connection. Therefore, the QoT guarantee is the premise of successful connection establishment in optical networks. QoT prediction before connections establishment can provide guidance for the routing and resources allocation of connections. In order to receive the correct signal at the receiving end, during network planning design margins are introduced to compensate the inaccuracy of the QoT prediction model itself and its inputs. Improving the accuracy of prediction can make better use of network resources and reduce margins. With the help of strong computing power and data acquisition based on software defined optical network (SDON), machine learning (ML) based models are more suitable for QoT prediction than analytical models that are difficult to derive and computationally heavy. This paper provides an overview on the applications of ML technologies in QoT prediction. Firstly, we elaborate the QoT problem in optical networks, including main QoT influence factors, QoT metrics, and QoT prediction strategies. Then, suitable ML algorithms, the generation of sample data, ML frameworks and the construction of QoT prediction model, are briefly introduced. Next, three solutions of QoT prediction using various ML technologies in recent studies and their practical feasibility are reviewed and discussed in detail. Finally, based on the existing researches, we present some future research directions about the improvement of QoT prediction.

Overview on routing and resource allocation based machine learning in optical networks

For optical networks, routing and resource allocation which considerably determines the resource efficiency and network capacity is one of the most important works. It has been widely studied and many excellent algorithms have been developed. However, theoretical analysis shows that routing and resource allocation belongs to the Nondeterministic Polynomial Complete (NP-C) problem no matter in wavelength division multiplexing (WDM) optical networks, elastic optical networks (EONs), or space division multiplexing (SDM) optical networks. At presents, there doesn't exist a polynomial-time algorithm for routing and resource allocation. In recent years, machine learning which shows great advantages in solving complex problems has been widely concerned and researched. Using machine learning to conduct routing and resource allocation has aroused a great interest of researchers. This paper provides an overview on routing and resource allocation based on machine learning in optical networks. At first, we briefly introduce the routing and wavelength allocation (RWA) problem in WDM optical networks, the routing and spectrum allocation (RSA) problem in EONs, and the routing, core, spectrum allocation (RCSA) problem in SDM optical networks respectively. Commonly used machine learning techniques in optical networks are briefly elaborated. Then, the problems of quality of transmission (QoT) estimation, traffic estimation, and crosstalk prediction which can help to routing and resource allocation are also elaborated. The machine learning enabled RWA algorithms, RSA algorithms, and RCSA algorithms are elaborated, analyzed and compared in detail. In addition, the applications of machine learning in the QoT estimation, traffic estimation, and crosstalk prediction, etc., are also elaborated. Based on the existing research results, we present future research directions about how to use machine learning techniques to conduct routing and resource allocation in multidimensional time–space-frequency optical networks and satellite optical networks.

Self-Taught Anomaly Detection With Hybrid Unsupervised/Supervised Machine Learning in Optical Networks

This paper proposes a self-taught anomaly detection framework for optical networks. The proposed framework makes use of a hybrid unsupervised and supervised machine learning scheme. First, it employs an unsupervised data clustering module (DCM) to analyze the patterns of monitoring data. The DCM enables a self-learning capability that eliminates the requirement of prior knowledge of abnormal network behaviors and therefore can potentially detect unforeseen anomalies. Second, we introduce a self-taught mechanism that transfers the patterns learned by the DCM to a supervised data regression and classification module (DRCM). The DRCM, whose complexity is mainly related to the scale of the applied supervised learning model, can potentially facilitate more scalable and time-efficient online anomaly detection by avoiding excessively traversing the original dataset. We designed the DCM and DRCM based on the density-based clustering algorithm and the deep neural network structure, respectively. Evaluations with experimental data from two use cases (i.e., single-point detection and end-to-end detection) demonstrate that up to $99\%$ anomaly detection accuracy can be achieved with a false positive rate below $1\%$ .

Introduction to the JOCN Special Issue on Machine Learning and Data Analytics for Optical Communications and Networking

This Special Issue contains a collection of outstanding papers covering several applications of machine learning in optical communications and networking, including: resource management in elastic optical networks, datacenters and ROADM-based networks; noise and signalquality estimation; and traffic prediction. We provide a brief overview of the current state of machine learning in optical networks, followed by a categorization of the eleven papers in this special issue.