Meta-ensemble learning for OPM in FMF systems

Abstract

Optical performance monitoring (OPM) is crucial for facilitating the management of future few-mode fiber (FMF)-based transmissions. OPM deploys fault detection and link diagnosis by measuring the physical layer states and provides feedback to the controller. Recently, machine learning (ML) has gained a lot of attention for OPM, and various ML algorithms were developed, wherein the selection of the proper method is a challenge. Ensemble learning (EL) solves this challenge by combining different ML models; however, this simultaneous employment suffers from increased complexity and dependency on the performance of each individual model. Meta-ensemble learning (MEL) provides a promising solution by intelligently selecting the proper ensemble at each instance. In this work, we employ MEL for OPM in FMF systems. We compare the proposed MEL-based OPM method with naive EL (NEL), which is a well-known EL method. The obtained results indicate that proposed MEL-based OPM method provides better performance with the loss data set size compared with NEL-based OPM. Furthermore, the proposed MEL-based OPM method does not need the feature preprocessing, which is an essential step in other ML algorithms such as NEL-based OPM.