Multi-labeled Random-forest Enabled Softwarized Management for Photonics Switching Systems

Modern optical networks demand more layer-bylayer flexibility compared to conventional networks. Softwaredefined Networking (SDN) may give the required degrees of freedom, but this requires the implementation of optical SDN down to the physical layer. This down to the physical layer implementation of SDN will provide the full abstraction of network components and functionality and thus enable their full control by a centralized controller. This paper provides a topologically and technologically agnostic data-driven abstracting of any N×N optical switching system for the calculation of Quality of transmission (QoT) penalties using a direct Machine learning (ML) design and the definition of its control states using an inverse ML design. The photonic design and simulation suite is used to generate a synthetic dataset for the simulated switching architecture. The results demonstrate that the proposed technique can define the control states of elementary switching units and QoT penalty with a good level of accuracy and minimize the complexity.