Hybrid-learning-assisted impairments abstraction framework for service planning and provisioning over multi-domain optical networks

This paper proposes and demonstrates a hybrid-learning-assisted impairments abstraction framework for planning and provisioning intra- and inter-domain services in a field-trial multi-domain optical network testbed. The proposed abstraction strategy consists of a parametric and a non-parametric machine learning technique to allow the control plane to implement impairments abstraction with different accessible data or monitoring technologies in the data plane. The hybrid-learning-assisted abstraction framework aims to abstract the property of segmental links along the lightpath and combine them for end-to-end performance evaluation. By deploying the proposed abstraction framework, network providers or operators are able to exchange the abstracted information for end-to-end impairments abstraction without revealing detailed information within each network. We experimentally demonstrate the proposed solution over a three-network field-trial testbed with real monitored data. The hybrid-learning-assisted impairments abstraction proves to be an accurate abstraction tool, with an average of 0.33 dB end-to-end signal-to-noise-ratio estimation error for services across the three networks. (C) 2020 Optical Society of America