A GAN Based Soft Failure Detection and Identification Framework for Long-Haul Coherent Optical Communication Systems
Huazhi Lun, Mengfan Fu, Yihao Zhang, Hexun Jiang, Lilin Yi, Weisheng Hu, Qunbi Zhuge
Abstract
The rapid progress of 5G, the internet of things (IoT), high-definition online video, and cloud computing have raised high requirements for the capacity of optical networks. To improve capacity, recent works begin to focus on the low-margin optical network. However, reduced margin may lead to soft failures caused by impairments of the physical layer, and if they are not handled in time, link disruptions may be induced. Consequently, an effective soft failure management system is of great significance. To build such a system, machine learning (ML) techniques have been studied and used. However, for the training of ML, massive labeled soft failure samples are needed, which are rare in practical systems. As a result, current algorithms are highly dependent on simulation systems or laboratory environments, which may be quite different from practical systems, leading to insufficient confidence to deploy them into practical optical networks. To solve this problem, we propose a generative adversarial network (GAN) based scheme for soft failure detection and identification. For soft failure detection, only normal samples are needed for the training, and for soft failure identification, with very few soft failure samples, a high identification accuracy can be achieved. To verify the proposed framework, we conducted simulations and experiments, where the filter shift and filter tightening are studied. Both numerical simulation and experiment demonstrate the superior performance of the proposed scheme.