The availability of coarse-resolution cost-effective Optical Spectrum Analyzers (OSA) allows its widespread deployment in operators’ networks. In this paper, we explore several machine learning approaches for soft-failure detection, identification and localization that take advantage of OSAs. In particular, we present three different solutions for the two most common filter-related soft-failures; filter shift and tight filtering which noticeably deform the expected shape of the optical spectrum. However, filter cascading is a key challenge as it affects the shape of the optical spectrum similarly to tight filtering; the approaches are specifically designed to avoid the misclassification of properly operating signals when normal filter cascading effects are present. The proposed solutions are: $i) multi-classifier$ approach, which uses features extracted directly from the optical spectrum, $ii) single-classifier$ approach, which uses pre-processed features to compensate for filter cascading, and $iii) residual-based$ approach, which uses a residual signal computed from subtracting the acquired single by OSAs from an expected signal synthetically generated. Extensive numerical results are ultimately presented to compare the performance of the proposed approaches in terms of accuracy and robustness.

Network slices combine resource virtualization with the isolation level required by future 5G applications. In addition, the use of monitoring and data analytics help to maintain the required network performance, while reducing total cost of ownership. In this paper, an architecture to enable autonomic slice networking is presented. Extended nodes make local decisions close to network devices, whereas centralized domain systems collate and export metered data transparently to customer controllers, all of them leveraging customizable and isolated data analytics processes. Discovered knowledge can be applied for both proactive and reactive network slice reconfiguration, triggered either by service providers or customers, thanks to the interaction with state-of-the-art software-defined networking controllers and planning tools. The architecture is experimentally demonstrated by means of a complex use case for a multidomain multilayer multiprotocol label switching (MPLS)-over-optical network. In particular, the use case consists of the following observe-analyze-act loops: 1) proactive network slice rerouting after bit error rate (BER) degradation detection in a lightpath supporting a virtual link (vlink); 2) reactive core network restoration after optical link failure; and 3) reactive network slice rerouting after the degraded lightpath is restored. The proposed architecture is experimentally validated on a distributed testbed connecting premises in UPC (Spain) and CNIT (Italy).

Focused on reducing capital expenditures by opening the data plane to multiple vendors without impacting performance, node disaggregation is attracting the interest of network operators. Although the software-defined networking (SDN) paradigm is key for the control of such networks, the increased complexity of multilayer networks strictly requires monitoring/telemetry and data analytics capabilities to assist in creating and operating self-managed (autonomic) networks. Such autonomicity greatly reduces operational expenditures, while improving network performance. In this context, a monitoring and data analytics (MDA) architecture consisting of centralized data storage with data analytics capabilities, together with a generic node agent for monitoring/telemetry supporting disaggregation, is presented. A YANG data model that allows one to clearly separate responsibilities for monitoring configuration from node configuration is also proposed. The MDA architecture and YANG data models are experimentally demonstrated through three different use cases: i) virtual link creation supported by an optical connection, where monitoring is automatically activated; ii) multilayer self-configuration after bit error rate (BER) degradation detection, where a modulation format adaptation is recommended for the SDN controller to minimize errors (this entails reducing the capacity of both the virtual link and supported multiprotocol label switching-transport profile (MPLS-TP) paths); and iii) optical layer self-healing, including failure localization at the optical layer to find the cause of BER degradation. A combination of active and passive monitoring procedures allows one to localize the cause of the failure, leading to lightpath rerouting recommendations toward the SDN controller avoiding the failing element(s).

The massive deployment of 5G and beyond will require high capacity and low latency connectivity services, so network operators will have either to overprovision capacity in their transport networks or to upgrade the optical network controllers to make decisions nearly in real time; both solutions entail high capital and operational expenditures. A different approach could be to move the decision making toward the nodes and subsystems, so they can adapt dynamically the capacity to the actual needs and thus reduce operational costs in terms of energy consumption. To achieve this, several technological challenges need to be addressed. In this paper, we focus on the autonomous operation of Digital Subcarrier Multiplexing (DSCM) systems, which enable the transmission of multiple and independent subcarriers (SC). Herein, we present several solutions enabling the autonomous DSCM operation, including: i) SC quality of transmission estimation; ii) autonomous SC operation at the transmitter side and blind SC configuration recognition at the receiver side; and iii) intent-based capacity management implemented through Reinforcement Learning. We provide useful guidelines for the application of autonomous SC management supported by the extensive results presented.

Autonomic optical transmission and networking requires machine learning (ML) models to be trained with large datasets. However, the availability of enough real data to produce accurate ML models is rarely ensured since new optical equipment and techniques are continuously being deployed in the network. One option is to generate data from simulations and lab experiments, but such data could not cover the whole features space and would translate into inaccuracies in the ML models. In this paper, we propose an ML-based algorithm life cycle to facilitate ML deployment in real operator networks. The dataset for ML training can be initially populated based on the results from simulations and lab experiments. Once ML models are generated, ML retraining can be performed after inaccuracies are detected to improve their precision. Illustrative numerical results show the benefits of the proposed learning cycle for general use cases. In addition, two specific use cases are proposed and demonstrated that implement different learning strategies: (i) a two-phase strategy performing out-of-field training using data from simulations and lab experiments with generic equipment, followed by an in-field adaptation to support heterogeneous equipment (the accuracy of this strategy is shown for a use case of failure detection and identification), and (ii) in-field retraining, where ML models are retrained after detecting model inaccuracies. Different approaches are analyzed and evaluated for a use case of autonomic transmission, where results show the significant benefits of collective learning.