Machine learning (ML) has been proposed for a variety of scenarios in optical networks, ranging from network design to network management to network security. One of the use cases that is currently receiving much attention is lightpath quality-of-transmission (QoT) estimation. Selecting the path/spectrum for a new connection request, along with parameters such as modulation format, requires that the received signal quality meet a minimum threshold to ensure that the BER is satisfactory, while maintaining the feasibility of already-established connections. As is well known, precise calculation of the QoT is computationally complex; this is especially problematic in a dynamic network. An alternative methodology is to use more tractable approximations, but with conservative assumptions to ensure an adequate QoT. This typically leads to an over-dimensioned network.
ML is a relatively new approach to QoT estimation. It holds promise to expeditiously and accurately evaluate the performance of a proposed lightpath. It may involve tracking, and learning from, the characteristics of live connections (in the current network, or in another network through transfer learning), and extending this knowledge to future connections. This may be accompanied by the use of probes to expand the coverage. There are numerous aspects that need further investigation, including: minimizing false positives while also minimizing the resulting system margin; scalability with respect to data tracking, storage, and processing; capturing the relevant optical impairments with good accuracy; application to networks requiring regeneration; greenfield vs. brownfield scenarios; application to real-time dynamic adjustment of connection parameters; application to open/disaggregated networks, etc. Furthermore, intermediary steps towards an ML-based approach should be investigated as network operators may be reluctant to adopt a completely new operational paradigm. For example, as a first step, an operator may use ML to improve the precision of input parameters for their more traditional offline QoT tool, which is still based on an analytical model.
The purpose of this special issue is to investigate the current status of, and future prospects for, ML-based QoT estimation, with a focus on practical application.
Specifically, the scope of the special issue includes but is not limited to the following topics:
Submissions to the special issue should be prepared according to the usual standards for the Journal of Optical Communications and Networking and will undergo the normal peer review process. Manuscripts must be uploaded through OSA's online submission system specifying from the Feature Issue drop-down menu that the manuscript is for the issue on Machine Learning Applied to QoT Estimation in Optical Networks.
Yvan Pointurier, Huawei Technologies, France (Lead Guest Editor)
Jelena Pesic, Nokia Bell Labs, France
Cristina Rottondi, Politecnico di Torino, Italy
Luis Velasco, Universitat Politècnica de Catalunya (UPC), Spain