Thanks to end-user demand for data and bandwidth-intensive services like cloud computing, edge computing, 5G and the Internet of Things (IoT), network operators are constantly trying to find ways to increase their capacity while keeping capital and operational costs as low as possible.

For those involved in the deployment and operation of the optical network infrastructure upon which all our applications depend, this is the key question:

What is the most potent role of Artificial Intelligence (AI) and Machine Learning (ML) in improving network agility and resiliency?

As the Chief Technology Officer of a New York-based network systems engineering company, I believe the answer lies in the relationship between optical devices and their analog data coupled with SDN application programming interfaces (APIs). From an optical networking perspective, AI is about improved data processing to enable smarter decisions and greater network efficiency. SDN, on the other hand, provides the mechanism for leveraging much of what AI/ML has to offer. While much attention has been paid recently to SDN and its iterations, in this article, I want to concentrate more on the power of AI/ML and how best to use this burgeoning technology in optimizing optical networks.

AI and optical networking trends

Network operators today struggle to balance three major challenges: network complexity, service level requirements and limited resources. That is, modern operators control networks with varying purposes that support the delivery of offerings at different service levels. To make matters even more challenging, the traffic that passes along these networks competes for resources like fiber, wavelengths, and time slots. To keep up with their customers’ ever-growing demands for speed and bandwidth, network operators and researchers are turning to AI to simplify and automate network operations.

When it comes to AI, the industry is driving toward the same goal: self-optimizing, autonomous networks that can automatically deliver what operators want without requiring the manual procedures commonplace today. The problem with conventional network operations is that they are both time-intensive in nature and susceptible to human error, which can cause disruptions that modern applications cannot tolerate. Thanks to advancements in Software-Defined Network (SDN) technology, network operators now have a powerful tool that can be paired with AI tools to program and configure networks optimally, through centralized management.

Many AI techniques are being applied to different aspects of optical networks right now. From transmitters and optical amplification control to OSNR monitoring and Quality of Transmission (QoT) estimation, engineers are finding unique ways to improve the intelligence of modern optical networks by making it easier to:

In all of this, machine learning (ML) has a strong role to play with engineers concentrating on supervised, unsupervised and reinforcement learning techniques. To function effectively, algorithms need vast amounts of data not only to learn patterns, but also to learn how to continuously optimize/upgrade network performance as in the case of Q-learning, a type of reinforcement learning. Real time analytics is one area of AI/ML innovation that offers network operators the foundation they need to eventually develop the autonomous, self-optimizing networks they seek. However, this is still a relatively untapped frontier and one that requires more work to be done on utilizing the analog data generated by the optical devices within layer 1 of modern optical networks.

Enhancing intelligence by leveraging data from the optical level

Many of today’s AI-enabled or ML-powered analytics tools examine the digital data from the higher layers of optical networks, which includes units like packets and frames as well as segments like TCP and UDP protocols. However, focusing on this information, while helpful, is only half the equation. Since digital data uses sampling to encode what is measured, it can only alert network operators as to the existence of errors. While AI/ML algorithms can solve these errors, the end goal for AI/ML applications is to prevent them from occurring in the first place.

Because analog devices use data that is continuous, they provide the real-time picture of everything that is happening at the most foundational layers of optical networks. As a result, focusing on the physical layer can help network operators detect issues when they are still at the anomaly level – before they turn into actual errors. SDN functionality as well as ML are key here. By acquiring and processing information from every optical device, operators can improve both the intelligence and resiliency of their networks.

Intelligence drives decision-making, which in turn paves the way for optimal performance. Equipment like transceivers, optical channel monitors and amplifiers are mission-critical components of any network’s operations. As a result, the future of ML-powered SDN analytics applications will likely concentrate heavily in this area. My team has already been leveraging an ML-powered SDN application of our own, Lightseer, to help our global customers automate their network monitoring and configuration. The future of optical AI is in applications that can aggregate the optical data of networks to predict and help solve issues before they become failures. This will be key in helping network operators boost network capacity and performance while reducing operational costs.

Summing it all up

The future of AI within the optical networking space is bright. Together, ML/AI and SDN technologies provide network operators with the tools needed to transform their networks for the better. While all current AI research is fascinating and completely necessary, those involved in the development of analytics tools need to ensure that they are paying attention to the analog data generated by the optical layers of networks. This provides the true jumping off point for the creation of the ideal autonomous networks that, coupled with SDN functionality, can centrally optimize themselves and provide the economic, efficient performance network operators want.

Chris Page is CTO at Precision OT, a system engineering company focused on optical transceivers and related active/passive optical components.