What AI brings to optical networks

Arthur Cole
AI robot eye

Optical networks are tailor-made to handle lots of data. With spectrum that far exceeds the practical limits of copper, fiber will be a key asset in the drive to establish 5G and the internet of things as the new cornerstones of the digital economy.

But unlike previous boosts in data usage, spurred by video streaming, teleconferencing and other high-data applications, the next wave of applications will feature small-packet transmissions that will not just place new bandwidth requirements on optical infrastructure but will rely on greater network flexibility and functionality.

Network intelligence

This will put pressure on the management side of the house, most likely demanding levels of performance that exceed the capabilities of even the most experienced operator. This is why research is ramping up on new forms of automated, even autonomous, management stacks, driven largely by machine learning (ML), neural networking (NN) and other forms of artificial intelligence (AI).

But exactly what can these new technologies bring to optical networking in particular, and to what degree should automation be allowed to oversee the flow of what is likely to be highly critical data?

At Light Reading’s first Optical Networking Digital Symposium in May, the message was clear: automation is already bringing tangible benefits to optical networks, and its full potential is not limited by technological constraints but by human imagination. While most observers may think the ultimate goal of automation is a fully autonomous network, the reality is that there is a wide range of possibilities that fall short of such an extreme. These range from functions like smart maintenance, which uses real-time data collection and other tools to monitor network health and initiate corrective actions, to things like traffic prioritization, customized resource allocation and service differentiation. In all likelihood, network operators will decide for themselves how to implement AI and automation based on service needs, reliability and performance.

When it comes to actually modeling AI tools to make decisions regarding optical networks, one of the key goals should be to maintain high quality of transmission (QoT) through active management and the ability to oversee heterogeneous network environments, according to a team of researchers at the Shanghai Institute for Advanced Communication and Data Science. AI brings many powerful capabilities to this challenge, such as the ability to estimate nonlinear noise with far greater accuracy than traditional analytical models. As well, properly trained ML algorithms can more easily adjust to the way network flows change according to the type of traffic being carried, the scale of the network and other factors.

Gauging the intent

But perhaps the most profound change that AI brings to optical networking is the shift from an actively managed environment to a more intent-based approach. A recent paper led by Beijing University’s Hui Yang described how two key closed-loop operations – closed loop strategies generation and closed-loop intent guarantees – can take over network configuration roles like adjusting command line interfaces and middleware scripts. With this in hand, network architects can craft intent-defined optical networks (IDONs) through self-optimizing, self-adapted generation and optimization (SAGO) policies. What’s more, through the use of natural language processing and neural networking, operators will be able to construct semantic graphs to understand, interact and operate required network configurations. In other words, you merely tell the network what you want and the necessary changes are made automatically.

While a fully autonomous, self-optimizing optical network is clearly possible, at least in the lab, there are still questions as to whether such a thing is practical, or desirable. At some point, human operators should be at the ready to make the kinds of insightful, intuitive decisions that even the most intelligent algorithms cannot make, no matter how much data they process.

Finding this sweet spot between organic and artificial intelligence will be one of the primary challenges going forward, even after machines have been put in charge of the key elements of optical network management.

Related articles