Nokia maps out AI/ML automation path for broadband operators

Thanks to ChatGPT, a chatbot capable of human-like conversation, awareness of artificial intelligence (AI) and machine learning (ML) — and how effective they can be — has gone mainstream. It’s hard not to be impressed by AI/ML. Aside from its chatbot capabilities, ChatGPT can generate essays, poems, art, software code and other types of text that give the impression (usually) of being crafted by human hand.

ChatGPT has made investors sit up, too. Since its release in late 2022 there’s been something of a rally in AI and tech stocks. Market expectation seems to be that AI/ML, following on from ChatGPT’s roaring success, can fulfil its broader potential of increasing productivity and cost-efficiencies across different industry sectors and perhaps, in the process, upend existing business models.

Broadband operators are among those that don’t want to be left behind in the AI/ML gold rush. Filip de Greve, product marketing director at Nokia, says there’s now much more urgency from them to tap into AI/ML and improve their network operations. The good news is that de Greve, looking beyond the inevitable AI/ML hype cycle, sees concrete business gains for broadband operators by harnessing AI/ML as part of software-defined networking in the access network (SDAN).

Nokia is working closely with many of its broadband operator customers to expose large datasets that feed AI/ML. This data can be used to train deep neural networks, in much the same way as ChatGPT does. The aim — aided by telemetry and data driven analytics — is to achieve much higher levels of automation in service provisioning, monitoring and network management.

"AI automation is expected to boost productivity in telecom, in much the same way as it’s doing for other industries.” de Greve told Light Reading. “AI can make operator systems more intelligent and automated, enabling predictive and near-real-time actions. It’s about detecting network behaviour that humans may not detect, and making links, associations and correlations that people wouldn’t easily make."

AI/ML prep: open up, go cloud native, be intent driven

In Nokia’s book, enabling access to large datasets that will drive AI/ML and reap benefits for broadband operator requires first and foremost an open and data-centric architecture. “You can’t automate something you can't measure,” quips de Greve.

He emphasizes that Altiplano, Nokia’s programable SDAN solution — which leverages open-source components — is built on the concept of openness. For a start that means open application programming interfaces (APIs). “Our open API is an API you can actually customize and change the parameters to make OSS integration and access to network data a lot easier,” explains de Greve.

Nokia has also made available the Altiplano software development kit (SDK) for third-party developers — another aspect of openness — so broadband operators are given the flexibility, in de Greve’s words, “to develop automation use cases we haven’t even thought of.” Altiplano customers, he points out, get enormous peace of mind that there are no proprietary custom-made solutions and vendor lock-ins.

“Network automation assists with the increased number of options, parameters and dimensions to optimize,” continues de Greve. “This is an area where AI/ML can complement, or even surpass, existing algorithmic power, and augment human capabilities in planning cycles, adapting to change and responding to threats.”

AI/ML, he stresses, broadens the range of network behaviour that can be captured. “It’s why we’ve integrated AI/ML capabilities in apps hosted in our Altiplano marketplace,” says de Greve.

Moreover, broadband operators want flexibility to run apps in cloud environments that suit them, whether it be on a local server, in a private or public cloud, or in software-as-a-service. Because Altiplano is cloud-native, capable of running microservices (small independent software components) in containers, de Greve says. Nokia has met the “mandatory RFQ [request for quote] requirement” from customers to be cloud agnostic. “The fully containerized microservices design of Altiplano allows smooth migrations between different cloud deployment models,” he says.

As good as all this sounds, de Greve warns broadband operators that if they haven’t yet implemented an intent-driven design, which involves a ‘closed loop’ that automatically verifies if an intent (a service provisioning request) is still valid or not, “then you don't have the right framework to be really data driven.”

If Nokia is anything to go by, however, this requirement shouldn’t be too much of an industry AI/ML showstopper in the near to mid term. De Greve says the intent driven model has already been deployed by more than 100 Nokia operator customers and that adoption - industry-wide - is at the “early majority” stage.

Automation in action

The Altiplano app marketplace with the help of clever AI/ML algorithms, provides broadband operators with a range of tools to boost operational efficiencies. They can automate tasks to derive an optimal set of rules, which is an unlikely outcome if things are done manually.

Bandwidth management is one of the AI tools available. Through use of high-frequency traffic telemetry data and ML techniques, Nokia claims it has built a unique model for residential traffic — based on a brand-new set of AI/ML based algorithms — that can predict and improve peak-rate availability for subscribers.

“We can be very accurate about congestion levels and full capacity planning, gaining up to 30% in peak bandwidth that can be offered,” asserts de Greve. Alternatively, he says, broadband operators can sometimes double the number of subscribers on a PON, from 32 to 64, while maintaining the original peak rate. “These are significant benefits,” says de Greve.

More accurate KPI threshold settings for fault detection, aided by multivariate analysis and ML, is another helpful AI tool. Typically, notes de Greve, broadband operators are conservative with threshold-setting, thinking only about worse-case scenarios, which can lead to lots of false negatives and missed detections. On the other hand, by setting thresholds too aggressively, there’s a danger of triggering an avalanche of false positives. AI/ML, says de Greve, can automatically compare performance outliers and avoids the pitfalls of setting thresholds manually.

Smart algorithms can also spot developing KPI trends that, if not addressed, will cause a service outage. AI/ML can also detect the problematic alarm patterns and correlate them to reduce problem investigation time. Predicative maintenance is part of Nokia’s AI/ML armoury.

“By our estimates,” claims de Greve, “Altiplano and its AI tools, taken collectively, can cut broadband operators’ opex by as much as 40%.” AI/ML excitement, then, should not be limited to chatbots. AI tools can be usefully applied to improve network management and make automation systems much more efficient.

Explore more about SDAN.