• The companies say they’re introducing a new machine learning (ML) algorithm to Vodafone’s pan-European mobile networks to detect and correct anomalies 
• Based on Nokia Bell Labs technology, the cloud based ‘Anomaly Detection Service’ is looking for ‘out-of-the-ordinary’ behaviour which might impact the quality of service
• And there is a big push behind getting some sort of standard architecture

Vodafone, in partnership with Nokia, is deploying machine learning (ML)-based monitoring tools in its pan-European mobile networks to detect and correct anomalies before services are affected. The machine detective work allows Vodafone engineers to address issues faster: Things like mobile site congestion, radio interference, unexpected latency, or difficulty in handing off calls between different cells, can be speedily identified and rectified, say the companies.

In addition to detecting anomalies, the algorithm will also identify patterns of change to allow Vodafone’s operating companies to proactively address issues at their first signs of  development - again, hopefully before customers are affected.

So confident is Vodafone with the power of the new service that it’s set as its target to automatically detect and address 80 per cent of all anomalous mobile network issues and capacity demands.

The companies say the algorithm has been tested on live networks to demonstrate its accuracy and to ensure that it works with equipment from all network vendors. Following its initial deployment in Italy on more than 60,000 4G cells, Vodafone will extend the service to all its European markets by early 2022, it says. 

Nokia says the Vodafone move is the system’s first commercial deployment.

The Anomaly detection is offered “as-a-Service” as part of Nokia’s Cloud Support Services.There has been much talk and publicity generated around the use of AI and machine learning for telcos, especially for managing and trouble-shooting their networks. The Vodafone/Nokia announcement highlights the fact that the techniques may be starting to be used as expected. 

“It’s  a massive area of focus for telcos,” says Aaron Boasman-Patel, Vice President, AI & Customer Experience at TM Forum. He says the forum is creating  a framework involving standard APIs and interfaces under which the machine learning and AI can be marshalled for different use cases and it’s now releasing a first version of the reference architecture. “The dream for many telcos (and big equipment vendors like Nokia) is a thriving ecosystem and marketplace for AI/ML applications, '' he says, so there is a big push behind getting some sort of standard architecture.

In the Vodafone/Nokia example, the solution runs natively on the public cloud, streaming data to Vodafone’s analytics platform and enabling the analysis of aggregated and anonymised network data from various points across multi-vendor environments all at once, rather than processing single variables independently.

 As a result, the solution can extract temporal patterns from the analyzed system to categorise network behavior and find anomalies; allowing Vodafone to quickly address issues that might impact their customers’ experience.

And that’s all fine and good, but there are problems. 

“The trouble is that telcos are currently automating for one domain and not across all of them.” says Boasman-Patel. “You have to have an architecture so you can scale up your (AI/ML) efforts,” he says. And that’s what the TM Forum is attempting to organise. “The point is that you have to work towards some commonality, otherwise you get lock-in and all those things. You have to manage automation at scale.” 

But it’s early days, he says. “Today it’s all about point solutions and I think it will be another five years before that changes.”