• Despite advances in network operations automation, human interaction is still required even in many ‘automated’ processes
• The topic was discussed during the recent Network Automation Summit
• While the industry strives for a zero-touch approach, operations teams will still be needed
• Telcos are encouraged to be more confident in automating parts of their networks where no service disruption can occur

Network Automation Summit 2022 – Although progress has been made towards the automation of multiple telecom operations processes in recent years, there are still many manual processes, even related to so-called ‘automation’, that can be minimised, even if human involvement can’t be removed from network operations altogether.

This was one of the key takeaways from a live audience Q&A discussion held during TelecomTV’s recent Network Automation Summit. Efforts to make networks smarter so that they run more efficiently and require less human intervention and involvement have been underway for years, and many operators around the world are now increasingly embracing software solutions that can automate at least some of the processes vital to operating a communications network.

And there are now examples of networks that show, at least, what is possible. For example, Rakuten Mobile now has a 4G/5G network that covers 98% of the Japanese population but which, according to its CEO Tareq Amin, has a network operations team of fewer than 250 staff compared with the thousands that would normally be required for a network of such size. But as many people know, Rakuten Mobile is a greenfield network, built over the past three years using cutting-edge software and disaggregated radio access network (RAN) systems that require far less maintenance than legacy technology and with a great deal of automated network management processes and cloud-native processes included from the start. As a result, existing network operators with legacy systems can’t simply replicate such a scenario, but it shows what might be possible over time for brownfield telcos.

And while there has been a lot of discussion related to the so-called zero-touch approach – with a framework that is modular, extensible and service-based, and expands across multiple domains, as industry specifications organisation ETSI puts it – when it comes to automating the networks, it appears this concept is not about removing all human intervention completely from day-to-day operations.

During the summit’s Building the network automation toolset session, the CMO of Rakuten Symphony, Geoff Hollingworth, stated that network automation should not be seen as “a removal of humans and a replacement with machines… humans will always be involved in the systems in terms of fine-tuning the efficacy and coverage of the actual machine learning (ML) or AI [artificial intelligence] systems”, or other decision-making systems that are handled by machines instead of humans.

He also noted that closed-loop (a process of continuously monitoring, measuring and assessing real-time network operations without manual intervention) is an approach that could be taken when it comes to some aspects of automating the network.

“The small scope ones [systems] that may use up a lot of time and effort today can absolutely be removed by closed-loop in a very safe and trustworthy way,” Hollingworth pointed out.

The Rakuten Symphony executive likened self-driving networks to running a self-driving car. “There’s certain things that we do closed-loop today, very isolated, very low impact” that are very common and repeatable, he noted, adding that people “have a confidence level” when it comes to autonomous operations. 

But that confidence level can and will grow.

“Coping with natural disasters where there’s chaos and there’s millions of [network] events, and trying to translate that, is not something that we do in an autonomous way today – and there’s a debate [about] whether we get to that or when we get to that,” said Hollingworth. (While he didn’t use a specific example, Japan’s mobile operators have been working on automated critical communications capabilities that can be activated and run autonomously in the event of a major natural disaster, such as an earthquake or tsunami.)

According to Grant Lenahan, partner and principal analyst at Appledore Research, fully-fledged AI solutions are not yet commonly used in telco operations. “There is a belief that [while] ML is programmed in Python, AI is programmed in PowerPoint – and I think that kind of summarises the state of the industry. There’s not a lot of real honest-to-goodness AI, with deep-learning, unsupervised [operations] being used in telecom”, he noted.

Lenahan highlighted a few levels where humans are involved in the process, such as creating the ML methods and algorithms, noting that ML is divided into supervised and unsupervised. In the first scenario, a “machine takes a guess, it submits it to humans or a human who says that was a good guess or a bad guess – maybe finding out after implementing it, suggesting it and seeing if the problem clears”.

In the unsupervised ML approach, the system is largely run based on its track record. But again, Lenahan explained, humans also play a role in the system.

However, Lenahan recommended that operators should recognise that to scale and get the benefit of real-time network management in the future, “we’re going to have to get human beings out – there simply aren’t enough human beings. Even if we had infinite trained human beings, they don’t operate in a couple of milliseconds. So, we need to get beyond that”.

Rahul Atri, advisor and head of transformation and strategy for telco at consultancy MNC, said the technology to enable closed-loop, automated processes already exists for telco networks. But he added that operators still take the usual route of having someone supervising the closed-loop action and having “clear guidance” about how such an action will reflect on the network as a whole.

He suggested that telcos start slowly, with small actions instead of drastic automation changes that do not affect the service or disrupt the network but “gives you a confidence to say – yes, this closed-loop is actually working, the model is tuned”. He believes the industry is ready, but “we just need to close the loop on how the chain management and overall process would work”.

Rafaqat Chaudhary, chief technologist of RAN automation at Hewlett Packard Enterprise, agreed that it is important to ensure there is enough confidence to run networks automatically.

He said the reluctance of putting activities traditionally done by network operation centre engineers into autopilot mode is something that needs to change – and that this will require different skill sets. “The transition does not happen just by rolling out new AI technology… [we need to] upgrade the people part as well in terms of the skills… Rather than focusing too much energy looking into logs and traces”, it’s important instead to “put energy and resources into learning how to implement, how to do training, and learning how to operationalise those tools”, explained Chaudhary.

He concluded that apart from shifting focus to skills set requirements, telcos also need to be “more open [and] less reluctant to put that technology into actual use, so that sustained autonomous operation can be achieved”.

- Yanitsa Boyadzhieva, Deputy Editor, TelecomTV