The 5G decade is about at its mid-point, which has triggered the usual upgrade focus on so-called “advanced” features targeted at driving differentiated monetization from those investments, with telecom equipment giant Ericsson focusing those efforts on software and artificial intelligence (AI).

Gabriel Foglander, strategic product manager at Ericsson, noted that forward-leaning operators have worked through their initial 5G deployments and are now sitting on well-established and stable networks. This has allowed them to initiate more efficient network services providing higher network throughput and basic advanced services.

However, the next network investment step toward 5G-Advanced can further unlock software-focused efforts for a more programmable network, which can improve performance, efficiency, and open the door to new revenue-generating services. This requires a service-aware network that can tell the difference between network traffic and is intent-driven to better manage that network traffic.

Foglander did note that this push toward 5G-Advanced would accelerate if more operators had implemented a 5G standalone (SA) core. This would make it easier for operators to support differentiated services.

“Certainly, at this point we would have hoped that more networks would have been standalone by now, but we also see strong pickup,” Foglander said. “If you could look back one year from now, we're seeing that it's coming into the equation and growing and scaling quite well now. And with that you open up another kind of avenue toward these monetization paths as well.”

A 5G SA core, which consists of the user plane, control plane, and shared data layer network functions, allows operators to deliver a more resilient core network. It also supports highly-touted 5G services like network slicing, automation, orchestration, and multi-access edge computing (MEC).

This compares to a 5G non-standalone (NSA) core that relies on an operator’s legacy 4G LTE core for base processing and routing.

A report earlier this year from Dell’Oro Group found only 12 new 5G SA cores were deployed in 2023, compared to 18 in 2022. “The biggest surprise for 2023 was the lack of 5G SA deployments by AT&T, Verizon, British Telecom EE, Deutsche Telekom and other mobile network operators (MNOs) around the globe,” Dell’Oro Group’s Dave Bolan wrote.

Foglander explained that 5G SA can support more advanced components that can then feed into a smarter overall network management structure.

“It's moving from I don't want to tell the network exactly what to do, I want to tell it what I want to achieve and then we let the network be, maybe not autonomous, but at least make interpretations based on those guiding principles when it does decision making,” Foglander said.

Is AI the ‘A’ in 5G-Advanced? This network autonomy will require greater use of AI to help steer this decision making. Foglander said these efforts are still in their nascent stages with a lot of experimentation to figure out best use cases for areas like the radio access network (RAN)

“That's what the people that work very actively tell me, it's really hard to do this without doing it or trying it and truly be data driven,” Foglander said. “The models sometimes you get unexpected results, and so you really [have to ask], we fed them this data, and then when we took that model it had a confidence level of this. Why did it go that low or that high?”

Ericsson is part of a consortium being led by telecom operator T-Mobile US that is building a test facility in Bellevue, Washington, that will be house the next push toward using AI to help control cloud-based RAN and open RAN architectures, a push that is becoming increasingly important as operators begin to deploy open RAN systems. That group’s other partners include Nokia and Nvidia.

T-Mobile US noted in a recent presentation that the goal is to integrate cloud-based RAN and AI using unified infrastructure that can scale to serve millions of mobile users at once.

“AI RAN will enable new AI algorithms to unlock the full potential of wireless networks,” T-Mobile US’ presentation noted. “These AI algorithms would be rapidly developed with software-defined RAN, trained on AI data centers, and fine-tuned with physically accurate digital twins. This will lead to dramatic improvements in spectral and energy efficiencies.”

Analysts have pointed to the benefits AI can bring to the deployment and management of cloud-based open RAN networks, which are more complex orchestration challenges due to the disaggregated multivendor ecosystem. The use of AI could help close performance gaps for open RAN architectures compared with legacy RAN models.

The T-Mobile US-led work is the latest push from the AI-RAN Alliance initiative, which began earlier this year, backed by founding members including Amazon Web Services (AWS), Arm, Ericsson, Microsoft, Nokia, Samsung Electronics, SoftBank, Nvidia, DeepSig, T-Mobile US, and Northeastern University. That number has since swelled to more than two dozen.

That group’s overriding goal is to steer the use of AI into RANs for better performance, lower operating costs, greater efficiencies, and to support new business models. This work will include using AI to improve RAN spectral efficiency, combine the two for more efficient network utilization, and deploy AI at the network edge to support new services.

The initiative recently gained leadership momentum when it announced Alex Choi as head of the group. Choi has previously served in a similar position at the O-RAN Alliance.

ABI Research open RAN research analyst Larbi Belkhit noted in a report that the hiring “may be indicative of the situation with open RAN, in general, and the industry placing greater emphasis on developing AI technology for the RAN over creating a more open ecosystem.”

Ericsson’s Foglander noted that these efforts are important in helping to support AI-related innovation that is quickly outpacing traditional telecom standardization efforts.

“What we're seeing right now is actual development is outpacing standardization by a long shot, and at this point it probably doesn't make sense to set standards because it could be limiting you six months down the road,” Foglander said. “I think right now we're probably in the phase where we need to figure out what works well based on some of the learnings of generic implementation, and that's where I think ecosystem collaboration is super helpful because there's a lot of people that do this really well outside of telco. We can bring our domain understanding, and we can learn a lot for generative AI competence, and at some point I think it will be de facto practices that emerge. And then I think it would make sense at that point to try to make sure that we follow at least roughly the same rules.”