SK Telecom avoids network tech hype as it goes all-in on AI

SKT, known as a network leader, has deployed little Cloud and Open RAN so far, and doubts that 5G SA can help it with 5G monetisation. But it's a different story with AI.

On its booth, SK Telecom is drawing crowds with a passenger drone that it has titled “AI Urban mobility”. There’s an AI-based 6G simulator. AI based RAN. Partner demos on Gen AI with AWS and Azure and Google. Its own Telco LLM is prominent, and the operator just announced it is tuning the Telco AI Alliance into a JV, with the target of producing further telco specific LLMs. There’s the Sapeon AI processor with Llama 2. There’s Deep Network AI which is using AI to identify network state and faults, including the use of very accurate positioning using map objects identified by, yes AI. There’s a PAA – Personal AI Assistant – demo. Just in case you didn’t get the message, the booth strapline and the shiny custom lanyards on the booth staff read “AI on the  point of inflection”.

Upstairs, in a meeting room,  Chiyoung Choi, Team Leader, Infra Strategy Team, talks to TMN about the operator’s network strategy. The network, after all, is still the thing that will sustain, and in turn be sustained by all this AI-driven, innovation. And SKT’s network is very good. It regularly wins benchmarking tests, and provides speeds faster than most other networks globally.

SKT is absolutely not afraid to lead with innovation. It has far more confidence in its brand power than many operators do – and confidence in using technology to underpin that position. Yet it becomes clear that the network itself is advanced but solid, with few of the latest buzzword technologies yet being deployed. No Cloud RAN. No Open RAN. No 5G SA. And no concrete, immediate plans for deployment.

In Korea, SKT has almost completed nationwide 5G coverage. Its current network is built using RAN from Ericsson, Nokia and Samsung. Samsung has about 50% of the network by share, mainly in the metro areas. Ericsson has about 35% and Nokia 15% .

Open RAN

99.9% of the network is dedicated RAN, not Cloud RAN, Choi says. There’s no Open RAN at all*, as SKT did its rollout before Open RAN was introduced to the market. Nor is there much scope for either technology to be adopted in the current investment round.

“SKT’s outdoor coverage is almost done. The largest scale buildings, over 10,000 square metres, we use DAS. At the end of this year we will complete our target to cover those. So what’s left? Medium or small sized buidings, or small underground stations. So there’s no chance to install Open RAN. There’s no room.”

So does that mean that, with the industry furiously discussing vendor diversity and the benefits (or otherwise) of vRAN for accessing new technology, SKT sits to one side?

“Our R&D team do their tests to make some technical progress, just as DT or NTT DoCoMo do, with Nokia and Samsung and others. The key point about Open RAN is firstly performance. In the Korean market it’s very competitive – so comparing performance in download speed is very important to operators in Korea. Dedicated systems have higher performance currently so there’s no reason to replace that. Secondly from an ESG perspective Open RAN still consumes more power than a dedicated system. Thirdly, Open RAN is more expensive than existing systems. So when SKT considers three factors, we want to go further with Open RAN, go into the Open RAN market and ecosystem to manage vendors’ walled garden. But it’s still not mature.”

Is Choi  not convinced by other operator metrics – or instance from Vodafone in its UK build, and from Rakuten in Japan, Dish in the USA – that say performance is as good, and costs are indeed lower?

“I met operators and we talked about the status. They agreed with some parts of our opinion because there’s still some difference in power consumption and performance.”

* In a later briefing SKT asked TMN to clarify that it has deployed some Open RAN interfaces in in-building deployments. It is also investing in R&D in Open RAN, as stated above by Choi, and is active in O-RAN specifications and plugfests. It is not able to say when it might deploy Open RAN in its network. 

No play for 5G SA

Let’s leave Open RAN and Cloud RAN, then, and turn to monetisation.

“Our concern,” Choi says, “is that we invested a bunch of money already, lots of capex, but as you know in 5G services there are not big differences between 4G and 5G. That is the huge problem for current operators. So, monetisation of the 5G network is very important and many operators cannot find a good solution for how to monetise.”

For some operators, monetisation hopes are pinned on the introduction of 5G SA, bringing with it a 5G Core that can enable them to meet “true” 5G capabilities in latency, and service flexibility via functionality such as slicing.

“Many operators agree they need it, but SA does not guarantee monetisation,” Choi says. “From SKT’s perspective it may reduce customer experience because in our situation we have five LTE bands, totalling 75 MHz LTE spectrum. And we have100 MHz contiguous at 3.5GHz for 5G.” At the moment SKT can aggregate the 5G band with the LTE bands to give customers very high throughputs, because NSA 5G is anchored in the LTE core and can provide Dual Connectivity between the layers. But SKT is worried that in 5G SA, as it loses the connection to 4G, it won’t be able to aggregate that spectrum.

“As we proceed to utilise the spectrum we are focussed on NSA. That means NSA can provide better performance for customers. At the moment if 5G coverage not quite good, customers can use LTE and 5G channels. But in the SA case the customer only download using the 5G channel, so that might make a worse experience.”

What, then, of 5G SA carrier aggregation?

“SKT has one [contiguous] 100MHz band, there’s no spectrum to aggregate.”

So we won’t see 5G SA for some time?  “We have talked about it internally and if there’s a chance to monetise we may do SA.The core is with Samsung and Ericsson. We are technically read so we can change our network from SA to NSA, but there’s no chance to monetise from SA network currently.”

But YES to AI

If 5G SA is off the table, what is SKT’s approach to monetisation?

“One approach of SKT is to go through with AI to transform our network and services.”

Choi describes embedding AI functionality in the infrastructure later, evolving its network with embedded AI functionality. A second layer of AI is to change the way the operator works – adopting AI functions like anomaly detection, auto-healing, analysing RF conditions to doing away with drive testing.

The third layer is to develop AI services – for example the Personal AI Assistant being demo’s on the booth.

SKT is also developing very accurate indoor positioning, combining WiFi positioning with time delay solutions with AI logic to create very accurate results. The DNA demo on the booth shows how the two can be combined into a powerful network analytics, monitoring and optimisation tool.

“Positioning is our strong point and we want to make an AI based service for positioning,” Choi said.

Then there the network API discussion. SKT has developed some use cases on network based APIs – one example being verification of location for financial services. Here the key is develop a common API across operators in the Korean market, so that enterprise developers can interface with just one API. Choi sees the Korean operators also basing their local APIs on global standards.