Nokia launches new salvos against Ericsson and Intel in open RAN

Open RAN, like an adaptable virus, has undergone several mutations since birth. Having begun life as the glue that would hold together a United Nations of telecom parts on the multivendor fringes of networks, it today justifies a $14 billion single-vendor deal between AT&T and Ericsson, the biggest contract in the Swedish company's history.

That's nuts, reckoned numerous visitors to Mobile World Congress (MWC) last week. Others were more guarded in their warnings about the increasingly single-vendor condition of open RAN. "If there's a democracy with only one candidate, is it really a democracy?" said Bruno Zerbib, the chief technology officer of France's Orange.

Ericsson heralds a new paradigm in vendor selection. Suppliers used to be given neighboring dukedoms in a telco network, each with full dominion over a geographical area. In future, telcos will choose the vendors they deem best for each sandwich layer of the network, whether service and management orchestration (SMO), baseband software in central and distributed units (CUs and DUs) or radios (RUs).

In some layers, such as radios, there will be room for multiple vendors per telco. Yet in others, it seems one vendor might butter itself across the whole bread slice of a large nationwide network. Fujitsu, as well as Ericsson, is down to provide radios to AT&T. But Ericsson is the sole named software vendor. Its own top executives have referred to the "single vendor" AT&T deal.

For Tommi Uitto, this Ericsson view of the "horizontal layers" is specious. Uitto runs the mobile networks business group at Nokia, whose equipment Ericsson will now dismantle across roughly one third of the AT&T network, the full extent of the Finnish vendor's footprint. He therefore has good reason to dislike what he sees. But he makes objections on philosophical grounds.

"What would be problematic from an open RAN point of view is the suggestion that layering should mean a sole supplier for the DU/CU layer in large networks and giving most or all of the RU business to the DU/CU supplier," he said when he met Light Reading at MWC. In defense of such criticisms, Ericsson says its compliance with open RAN specifications would allow AT&T to replace it whenever. But Uitto is dismissive.

"We all know that changing the complete DU/CU layer will be equally difficult in practise as changing a supplier with a 'vertically integrated' basestation in a geographical region," he said. "It will be even more difficult to replace the whole DU/CU if you have given the supplier a good part of the RU business."

The situation for the operator might be even worse if the DU/CU supplier also lands a contract for the SMO layer, according to Uitto. Open interfaces in this part of the network remained incomplete late last year, warned NTT Docomo and Vodafone in a co-authored white paper. If the DU/CU vendor creates "sticky functionality" between this software and the SMO systems, it would be even harder to replace, said Uitto.

"This is because the SMO layer is so deeply linked with the IT systems of the operator," he added. "In effect, the DU/CU supplier can extend the stickiness and customization of IT systems all the way from IT to SMO to DU/CU, making it extremely sticky."

The algorithmic hurdle

Open interfaces, clearly, will go only so far in helping telcos to combine vendors at scale and with ease. Massive MIMO, an advanced, antenna-rich 5G technology, remains a particular challenge because complex software algorithms split between the DU and the RU need to match. "If I have to connect my DU to someone else's massive MIMO RU, he will have to make changes to his software, I will have to make changes to my software, or both will," said Uitto. "I already have software in the field, so what do I do with that? Am I going to have two software tracks?"

Others on both the vendor and the telco sides agree with his assessment. While Vodafone recently flaunted its first multivendor massive MIMO site in the UK, pairing Samsung's DU software with an NEC radio, most commercial deployments are single vendor in massive MIMO. That includes Nokia's 3,000-site open RAN deployment in Deutsche Telekom's network. For more basic radios, Nokia's DUs are paired with Fujitsu's RUs. In massive MIMO, it is all Nokia.

One option is to make use of simpler algorithms. A beamforming algorithm known as grid of beams (GoB) allows for a reasonably straightforward integration of vendors, according to Uitto. But far more sophisticated algorithms have subsequently emerged. Compared with these, GoB would reduce throughput by anywhere between 20% and 80%, he reckons.

"Today, I claim it is not possible to get market-leading throughput with open RAN massive MIMO," he said. "It can be done, but the performance is lesser, and some work is needed to make it feasible, make it practical, make it something that actually can be used in the field. But it will be done. We will do it."

Doing the splits

Massive MIMO concerns do not end here, though. Efforts to reach consensus on the precise division of uplink network functions between the DU and the RU foundered last year. A meeting in Japan has instead left the industry with several options – the old 7.2x category B split, deemed inadequate by various stakeholders, and its class (or operation mode) A and B variants.

The main difference between them? Class A shifts the equalizer, an important function that addresses interference, into the RU, necessarily making it more complex. Free of equalizers, class B radios are simpler. Yet DUs must include equalizers to comply with specifications, ensuring they can support both types of RU.

Uitto says Nokia will cover all bases and make both class A and class B RUs. That differentiates his company from Ericsson, which has thrown its full RU weight behind class A while insisting its DUs will include equalizers and therefore support both modes to comply with the specs. It already has more than 1 million compatible RUs in the field, it has said. Without naming Ericsson directly, Uitto is critical of the strategy.

"If you have baseband in the field, like one of my big competitors, that doesn't support the equalizer, they cannot connect the baseband that doesn't have the equalizer to an RU that doesn't have the equalizer because there is an equalizer nowhere," he said. "And that is why a big competitor has a problem with the installed base. You have to change to a new baseband or do cloud RAN. You can't use the existing baseband." Ericsson, for its part, says DU support for open RAN specifications will come with baseband products it plans to launch this year.

Disaggregation and re-aggregators

Multivendor rollouts would conceivably be less problematic if massive MIMO had not been allowed to advance so far in a single-vendor world. "Somehow the rabbit got out of the cage," said Uitto. "If we were coding the software and algorithms from scratch, it would be easier to align." Even NTT Docomo, which boasts a multivendor network along with systems integration expertise in open RAN, has avoided making any investments in massive MIMO, he said. Other sources agree.

Regardless, the future universe of disaggregated networks will require someone to be the "re-aggregator," as Uitto puts it. The question is who. With their current expertise and development budgets, the giant vendors would seem to be in pole position. In any contract, Uitto thinks it is the DU/CU vendor that is more likely to be handed responsibility. "So much of the network functionality and performance is decided by the DU and the CU," he said. But this could explain why new entrants appear to be having more success on the RU side.

While some telcos would prefer to play the systems integrator role, even AT&T executives think operators must club together to challenge what Robert Soni, AT&T's vice president of RAN technology, has called the "closed ecosystem" world. "The only opportunity for open RAN to catch up to that ecosystem – and provide performance and feature parity but also leapfrog in terms of capabilities – is for us to pool," he said late last year.

What seems increasingly unlikely is that any third-party systems integrators – the likes of Accenture, Capgemini and Tech Mahindra – will feature prominently at all. Many telco executives seem to regard them as an avoidable cost, and there has even been the suggestion they could become a different form of "lock-in." Uitto does not evidently see them as a competitive threat. "In one early case, an operator said they would consider giving it to a systems integrator, and one of the leading ones said we don't have domain expertise in this space," he said.

Chip choices

When it comes to virtual or cloud RAN, Nokia remains vehemently opposed to "lookaside," an approach that keeps most functions on the central processing units (CPUs) typically supplied by Intel. Its preference is an alternative technique called "inline," which shifts the entirety of Layer 1 – the most demanding part of the RAN software stack – onto third-party custom silicon instead. In Nokia's case, it is provided by Marvell Technology and hosted on smart network interface cards (SmartNICs) that plug into the servers.

Ericsson is firmly in the other camp and its recent momentum in North America has added to the impression lookaside is ahead in the fight. The opposite is true, insists Uitto, who outspokenly regards lookaside as an Intel monopoly. "Most of the large operators agree with us and say it should not be a monopoly and there should be diversity," he said.

A common retort is that inline – with its tight coupling of proprietary software and custom chips – excludes Layer 1 from virtualization. But Intel's move to integrate hardware accelerators and fronthaul connectivity with its CPUs makes them look just as customized to Uitto. "How is that different from any other custom chip?" he said. "It's not a general-purpose processor. And if you build servers with only those CPUs, it would be ridiculous because you'd have all this overhead of hardware acceleration in the product cost and power consumption."

He points out that inline options are now available from a multitude of suppliers, including Nvidia and Qualcomm, besides Marvell. But what of the view that an operator with Intel servers and lookaside architecture could more easily switch DU software vendors by avoiding the need for hardware changes? Canada's Telus recently cited this as one of its reasons for selecting Samsung and lookaside technology to cloudify part of its RAN.

"You can still use the server and the SmartNIC is inexpensive," said Uitto. "The value of the SmartNIC is that it is extremely cheap compared with the overall server price. It is not a big deal. If you want to change the DU/CU supplier, you can still use the same server." The hyperscalers now eyeing jobs in the RAN use SmartNICs every day for efficient offloading of specific workloads, he added.

Technology disputes notwithstanding, open RAN has so far produced few smaller challengers to the old guard. South Korea's Samsung, an even bigger technology giant, has perhaps been the main beneficiary. Japan's Fujitsu and NEC have also picked up work. But Uitto says both Nokia and new entrants can build market share by doing it at the expense of vendors still opposed to open RAN. For now, that puts all eyes on Huawei.