Samsung is open RAN's big winner so far as specialists miss out

After deploying thousands of 5G radios engineered to support one of the original open radio access network (RAN) specifications, Samsung was obviously unimpressed when Ericsson and pals complained it was not good enough. The O-RAN Alliance, the body in charge of specs, last year approved optional changes in response to those complaints. But the multiplicity of choices now risks a fracturing and even shrinkage of the ecosystem, the South Korean electronics giant has warned.

"We worry a great deal about fragmentation of the ecosystem," said Alok Shah, the head of strategy, business development and marketing for the networks business of Samsung Electronics America. "The goal of open RAN was to expand the ecosystem, and I think that putting more intelligence into more advanced radios actually runs the risk of shrinking the ecosystem because there are only so many vendors that can build the kind of advanced radios being discussed."

Ericsson and other stakeholders, conversely, had argued that leaving uplink functions in the distributed unit (DU), a server box for baseband processing, instead of moving them to the radio would upset performance in massive MIMO, an advanced, antenna-rich 5G technology. Unable to arrive at a consensus, the O-RAN Alliance signed off on two optional modifications at a meeting in Japan last year. One, backed by Ericsson, adds radio support for uplink functions including the equalizer, which addresses interference. The other modifies radios to handle uplink functions except the equalizer. Any compliant DU must be equalizer-equipped to sync with both types of radio.

Including the original 7.2x specification, this means there are now four flavors of open fronthaul, notes Shah: Category A, for more basic radios; Category B, for massive MIMO; Class A (the radios-with-equalizers update); and Class B (radios without equalizers). And yet there was nothing wrong with Cat-B radios in the first place, insists the Samsung executive. "We've already deployed many thousands of massive MIMO radios using option 7.2x Cat B and performance has been very good," said Shah.

Nevertheless, unlike Ericsson, which has said it will not produce radios minus equalizers, Samsung aims to do whatever the market dictates, Shah told Light Reading at this year's Mobile World Congress event in Barcelona. "We will certainly support whatever the ultimate decisions are, and our DU will support the four variants of radio," he said. "I guess our customers will lead us in the direction they want to go."

Single vendor suits the strongest

Covering all the bases, whatever the early investment implications, seems to be a critical tenet of the South Korean vendor's 5G strategy as it targets a RAN market that has been dominated for years by Ericsson, Huawei and Nokia. Open RAN, whose specs allow a telco to combine vendors at the same site more easily, was supposed to boost specialists previously excluded from tenders because they lacked a full product portfolio. Yet the giant Samsung, part of a group that dwarfs Ericsson and Nokia, has been open RAN's big winner so far.

"We've been number one in open RAN and number one in virtual RAN for a number of years now," said Shah, pointing out that Samsung currently provides products that tick both the open RAN and virtual RAN boxes at more than 38,000 sites. After setbacks in India – where market leader Reliance Jio has moved from a 100% Samsung rollout in 4G to introducing both Ericsson and Nokia alongside it in 5G – Samsung has watched its RAN market share outside China fluctuate between 9% and 11% in the last three years, according to Remy Pascal, a principal analyst at Omdia (a Light Reading sister company). But he agrees that Samsung has been the market leader in open, virtual RAN, ahead of Japan's NEC and Fujitsu.

Several factors may explain this lead. The first is the phenomenon of single-vendor open RAN, whereby a telco demands specs-compliant products but takes them mainly or entirely from the same supplier. Dell'Oro, another market research firm, reckons between 50% and 83% of all open RAN deployments will fall into this category by 2028. The second factor in this environment is Samsung's end-to-end capability, which specialists lack by their very nature. Finally, there is Samsung's legacy position as a provider of traditional and highly regarded 4G equipment before open RAN became an industry topic.

While others on both the telco and vendor sides think single-vendor open RAN is a contradiction in terms, Shah – perhaps unsurprisingly – does not see it as problematic. "Our view is that single vendor as a starting point makes a lot of sense because the truth of the matter is that virtual RAN or cloud RAN is quite a complex transition for an operator," he said. "But you need to make sure you have the optionality so that down the road if you want to introduce a third-party radio, and are unhappy with the existing vendor, then you have the option to pivot."

Opponents will undoubtedly view Samsung's criticism of the recent optional modifications to 7.2x with suspicion. Just like Ericsson, which already claims to have 1 million radios in the field compatible with Class A, Samsung is probably more interested in safeguarding its investments and installed base of radios than in nurturing a multivendor market. Yet Shah says his employer can boast a "longer list of third-party OEM [original equipment manufacturer] partners" than any of its rivals.

On the radio side those include NEC, recently paired with Samsung at a Vodafone site in the UK, as well as Fujitsu. Less well-known partners are Jabil, a manufacturer headquartered in Florida, along with South Korea's Ace and Solid. Samsung also collaborates with Ciena and Juniper on cell-site routers (CSRs), and it has recently been able to move some of Juniper's CSR functionality to a DU server. "The goal is to take as much hardware out of that site as possible," said Shah.

The latest addition to Samsung's product portfolio, though, is systems integration expertise. "We feel confident we understand the scope of what's required to be an effective systems integrator," said Shah. "That is a capability we are offering now to our customers." It seems a necessary move simply because many operators will probably continue to lean on their DU software vendors for systems integration.

Indeed, even Vodafone, which insists it is the systems integrator for its own UK deployment, appears to have relied heavily on Samsung when pairing the South Korean vendor's software with NEC's radios. "At that level, when it comes to integration with a third-party radio, I would say Samsung is pretty heavily involved," said Shah. "The operator helps to ensure both parties are swimming in the same direction – that we are communicating effectively and all those things – but I think the vendors have to take a pretty active role."

Practicing agnostic

Samsung also sounds less religious about the approach to RAN virtualization than either Ericsson or Nokia. The schism between the Nordic vendors is essentially about the extent to which the RAN software should rely on the server's central processing unit (CPU) rather than on specialist silicon. Ericsson thinks those CPUs, provided mainly by Intel, are today good enough for all but one chunk of resource-hungry software that does forward error correction (FEC), which falls under the Layer 1 umbrella in the RAN software stack. But Nokia believes the whole of Layer 1 needs to be "accelerated" with specialist silicon.

So far, Samsung's virtual RAN deployments have used Intel's Ice Lake processors along with an accelerator card for the FEC. It is, however, moving to a newer Intel processor called Sapphire Rapids EE, which puts the accelerator on the same die as the CPU. Tests have also been carried out with Granite Rapids D, which accelerates functions besides the FEC and is due for commercial launch next year.

While Shah downplays the effort involved, these upgrades demand some degree of software reconfiguration by Samsung. "The architecture continues to advance on the Intel side of things, and so it isn't as simple as going from 32 cores to 48 cores and nothing else changing," he said. "There is further acceleration we want to take advantage of."

In addition, Samsung has been lab-testing Siena-branded processors from AMD, Intel's sole rival in the market for x86-based CPUs. Thanks partly to this x86 commonality, Samsung has been able to avoid a big rewrite of the software when deploying on AMD rather than Intel, although its tests with AMD have not employed any accelerators.

But the real evidence of Samsung's agnosticism comes via other partnerships. Besides relying on x86-based CPUs, it has been able to put RAN software bar the Layer 1 functions on the Arm-based Graviton chips developed by AWS, running inside the Internet giant's Outposts servers. For Layer 1, in what has so far been a test environment, Samsung used silicon from Marvell Technology, which also caters to Nokia and features in Samsung's purpose-built 5G kit.

All this essentially means Samsung has two software development tracks – one, sometimes called "lookaside," that relies more heavily on the CPU, and the "inline" alternative of using specialist silicon for Layer 1. Shah insists "there is a fair amount of portability," but it is likely to exist outside Layer 1. Proprietary code written for Marvell's chips would look very different from anything usable on an Intel CPU. Here, at least, Samsung can probably replicate what it has done for its purpose-built products.

The company's view seems to be that lookaside is more prevalent, while inline may prove necessary in the most congested areas. "So far, Intel has been the platform of choice in all commercial deployments," said Shah. "We feel that ultimately there may be situations where an operator wants to use lookaside for the most part. But if there are sites that are high traffic, where they want to get an additional boost, then an inline card may make sense."

Samsung's hedging-all-bets approach extends beyond radios and chips into server equipment and the cloud. It is collaborating with Google and Microsoft, besides AWS, and has relationships in place with Dell, HPE and Supermicro, the three big makers of common, off-the-shelf servers, said Shah. "It's important for operators that have already made decisions around their IT infrastructure to be able to choose their preferred server provider, and that's fine with us," he said.

If there is one related technology area where the South Korean vendor wants to avoid dirtying its hands, it is the development of virtualization and cloud platforms. "We focus on things we do really well, and if there are partners out there that already have very good products, like Wind River, VMWare, Red Hat and AWS, we don't know why we would need to build our own," said Shah.

Operators increasingly say they want "multitenancy" cloud infrastructure that can support the full spectrum of third-party applications. Yet Ericsson, perhaps Samsung's chief RAN rival, still markets a cloud-native infrastructure solution (CNIS) that was clearly not designed to be this kind of all-accommodating host. After publicizing a $14 billion "open RAN" deal with Ericsson last December, AT&T acknowledged CNIS would be in the mix alongside Azure Operator Nexus, the Microsoft platform it is using for its 5G core. Some telco peers will regard that as one platform too many.