The "open RAN" love affair between AT&T and Ericsson was on gratuitous display this week in Dublin, the new location of the Telecom Infra Project's annual FYUZ event. Last year, when the show was still in Madrid, Ericsson had been confined to a fleeting televised appearance by Fredrik Jejdling, head of the mobile networks business group, after the Swedish vendor had persuaded the O-RAN Alliance, the group in charge of open RAN specifications, to accommodate its "uplink" demands (more on that later).
This year, Ericsson is a full convert. Quietly announcing its membership of the Telecom Infra Project, Ericsson seemed to be running parts of the show. Entire sessions were dedicated to proving that its $14 billion contract with AT&T, struck weeks after the Madrid 2023 event, really is an open RAN, despite the absence of many other suppliers.
Helpfully, for Ericsson, FYUZ had also switched to a new chairman, introducing AT&T's Rob Soni in place of Yago Tenorio, who recently moved from Vodafone to Verizon. "The O-RAN Alliance was out to lunch," railed Soni in front of FYUZ attendees, blaming the specs body for a "five-year deadlock" on talks about managing the uplink for massive MIMO, an advanced 5G technology. It took the combined muscle of AT&T and Ericsson to break that deadlock, he said. "It would have been a pyrrhic victory to have no support for massive MIMO and limited our ability to deploy and look at the use of a third party."
But on the same day that AT&T and Ericsson were lauding their achievements, Ericsson published a curious update on open RAN progress. It includes this interesting sentence: "To further strengthen the open RAN portfolio, Ericsson is adding OFH Category B with ULPI to our cloud RAN and selected massive multiple input multiple output (M-MIMO) radios, starting in 2025."
Romancing the Soni
In plainer English, and without any reference to purpose-built products, this essentially promises support for the interface described by Soni only in Ericsson's cloud RAN products – and, even then, the implication is that only some of its radios (the critical word being "selected") will include that interface. It does not seem like an accidental omission because Ericsson highlights purpose-built RAN in the next sentence about more basic radios (as opposed to massive MIMO). "As for Category A, it is being incorporated into our latest generation of purpose-built RAN, cloud RAN, and enhanced Common Public Radio Interface (eCPRI)-capable radios, also starting in 2025," it says.
Purpose-built RAN, of course, accounts for a much bigger share of the global footprint, a fact that won't change for many years, however quickly AT&T moves. If Ericsson did have a plan for O-RAN-compatible massive MIMO products in this sector, it would seem unlikely to be implemented before 2026, at the earliest.
This all sounds very wobbly in comparison with the much firmer commitment that Ericsson previously made, when it said: "Our purpose-built RAN solution, equipped with Ericsson's specialized RAN compute hardware, will fully support open RAN fronthaul. Furthermore, all future Ericsson radios will also be designed to incorporate open RAN fronthaul."
Subsequent exchanges with Ericsson, however, have revealed some extra details about the latest update. Cat B ULPI, a flavor of open RAN for massive MIMO, will feature in eCPRI-capable radios, eCPRI being a more "open" update to the "closed" CPRI fronthaul link. Older radios will have to continue living with CPRI and therefore won't ever be opened, although at some point they will presumably be replaced with open RAN equipment.
But there is no commitment to open RAN for massive MIMO on the purpose-built RAN compute side of things. There are plans to support Cat A, the open RAN spec for less demanding 5G, in the 6672 processor, which Ericsson previously described as high-capacity hardware for the most advanced 5G deployments, including rollouts of massive MIMO. What about support for Cat B ULPI in 6672? There are no plans to support this, Ericsson confirmed by email.
On cloud RAN, moreover, AT&T is not moving very quickly at all, say multiple sources. The network as it stands is largely purpose-built and the investment in cloud RAN has been tokenistic so far. Nor have the companies said what share of the network will ultimately be cloudified. If the RAN is truly open, the FYUZ session led by Soni before a packed audience certainly wasn't, giving attendees no opportunity to ask those awkward questions ("What's the footprint split between cloud and purpose-built RAN?" for instance).
The faster AT&T moves on purpose-built RAN, the smaller the gap left for cloud RAN. This doesn’t by itself shut the door on other suppliers if those O-RAN Alliance interfaces feature in cloud RAN products. But Ericsson will be AT&T's only provider of purpose-built RAN compute after the telco has finished replacing Nokia at a third of its sites. If the 6672 range, highlighted in this release from November 2023 as a groundbreaking tech, doesn't support Cat B ULPI, Soni may be staring at a pyrrhic victory after all.
Small sell for Fujitsu
For anyone hoping to see AT&T invest in a multivendor massive MIMO network, the other signals are not great, either. Fujitsu is the only radio unit partner that AT&T has announced besides Ericsson so far. As with cloud RAN, there has been no indication of how many sites it might support. Fujitsu does boast a portfolio of massive MIMO radio units, but the latest update from FYUZ this week suggests they will not feature in the AT&T macro network.
"We announced Fujitsu as a partner specifically on the journey for the deal we did with Ericsson and we've been working quite closely with them on the requirements of the products specifically for small cell radios and they have committed to support us and Ericsson on that," said Soni. "Hopefully mid-next year will see an opportunity for them to demonstrate their capabilities on the radio side."
What's more, there has been almost no progress on diversifying the market for cloud RAN processors beyond Intel, whose close partnership with Ericsson and current difficulties fuel constant speculation. A table in Ericsson's latest update names Intel as the sole "commercial-grade" solution for central processing units, accelerators and NICs (network interface cards). AMD and its Xilinx subsidiary are the only other chipmakers Ericsson identifies, falling into an "active engagement" column.
But they have been actively engaged for more than a year and a half with nothing to show for it. This week, AMD said Intel's move to integrate an accelerator and a CPU, on the same die, would create "lock-in," rather than the hardware and software separation that Ericsson claims to champion. Others, meanwhile, continue to argue that the FPGAs used by Xilinx for acceleration are too power hungry. Arm, a rival architecture to the x86 technology produced by Intel and AMD, is seen as an option for cloud RAN. But not a single company building Arm-based chips is even in active engagement with Ericsson, it seems.
The uplink specification that took shape after AT&T, Ericsson and others intervened is already controversial, it is worth noting. In the interests of keeping radios as simple as possible, and helping to cultivate a market of smaller vendors, the O-RAN Alliance in its original specification had kept most of the uplink functionality in the distributed unit (DU). As far as Ericsson and its pals were concerned, this would place a heavy traffic burden on the fronthaul link.
But the fix preferred by Ericsson means building more complex radios, featuring equalizers as well as other uplink functions. Vendors will still be able to build O-RAN-compliant radios without equalizers, but DUs must always include them under a "compromise" reached last year to satisfy the demands of other parties. AT&T, interestingly, was on the opposite side of the fence from Ericsson, which has also indicated it will not build radios without equalizers. Vendors including Samsung and Mavenir, meanwhile, have insisted there was nothing wrong with the original spec. If all this sounds confusing, complicated and the opposite of what standardization and streamlining should be about, that's because it is.
You May Also Like
