At its pre-MWC briefing held in London on Thursday 13 February, Ericsson’s topline messaging emphasised that the industry needs to grow by creating a network and service infrastructure that enables new service and revenue opportunities. It is based on the horizontalisation strategy that it is being incorporated most publicly at AT&T.
The key buttons that kept being pressed were “open” and “programmable”. It is only by creating open, software-driven and AI-enriched networks that in turn enable the creation of differentiated connectivity that the industry will grow. CTO Erik Ekudden said, “It’s not so much [about] our share of the pie but the total size of the pie. Unless we extend that we will not see the investments in networks that are needed. Differentiated connectivity has not been exposed and monetised to the level it could be, so far.”
Head of Networks Fredrik Jejdling said, “Programmability is where we think we can differentiate. It’s the reason we moved to the horizontal architecture, so we can manage based on intent. For a base station to deal with multiple service intents at the same time it has to do it in real time. That requires significant amounts of AI-driven algorithms. For example, the sensing to enable MIMO requires 40 billion computations per second. It cannot happen without AI. So to create the services we have talked about we need to define a slice we must automate radio resource partitioning, get the right radio resources allocated. ”
Ericsson’s software launch as part of its 5G Advanced was about creating that independent software layer that can set energy targets, automate resuorces and enable that new programmability for the RAN.
This week, the company announced seven new radio products at the event and again the messaging here was around Openness and programmability. The company said by the end of 2025 it will have 130 radio products that carry the necessary hardware to support its chosen implementation of O-RAN Alliance fronthaul interfaces. That means, in the case of massive MIMO units, the RUs have the compute required to support OFH between the O-RU and the O-DU, using eCPRI to support the O-RAN Alliance’s ULPI categeory of fronthaul, which Ericsson helped push through specifications.
It added that the closest competitor has 20 such units, currently, but when asked later to specifiy which competitor they were referring to, chose not to do so. (It should be noted of course that O-RUs and O-DUs do not have to support the OFH category that Ericsson has chosen to support, in order to delilver M-MIMO.) The company added that during 2025 more than two thirds of shipments will be on its Open RAN-ready hardware.
Fredrik Jejdling, Networks, said that the company did not want to get into specifics of Open RAN interoperability.
“I’m not obsessed about integrating this radio partner with this baseband, that’s not the question, really. The way we try to design networks differently is based on open standards, and cloud, so that enables us to do different things in the software layer. It enables us to work with partners and to think about software differently. And for us on the radio layer to be more a more attractive partner, offering optionality if we don’t perform, gives us a bit more credibility with the customers.
“We are not going to be the one that associates ourselves with “now we integrated this Samsung radio with this baseband”, that to me is table stakes – openness in the interfaces. I’m more interested in talking about what it does, what type of use case we can enable, the benefits for our customers. Openness, cloudification, virtualisation, they’re all enablers to create something that is an architecture that is horizontal, that allows us to build more value for our customers.”
This mirrors Jejdlings comments this time last year, where he again avoided being pinned on Ericsson’s third party integrations. Since then, however, the company has announced that it will interface its SMO-RIC platform with radios from other vendors within AT&T’s network, notably Mavenir, Fujitsu, Corning and Commscope.
This openess should also extend, Jejdling said, to the company’s ability to run its Cloud RAN workloads in any hardware setup, even ones that rely on different architectures to Ericsson’s own integrated solutions and Cloud RAN implementation. That’s a discussion that has arisen not just around Open RAN, but is likely to become more relevant if the concept of the AI-RAN means that Nvidia, for example, is successful in convincing telcos that RAN workloads should run in GPU platforms that can then be leveraged for other compute functions. With Nvidia also marketing its own RAN software capability, the relationship between the GPU company and the major RAN vendors is muddied further.
Ultimately, if you are a virtualised RAN, if you’ve got virtual software, it needs to run on most types of underlying compute platforms, including that of Nvidia.
Hardware independence
Jejdling said that the company would need to consider porting its RAN software to Nvidia as part of its commitment to openness.
“Of course, we work with Nvidia, and look at the possibility of running our software. Ultimately, if you are a virtualised RAN, if you’ve got virtual software, it needs to run on most types of underlying compute platforms, including that of Nvidia.”
He recalled Ericsson’s earlier work with Nvidia, but counselled that any port to Nvidia GPUs, or Arm-based structures featuring in-line acceleration, would not necessarily come quickly, but would be led by customer demand, just as Ericsson’s decision to use its own customisation of Intel silicon was led by customers.
“Our first Cloud RAN software ran on Nvidia. We talked about it with Jensen (Huang, Nvidia CEO) at MWC Los Angeles in 2019.
“Then our roadmap partly got defined by our leading customers to choose Intel for their compute platform. So we evolved our own software stack with a partly bespoke new L1 implementation. ”
The customers will choose the underlying compute and hardware platform. So we need to be able to have a stack that is deployable across those various hardware instances
“Everything has a time, though you need to have parity and there’s a lot of things that need to be done, but NVIDIA, Arm-based architectures, AMD… ultimately, the customers will choose the underlying compute and hardware platform. So we need to be able to have a stack that is deployable across those various hardware instances.”
Nokia’s approach to this hardware diversity has been to encapsulate its L1 processing within an vDU card with inline acceleration on a Marvell SoC that can be plugged into other server chassis. Ericsson has long said it sees its approach being to fully abstract its RAN software from the underlying hardware architecture.
My point is that if you move into virtualising software, if you want to be true to that commitment, you need to run it on multiple hardware platforms. Otherwise you might as well continue buildling basebands
“We also ported over our current stack from Intel over to AMD,” Jejdling said, “I think we proved that three, four years back in Barcelona. It was actually quite little changes, even down on the L1 level, to port that software over to different hardware. We work with a bit of a different architecture, which is also important, because we work with look-aside [acceleration], as opposed to inline. You could argue that we’re almost too generous in a sense because it means that it is truly a virtualised layer, so you don’t need to move over your accelerator platform, it’s not integrated at all.”
“In my view a true virtualised platform needs to be cut above the hardware layer. So it takes a little bit more software engineering to be able to port it over. And of course, going from x86 on the accelerator level into GPUs, it takes a bit of work honestly at the application level. And I think to certain extent, our priorities on who we build on top of get defined by also what our customers choose.
We’re choosing to work all the way in on O-RAN Alliance specification. That’s the only thing we’re building on
“But my point is that if you move into virtualising software, if you want to be true to that commitment, you need to run it on multiple hardware platforms. Otherwise you might as well continue buildling basebands.”
Jejdling, though, did not want to be drawn into any likely timelines.
“It’s about resourcing what our customers want to do. We are of course heavily engaged with AT&T and it’s quite a massive part of Cloud RAN being introduced in that network. It’s a matter of securing what we do promise, and then evolving the deployability of the software.”
“You know, everything takes time. I just want to secure that commitment – we’re choosing to work all the way in on O-RAN Alliance specification. That’s the only thing we’re building on.
“And similarly, we choose to disaggregate the software layer to the hardware but we see those more as enablers for what you can actually do with the network.”
NEW RADIOS:
As at its 2024 event, Ericsson launched seven new radios.
Three flagship are the AIR 3266, a wideband Massive MIMO with 400W output, weighing 17kg. This will be available in Q2 or Q3 of 2026, the vendor told TMN.
The 4455 is an advanced FDD radio, with new wideband PAs, dual band FDD with 4x100W output power in 19kg, reducing weight by 20%. It’s available in Q4 2025.
Then there’s the AIR 3285, a “game changer”. Dual band FDD Massive MIMO radio, at 30kg the lightest in class, and also available Q4 2025.
And there was the launch of a new portfolio known as RAN Connect, a new fronhaul, synch aggregation unit designed to enable operators to aggregate more radio traffic to the fronthaul point. It would also serve to support Cloud RAN migrations, the company said.
Here, Ericsson was clearly making an appeal to authority and knowledge to emphaises the performance benefits it claims for its antennas and Radio Units.
The company’s Anna Dicander, Head of Radio Engineering said that radio engineering is both an art and a science, requiring mastering of disciplines accross silicon, algorithms, passive cooling, filter technologies and advanced PAs, “not in isolation but in how they interact”.
You could certainly draw this messaging line across not just to other major vendors, but to new entrants who are seeking to use Open RAN to open up the possibility of new markets within operator networks, by building units based on available commercial radio silicon and their own software, allied to cheaper manufacturing.
Jejdling said, “Radio development is becoming bespoke. I mean, our ASICs are bespoke radio ASICs. PAs [Power Amplifiers] are quite an advanced radio type of technology. Filters is something we’ve spent time on.
“I would argue if you go back a few years when it was about 1800/900 FDD radios the differentiation was not so strong. Now, when you build this kind of complex compute platform, which even radios are becoming, then differentiation and specialisation is actually higher,
“So our ability then to build PA technologies, filters, ASICs with software on top – that’s why we can build 400 MHz bandwidth radios at 16 or 17 kilos now. That cannot be done without the new generation of ASICs in them.”
Jejdling also emphaises that Ericsson’s scale can play a role in securing the supply chain for these crucial parts.
“For those three element to be secure we need to get enough scale and volume in the Western ecosystem. What used to be a highly specialised global business is going into segments, basically, so we work very tightly with a few vendors or partners both PAs, filters, ASICs, to secure that we get a competitive, innovative Western ecosystem.”
