Ericsson is taking a very different approach to the open, virtualized RAN, to that of Nokia. The defining differences, currently, between these two mighty rivals are their future RAN propositions, and their approach to enterprise. Nokia moved more quickly into fully software-based networks than its rivals, and then adopted the persona of open RAN’s cheerleader to help offset its 5G difficulties. It has also been more wholehearted about selling services to enterprises, with or without the help of its operator customers (see separate item), and has only intensified this effort as its core business has run into challenges.
Ericsson pulled back from direct-to-enterprise sales when its current CEO took the helm, though it has been steadily building up its industrial 5G offerings and may relax its view in future. Indeed, a clear target for its new Cloud-RAN products is likely to be enterprise and campus networks.
In the 5G RAN, it has performed strongly in the first, conventional deployments, particularly with its 5G-ready Ericsson Radio System. For the next wave of 5G, it faces bigger dilemmas, one of which is how far to assume operators really will deploy large-scale, open, disaggregated vRANs in their macro networks within 3-5 years.
It has unveiled its 5G virtualized RAN portfolio, a few months after Nokia and Samsung launched theirs, but it is notably more cautious about open RAN architectures. There is plenty of virtualization, but very limited openness. In fact, there is a clear line from Ericsson’s previous launches of cloud network systems to this disaggregated, 5G-centric platform (it is easy to forget that the big vendors have already launched Cloud-RANs for 4G, even if they have been only rarely deployed at scale, and Ericsson has been more cautious than Nokia in this area).
The cautious approach to open RAN is in line with the Swedish company’s recent stance on this topical subject. A member of the O-RAN Alliance, it has nevertheless raised concerns – which are bound to seem self-interested, but do carry some weight with operators – about potential security and performance trade-offs. Its attitude seems to be to watch and wait, trying to offer a reality check on the wildest hopes of the fully open network, and so defending the closed ecosystem that has served it so well.
In the greenfield and localized (enterprise or rural) networks, this caution is a risk because challenger vendors, as well as Nokia, could establish an unassailable position in a market where there are no entrenched suppliers and where the performance and scale requirements are far lower than in the macro RAN. In that heavily loaded 5G macro network, the caution looks far more realistic, however self-motivated it may be.
Last month, CEO Börje Ekholm, announcing solid quarterly results, took a somber line on the migration to new networks, saying: “In the legacy portfolio, sales are falling faster than we previously expected. In order to combat this we have increased our investments in the new cloud-native 5G portfolio and we are seeing a good win ratio, but it’s also fair to say it has not yet generated any significant sales and therefore does simply not compensate for the fall in the legacy portfolio.”
He added: “I don’t really see Open RAN to have a major impact in 2021-22 timeframe, but after that I think it will start to impact revenues for us – it will start to impact the way business models evolve going forward.”
And, launching the new Cloud-RAN offerings, Per Narvinger, head of product area Networks at Ericsson, told SDxCentral: “Many of our customers are still not even fully evaluating open RAN”. While some customers were running trials with Ericsson and others, most had yet to consider virtualized RAN seriously, let alone a fully open architecture.
However, this was a different tone from that sometimes struck by Ericsson executives, which have sometimes been wholly dismissive of open RAN. Caution about timescales, but recognition that Ericsson needs to be prepared for a significant change in architectures and supply chain – whether that is based on O-RAN itself or something else – were the hallmarks of Ekholm’s statements, and of the Cloud-RAN launch.
Where the new portfolio does tick the modern 5G boxes is in the cloud platform. The first family members are a virtualized distributed unit (vDU) and centralized unit (vCU), both of which can run on any Intel-based COTS hardware without special acceleration and are for low band 5G spectrum. The hardware can be Ericsson’s own boxes, but it may also be the operator’s edge servers, or a third party cloud – a significant change of emphasis from previous Cloud-RAN announcements.
There is also a Network Gateway that interfaces between the vDU and already installed Ericsson remote radio heads. The gateway translates the vDU’s native eCPRI fronthaul protocol into CPRI, in order to support legacy radios. The gateway also hosts the Ericsson Spectrum Sharing (ESS) technology, which allows 4G spectrum to be allocated dynamically to 4G or 5G, greatly boosting spectral efficiency. This has been an important differentiator for Ericsson 5G, since it appears to have stolen a lengthy march on its competitors in getting this technology to market. That will be particularly significant for operators with limited access to midband spectrum, such as Verizon and AT&T, or which want to launch 5G before new spectrum auctions are held, like Vodafone Ziggo in The Netherlands.
The focus is all on sub-GHz spectrum for now, since those networks have limited capacity and therefore lower traffic and computation demands than 5G RANs in midband or millimeter wave spectrum. More units will follow eventually, but Ericsson pointed out that some vDU implementations, especially those required to support demanding near-real time operations such as dynamic spectrum sharing, may stretch the definition of ‘COTS’, and require cloud hardware with sophisticated (and possibly expensive) acceleration.
The Cloud-RAN systems will be available in late 2021, by which time it is possible that some of the performance concerns about COTS hardware for vDUs may have been addressed by the chip providers. This will be important for driving down the cost of Cloud-RAN, which is a significant barrier to some operators.
Narvinger told LightReading: “It would be sad if we build something that is only for one purpose and we cannot do it more cost effectively than with general-purpose hardware.” But he did acknowledge that there are other potential benefits to a cloud platform – “you can potentially have a richer ecosystem around the solution, and you can leverage automation and AI in a larger way”.
The vCU can also use the 3GPP F1 interface to Ericsson’s midband ERS and to its higher band mini-macro base stations, which can be mounted on street furniture. The key in all this is to introduce Cloud-RAN gradually, where it is required to boost performance and flexibility, alongside an existing RAN. That is an attractive prospect for many operators, since it does not entail the risk and disruption of a big bang migration or even a rip-and-replace exercise – and of course, it is very attractive to Ericsson, since it defends its installed base and keeps the Cloud-RAN transformation firmly single-vendor.
However, it is sure to tick open interface boxes where these are required by operators for the reassurance that they are not really locked-in. And Ericsson is not dismissing the O-RAN Alliance, of which it is an active member. In line with Ekholm’s view, executives said the vCU was designed to support some O-RAN interfaces, notably to the O-RAN Non-Real Time Radio Interface Controller (NRT-RIC), when this becomes commercialized (or ‘if’ it does, was the subtext).
The O-RAN Alliance’s fronthaul interface may be the most basic enabler of an open RAN, allowing radio units and distributed or centralized units from different suppliers to interoperate. But its more disruptive and significant contribution to the architecture is the RIC, which moves many functions traditionally embedded in the base station up into a higher layer of software based around xApps. These could come from different vendors from the virtualized base station hardware and software, and as such are a target for suppliers from many fields, and a potential threat to a key area of expertise and control for the traditional RAN vendors. So far, Nokia has worked hard to neutralize that threat by contributing much of the seed code for the O-RAN Alliance RIC, though a greater challenge may come from a separate, more operator-driven implementation being driven by the Open Networking Foundation, called SD-RAN.
Ericsson, unlike Nokia, has not yet sought to seize control of the RIC itself. It is taking the view that the non-real time RIC is an area which can logically be abstracted from the base station and even opened up, hence the support in its new vCU – indeed, it says it already supports many of these functions in software anyway. But when it comes to the near-real time RIC, the centerpiece of O-RAN’s disruptive efforts, it believes there are too many performance and reliability trade-offs compared to embedded solutions (and it may well be right). Ultra-low latency, ultra-reliable services such as dynamic spectrum sharing will struggle to be fully supported in xApps living above the base station, argues Ericsson.
And for now, even the non-real time RIC is a future consideration for operators, it says. The concern for MNOs with existing RANs is to introduce vRAN in a complementary manner, and to make the new vCUs and vDUs work smoothly with existing network management systems (NMS). It will study which open interfaces make most sense in future, but it seems likely that the O-RAN Alliance’s O1 interface to the non-real time RIC will be commercialized in the near future but not the E1 interface to the near-real time RIC.
But such support is bound to look grudging when set against Nokia’s enthusiasm. Ericsson remains clear that an integrated network, in which the radio units and basebands are optimized by their own supplier, will perform better than a multivendor network. It will only support third party software through its own ‘system-verified’ process, and it questions the idea that open RANs will be integrated by third party system integrators, arguing that large-scale RANs can only be effectively optimized by the makers of the radios.
The trade-offs become really glaring in Massive MIMO systems, especially in millimeter wave bands, Ericsson argues, and this is where its claims are most persuasive, since most operators are struggling with the challenges of optimizing networks in high frequencies anyway.
We would expect any new open interface support to be seen first in lower bands and simpler radio/antenna units with only 2T2R or 4T4R configurations. And even in midband spectrum, those computational performance loads become challenging for COTS and for open fronthaul, argues Narvinger, which is why support for popular bands like 3.5 GHz will come later than low band spectrum. He said: “Those networks become so computationally heavy that you need accelerators”. Ericsson is investing heavily in R&D in this area, which could in future support a open RAN offering – one project, working with Nvidia, is experimenting with the use of GPUs (graphical processing units) as the basis of accelerated vDU platforms.
Of course, Samsung and the challenger vendors are bound to be more keen than Ericsson on open interfaces, because these lower barriers to the RAN fortress, but Nokia has the same territory to defend as Ericsson and the even more cautious Huawei. The Finnish company has clearly decided that it will try to make a virtue out of necessity and seize leadership of the open RAN, and in so doing, hope not only to gain market share but to shape the new platforms in a way that suits its own technological strengths.
In June, Nokia introduced the new generation of its Cloud-RAN architecture, with strong support for the O-RAN Alliance’s open interfaces. Nokia has been more advanced than its competitors in Cloud-RAN. It adopted elements of a virtualized RAN architecture almost a decade ago in its Liquid Net portfolio, and opened itself up to the idea of running its software on third party hardware, when that was still a radical idea in telecoms networks.
Arguably it moved too early, adopting Cloud-RAN architectures before current thinking – about containers, for instance, or distributed units – was formed. But it has stored up considerable software expertise, and just as important for the large MNOs, experience of integrating large-scale virtual networks.
Nokia has made it clear that it intends to take the lead in supporting open interfaces, despite all the risks that presents for its traditional economics and its hold over customers. In O-RAN, it appears to be hoping that, if O-RAN (or another open architecture) prove to be unstoppable, it will have seized competitive advantage from its main rivals, while also offering an end-to-end, fully telco-grade solution.
In the best case for Nokia, operators will opt for single-vendor O-RAN implementations, comforted by the ‘get-out clause’ of the open interfaces and swappable hardware and software elements. In a less ideal scenario, they will want fully multivendor networks but will need a trusted, large-scale partner to do the integration and roll-out, which would see Nokia working with challenger suppliers, but keeping hold of the pivotal role in the value chain.
Nokia’s initial open RAN offering is based on key elements it already offers, such as the second generation 5G AirScale Cloud-RAN platform, and on some of the work it did for Rakuten to open up its interfaces for third party virtual network functions (VNFs) or hardware elements.
The “fully cloudified and disaggregated 5G base station” will be commercially available to selected customers this year, with wider availability in early 2021. And while O-RAN support has grabbed the headlines, the really important improvement on the first 5G AirScale release is to introduce the virtualized distributed unit (vDU). This is where Nokia can be argued to have moved too early in vRAN, adopting a very centralized architecture – which was the early thinking about the platform.
However, that is challenged by any RAN services that require low latency, real time response, or which involve very high levels of signal processing. These are better supported in distributed units, closer to the radio units on the cell site (and sometimes collocated with it), and based on cloud hardware with specialized accelerators to handle the very intensive processing. The simpler CU/RU architecture has been implemented in some Nokia projects, such as AT&T’s and Verizon’s millimeter wave 5G networks, but the fully virtualized DU option will be important for wider appeal, especially when operators starting thinking about low latency applications with their upgrades to the more industrial and IoT-focused 3GPP Releases 16 and 17.
With the new vDU, both non-real time and real time RAN functions are virtualized and run on a cloud stack in x86-based Nokia AirFrame Open Edge Servers and Rackmount Servers.
And Nokia points out another of the key advantages of the three-tiered architecture – which also lies at the heart of O-RAN. This is the ability to use the same distributed cloud infrastructure to run other applications as well as the RAN functions, such as low latency IoT services and analytics for enterprise users, or slice management.
The new AirScale architecture supports the O-RAN taxonomy – radio unit (RU), DU, radio access point (RAP – an option which combines DU and RU), CU and RAN Intelligent Controller, all connected by the various O-RAN interfaces. The vDU and vCU can be colocated at a central or regional edge cloud, or the vDU can be decentralized to an edge or far edge cloud, or the vDU can be located at the cell site. Nokia said: “Such flexibility is of paramount importance due to varying topologies and transport network needs even within the same network.”
The O-RAN interfaces, and the RIC, are in different stages of maturity, but Nokia has been careful to stamp its mark on several of them, most notably the RIC. The RIC only operates within O-RAN networks so will not always be present in AirScale deployments. Nokia said: “Not every Cloud-RAN implementation is an Open RAN case. Also, not all Open RAN cases will be implemented using Cloud RAN/vRAN, even though it is fair to assume that most will.”
Samsung followed just a month later with the launch of a fully virtualized RAN (vRAN) with support for the open RAN interfaces and architecture. Its first 5G vRAN offering includes vCU and vDU, and software that can run on dedicated hardware – for more demanding, real time processes that require acceleration – or on x86 COT servers. The software is also designed to make it easier for operators to combine different types of hardware and move the network functions between them.
The CU is already in commercial use in Korea, Japan and the USA and the new step is to move the DU to the virtualized architecture, a far more demanding task. Samsung insists that its vRAN can match the performance of a vertically integrated RAN with dedicated hardware, though the DUs will rely on sophisticated accelerators.
Alok Shah, VP of networks strategy at Samsung USA, told SDxCentral: “This idea of moving the functions into x86-based servers, just a whole new ecosystem of hardware that’s available that’s easier to upgrade, easier to obtain, easier to spare — there’s just a whole set of ways that the opex can be substantially improved. Samsung doesn’t build that equipment but our software runs on top of it.”
The hardware Samsung does provide includes the radio unit (RU), and the company offers a combined RU/DU in a single box, which it calls an access unit. This is mainly designed to reduce cost, space and power in dense millimeter wave networks. Partners include HPE for 5G core software and edge computing, and Qualcomm, Marvell and Xilinx for base station components.
Like many vendors, Samsung has moved away from a purist view of the vRAN, or of cloud-native networks in general, and is presenting operators with a possibly less daunting vision – a pragmatic roadmap in which existing and even new physical elements will coexist with virtualized and cloud-native functions for many years, all coordinated by a single orchestration platform, and with migrations and swaps simplified by adoption of open interfaces.
“Our recommendation is not that every operator run out and deploy vRAN for every portion of their network,” Shah continued in the interview. “You have to consider the processing requirements that a given function takes based on spectrum, based on bandwidth requirements, and so you have to look at the economics of doing that all in software versus kind of more traditional hardware-based approaches.”
He believes, for instance, that vRAN is most economic, and has the biggest impact on flexibility, in low and middle spectrum bands, while in millimeter wave bands, “dedicated hardware can more cost-effectively support the amount of processing required for large swathes of spectrum and at least 1Gbps speeds on data throughput,” because of the huge processing requirements.