While there are many discussions about virtualized, disaggregated RAN – often in the context of Open RAN initiatives – for most operators this remains a next-generation option. Indeed, many are only just deploying the less disruptive Centralized RAN (C-RAN) at scale.
C-RAN involves a shared baseband processing platform that can support a number of cell sites, potentially a very large number. The key difference from vRAN is that the baseband does not need to be virtualized, and so operators can achieve some of the upsides of resource pooling without the daunting migration to cloud-based architectures, at least until these platforms become more mature and de-risked.
The bigger the number of sites supported, the greater the benefits of scalability and resource flexibility, but there are trade-offs in terms of latency between the radio unit and baseband. While early C-RAN attempts (called base station hotels) often focused on high degrees of centralization and sharing, in 5G, the need for increased numbers of smaller cells, together with lower latency requirements, is tending to drive a more localized topology, or at least a multi-tiered design in which cells are first centralized fairly locally, and then non-time-sensitive processing from these ‘clusters’ is aggregated to a data center.
This multi-layered approach prefigures the disaggregated vRAN architectures that are just starting to arouse interest, in which baseband processing is split between localized distributed units (virtualized or not) that are close to the cell site, and centralized units (always virtualized) in the data center or central office.
As with vRAN, the complexities of implementing even simpler C-RAN are increased in 5G compared to 4G, and so some operators that embarked on 4G centralization are yet to do the same in 5G, while others are waiting until they can also introduce vRAN. But some are seeking near-term benefits of C-RAN, including US Cellular, whose principal RAN engineer, Reggie Collette, was speaking at the recent LightReading Transport & Networking Strategies event.
Collette described a recently completed, though limited, C-RAN deployment and said:
“We’re accelerating our centralized deployments.” The operator, the USA’s fourth largest MNO with 5m mobile subscribers in 24 states, now aims to deploy C-RAN in 10% of its network by the end of next year and 20% a year after that. This will mainly be associated with its 5G roll-out in its C-band spectrum but there are no plans to introduce vRAN to the macro network architecture until a later stage, possibly 2025.
Collette said the C-RAN sites delivered significantly improved data rates – between 10% and 50% better – to customers because of the efficient site coordination enabled by C-RAN. Capabilities such as Coordinated Multipoint (CoMP), first introduced in 4G, have always been a key driver for C-RAN, as they allow multiple sites to cooperate and interact to handle large traffic loads in the most efficient way.
Collette also said US Cellular reduced costs as part of its C-RAN deployment because, at the same time, it moved from Ethernet to dark fiber for backhaul, implementing a 10-Gig circuit to its Central RAN hub. This highlights the critical importance of high quality fiber for the fronthaul links between radio units and the centralized baseband, and the availability and cost of low-latency fiber is a key determining factor in whether C-RAN is appropriate for certain operators and markets.
Common fronthaul interfaces will help to reduce risk and future-proof deployments, as well as allowing operators to choose from a wider range of radio unit suppliers than in the traditional network – some Taiwanese vendors such as MTI, for instance, make remote radio units that can interact with third party basebands provided the interfaces are opened up. The Open RAN movement, though mainly associated with vRAN, has defined Open Fronthaul as an interface that can be used in virtualized or non-virtualized networks, and some of the early (mainly rural) deployments have included non-virtualized baseband units and a variety of different radio form factors and suppliers.
Collette added that deploying C-RAN is “half art and half engineering” – the RAN and transport network teams need to cooperate in a closer way than before, and coordinating their respective workstreams has been one factor in the relatively long timescale to deploy the new architectures.
He added: “We’ve removed all the basebands from those macro sites, and we condensed them all into a macro hub. In this particular case, we’ve got three hubs in this cluster. But really, the only thing left out at macro sites anymore is just radio,” said Collette.
Fronthaul is based on active-active wave division multiplexing (WDM) fiber solutions to maximize efficiency of roll-out and network visibility, but Collette cautioned that C-RAN and later vRAN are far from being plug-and-play – the ultimate goal of initiatives such as Open RAN.
At the same event, Ericsson’s director of sales support, Vikas Khera, said: “I think we’re trending more toward centralized RAN” (this from a vendor that has been cautious about the downsides of these architectures in 5G).
But others are more cautious. Verizon, which has deployed some C-RAN clusters in 4G and 5G, and is now moving towards Open RAN, pointed out that there are many barriers to C-RAN. “The use cases have been narrowed down over the years,” said Mark Watts, a member of Verizon’s technical staff, highlighting the need for fronthaul fiber and power to be available and affordable to make the case. He believes Verizon would need to invest in 100-Gig backhaul to support an aggregated C-RAN, but it has 10-Gig links at most 5G cell sites, so this capacity increase would need to be cost-justified.
Like many operators, Verizon has mainly deployed C-RAN not in the main public network but in dense environments such as stadiums, where many radios are required to support high, concentrated traffic levels, but sharing does not have to take place over long distances. He said that a step towards macro C-RAN might involve “hubbing” a few sites, such as a macro site together with a rooftop radio and a small cell.
The pace of deployment in the USA, as elsewhere, is slow, and that makes the timeframes touted for full vRAN look over-optimistic, since that architecture brings the same issues of fronthaul capacity and latency, but with the added complexity of virtualization.
Back in 2019, Nokia laid claim to the world’s first fully cloud-based 5G vRAN, deployed Verizon in Dallas. The roll-out used a combination of central cloud-based functions and those based on distributed units at the cell site. Yet still, only a handful of operators, mainly greenfield, have deployed a pure cloud-native, containerized RAN.
Of all the technologies which have been touted as transformational in mobile network economics, the virtualized RAN (vRAN) has been the one with the biggest promises and the slowest progress. The industry has to hope that this is a case of a tortoise, making a slow but steady path towards a genuinely game-changing destination – and not a Galapagos turtle, large and impressive but doomed to an evolutionary dead end.
Because the vRAN really is key to the promised cost economics and service flexibility that will make 5G a different beast from 4G. It will be essential to full cloud networks and to network slicing.
It seems that we have been waiting for many, many years for the vRAN. It is eight years since China Mobile published its seminal white paper setting out its concept of a green, low-opex Cloud-RAN and the industry leapt on this as the architecture that would change the economics of mobile broadband. Yet in 2019, there are very few vRANs in evidence.
The original idea, that vRAN would become mainstream with LTE, has died – there were a few deployments in Korea and Japan, but they were proprietary and expensively engineered. In fact, most first-phase 5G RANs are being deployed in the conventional way.
Standard servers need a great deal of acceleration to be able to cope with the demanding digital processes of a RAN; some ‘vRANs’, as a result, have been quite proprietary, with non-standard VNFs running on separate, but specialized and expensive, boxes. Even companies trying to do a vRAN on common cloud infrastructure have had to invest large amounts of time and money in tuning and retuning the hardware and software to run together so that they could deliver anything close to the performance of a dedicated RAN – the much-vaunted Rakuten deployment being an example.
Other challenges complicate and delay the roll-out of a macro vRAN, such as how the digital functions should be split between the cloud and the cell site (there has been a backlash against full centralization because low latency functions need to be closer to the user). In future architectures, some of those cell site functions could run on edge cloud nodes rather than on proprietary appliances, but those platforms are not mature yet.
That sums up the problem with vRAN – just as one set of enablers stabilizes, the industry moves on with its demands. So now, in the RAN and the core, operators are reluctant to deploy virtual machines on OpenStack cloud infrastructure, because this approach is now seen as cumbersome and old-fashioned. Better to wait for full cloud-native, container-based designs, which will be far more flexible and automated, and should do away with much of the upfront tuning – but which are not fully commercially available yet.
In 2019, Mark Atkinson, head of Nokia’s 5G and small cells business unit, suggested that the vRAN would not be a macro layer deployment in the short term, but an urban small cell roll-out geared to enterprise. Atkinson said Nokia’s cloud base station had been used “in one of the busiest cities in the USA” to “split traffic to ensure each connection gets the service it needs”.
He added that, in the Nokia AirScale architecture, distributed units processed time-critical functions at the cell site, connected by Ethernet fronthaul to the radios, while non-real time functions were fully virtualized and run in a central cloud data center, presumably with fiber fronthaul.
“This flexible mix of local and cloud-based processing a real game-changer,” wrote Atkinson. “This means that we can combine performance, scalability and efficiency at its best – in the radio unit (RU), distributed unit (DU) and centralized unit (CU).”
Nokia has been a strong flagwaver for vRAN, although it has sometimes, like first mover operators, been wrongfooted by the changes in architectural thinking in this area. It has worked closely with Verizon on its vRAN roadmap. In May 2019 the operator demonstrated a fully virtualized RAN baseband, enabling edge computing, in its network in California, working with Nokia and Intel. An earlier trial, held in January in Houston, Texas, used multi-access edge computing (MEC) equipment and software in a 5G testbed and claimed to have cut latency in half.
Verizon has conducted vRAN trials with all three of its 5G access network suppliers (Nokia, Ericsson and Samsung), and has made some implementations, particularly in localized small cell networks and in pre-commercial 5G. The first trials, in the first quarter of 2018, took place in Oklahoma City, with the virtualized baseband functions running on Nokia’s AirScale cloud base station server, based on Intel’s Xeon processors and its FlexRAN reference architecture, which supports the RAN stack running on servers. The architecture builds on existing centralized, but non-virtualized, units which Verizon calls cRAN hubs.
The operator now has a roadmap to convert those hubs to full vRAN units and pledges implementation of Open RAN in 2023, though this is unlikely to be based on vanilla O-RAN specifications, but rather on a platform heavily customized for Verizon by its close suppliers – but with open interfaces so further vendors can be added or swapped at a later stage.
Verizon has said: “We have thousands of these cRAN hubs throughout the US. They’ve already been identified. They are built out and equipped. And we have been in the process of centralizing those baseband units.” As a result, 15 to 50 centrally managed cells could work together, using LTE-Advanced features like remote electrical tilt (RET), which can point an antenna array toward high traffic areas during the day and then tilt them up at night to improve coverage.
The next step is to “do a lot more … The cRAN hubs become vRAN hubs, and now you can communicate across cRAN hubs. … You can scale horizontally instead of vertically.”
Verizon has also started to virtualize some functions in the upper layer, or Layer 3, of the network, within its own cloud architecture. This is not all about 5G, though it will go hand-in-hand with 5G NR roll-out as that progresses. Some of the benefits of vRAN come from having a common set of functions in the centralized, white box baseband unit (BBU), which can work with remote radio units (RRUs) supporting different standards, and potentially from different vendors. This gives the MNO flexibility to swap new radios in and out only as required, rather than engaging in a complete upgrade.
There is still a long road ahead to make vRANs mainstream, and easily deployable by operators without the technology prowess and budgets of Verizon. Many MNOs remain sceptical of the architecture, and even some vendors – notably Huawei, despite its early involvement in the China Mobile tests – have questioned whether it can deliver the promised automation and reduction in TCO and power consumption.
Yet the RAN is by far the most expensive part of the mobile network to deploy and run, and without a radical change in architecture, it will he hard to roll out 5G in a way that can be cost-justified. The move to a cloud-native network, at least in the second phase of 5G, underpins a very large part of the 5G proposition, in terms of reducing TCO by leveraging cloud economics, and enabling full service flexibility and developments like dynamic slicing. If the case for vRAN cannot be made convincingly, in the macro as well as the small cell layer, the whole 5G story will need to be recast.