One of the great dilemmas facing vendors in a standardized industry is how to maintain differentiation and margins while also supporting standards-based interoperability. As the mobile industry has matured, operators have placed multivendor interoperability higher and higher up their priority lists. So will they finally achieve full openness in the 5G era, free of the compromises with have dogged previous generations?
In a recent survey of 68 Tier 1 MNOs by Rethink Technology Research, interoperability emerged as the third most important enabler of a successful 5G deployment model – 48% cited this as one of the three factors they considered essential to such a model, and it was outranked only by ‘lower total cost of ownership than 4G’ (72%) and ‘end-to-end virtualization’. Worryingly perhaps, interoperability came out higher than security (38%).
Yet the calls for interoperability have been heard before, as a new technology emerged, but the cellular world has never achieved the kind of mix-and-match platforms which are seen in WiFi. Of course, devices and equipment interoperate interchangeably; and there has been significant progress in RAN-neutral options for the packet core; but when it comes to network equipment, most operators’ ‘multivendor’ deployments go no further than hiring different suppliers to roll out different regions or zones.
Standard interfaces exist, which should allow elements from different vendors to be mixed within a single RAN. But too many come with performance compromises, or with vendor-specific implementations which effectively make them incompatible. A good example is CPRI, which has been implemented in a semi-proprietary way, ending the dream of easy interworking between different suppliers’ basebands and radio heads.
Now there are new network architectures which will need interoperability even more urgently if they are to move into the commercial mainstream. These are densification, virtualization, and the 5G New Radio. Together, these make up a radical approach to the RAN and they will often be inter-dependent – virtualization makes it far more cost-effective to deploy and manage huge numbers of small cells in a dense network; 5G is expected to rely heavily on small cells in higher frequencies, to augment wide area 4G, and on virtualized RAN to boost resource flexibility and efficiency.
In all three cases, there are significant efforts to define standard interfaces and ensure open interoperability, and as operators move to 5G this will be critical – many will want to migrate gradually, keeping LTE running in parallel for many years, and so will need to be able to swap elements in and out flexibly, without being locked into a single vendor.
But the success of these efforts is not assured yet, and the failure of CPRI to be fully unified casts a shadow over the industry. The Small Cell Forum made a strong impact with its Iuh interface between residential access points and gateways, but its FAPI and nFAPI specifications, for physical and virtual small cell HetNets, have not yet become well established in the market, where many densification projects are still centering on single-vendor roll-outs.
In virtualization, there have been a couple of demonstrations of multivendor NFV (Network Functions Virtualization) systems interoperating, but gradual progress is frustrating early movers like Vodafone, and the vendor politics which afflicts this process is spreading to 5G. Although 5G is expected to be deployed in a Cloud-RAN formation in many cases, the 3GPP RAN working group has yet to agree on the functional splits between the different parts of the network.
In some areas, service providers are turning to open source initiatives to accelerate progress towards open multivendor platforms. The recent move by AT&T and its allies to accelerate the schedule for initial 5G NR standards showed how quickly some MNOs want to move towards new architectures, to improve their own business case and to fend off new challengers from the webscale world. These fast movers may be a minority, but they are very influential, and if the conventional standards bodies do not respond to their demands for faster progress, they will look to the open source approach.
This has been seen in the embrace of Openstack for orchestration of virtualized networks and in Small Cell Forum’s open sourcing of nFAPI, among other examples. The influence of open source processes on 5G standards remains unclear, but 3GPP’s decision to accelerate the standardization of 5G NR Non-Standalone must reflect, in part, a desire to stay responsive in a world where, beyond the radio itself, many elements will be defined by initiatives hosted by the Linux Foundation, or by web-focused projects like Facebook TIP (Telecoms Infra Project).
It is clear that, for most operators to adopt the virtuous triangle of small cells, 5G radio and Cloud-RAN, there will need to be significantly more confidence that standard and open interfaces will be agreed in the coming two years.
Will 5G virtualized RANs avoid repeating the CPRI mistakes?
There are two issues with defining interfaces for a virtualized RAN. One is to secure full vendor support for a completely unified implementation, which failed in CPRI. The other is that one interface is no longer enough – there are likely to be different ways to split virtual and physical elements, according to use case and deployment scenario.
Of course, where the RAN differs from other parts of the network, such as the packet core, is that some physical elements will always remain, at least for the RF functions. In 4G, the only standardized split between different network elements was between the remote radio head and baseband unit, supported by the CPRI and ORI interfaces, but these were not defined with virtualization in mind. To support fronthaul – the link between physical network functions (PNFs) and their centralized virtual equivalents (VNFs) – these interfaces will need to be greatly enhanced to improve latency and bandwidth.
Beyond that, most industry players agree that, in 5G C-RAN, there will need to be more than one functional split between physical and virtual. For instance, NTT Docomo has outlined a topology in which some parts of the baseband are collocated with the PNFs at the radio site, and this idea will become increasingly important as operators look to distribute cloud functions to the edge of the network. One way to do this is to deploy a technology like ETSI Multi-access Edge Computing (MEC) on processors integrated into small base stations (though others favor implementing localized cloud platforms on their own hardware or on gateways).
This example indicates the need for flexibility as to where the PNFs and VNFs reside – the initial concept of Cloud-RAN, in which all baseband functions were virtualized in a central location and cell sites were stripped down to simplified radio/antenna units, will not work for all applications, or for operators without access to ubiquitous and affordable fiber for fronthaul. Low latency services will often work better with much of the cloud resources and content distributed close to the user, and with zones of capacity provided by localized small cell networks, connected to virtualized controllers via Ethernet IP transport.
The split of the radio from the virtualized resources also makes it easier to mix and match different access technologies within a single network, which is particularly important in the Internet of Things (IoT), helping to neutralize the fragmentation between different technologies.
As Cisco wrote in a recent white paper on 5G: “The evolution toward new RAN decompositions with open, standardized fronthaul interfaces is also beneficial for supporting IoT use cases. Whether it is LTE-M; narrowband IoT (NB-IoT); one of the leading low power, wide area (LWPA) technologies (such as Sigfox, LoRa, Weightless or OnRamp); or some new 5G RAT, efforts to centralize and virtualize as much of the RAN as economically feasible help drive complexity and cost out of the access points for IoT.”
However, this epitomizes the double-edged sword for vendors – interoperability can help drive a market to large scale quickly, but it means they can no longer lock customers into their products, or their preferred technologies.
All these issues are occupying the thoughts of 3GPP’s RAN working groups, which met in Spokane, Washington in January and then in Athens the following month (as well as at the RAN plenary meeting in Dubrovnik two weeks ago). The complexity of the issue is illustrated in the figure, from the 3GPP RAN3 Technical Report TR38.803v1.1.0, which summarizes all the options for the functional split.
The group has three main issues to address:
the split between the central unit and distributed unit
the fronthaul split for the remote radio head
the RAN internal split between user plane and control plane
And for all the options it needs to decide whether there should be a standard 3GPP interface or not, and whether that should define just the functional architecture, or all messages, bits and bytes.
Eiko Seidel of Nomor Research, who attended the RAN 3 meetings, made some interesting comments in his write-up of the event. He cautioned: “Bear in mind that in every generation we standardized interfaces in RAN3, which in practice still did not allow for multivendor interoperability.” And his notes hint at some of the commercial interests which may still stand in the way of full interoperability in 5G.
He wrote in a paper entitled ‘3GPP 5G Adhoc: Any Decisions on RAN Internal Functional Split?’ that “The general attitude is that most infrastructure vendors are opposing to standardize any RAN internal interfaces in detail. The operators on the other hand insist to standardize at least one fronthaul and one midhaul interface in detail. This shall enable a multivendor infrastructure market.”
He believes most vendors are willing to compromise and support a functional description but that would not satisfy the operators. Yet there are risks if interfaces are standardized outside 3GPP, as CPRI was, since that opened the way to more vendor-specific adaptations.
And even among the operators, there seems to be little consensus over the best options for each split. For the central unit/distributed unit interface, there is a divide between those favouring Option 2 or Option 3 on the figure; for fronthaul, Options 6, 7 and 8 are all under consideration.
“Considering that there is not a clear view from the operators on the split, and the lack of expertise in RAN3 on a physical layer internal interface, I doubt that we will see quick progress on a fronthaul interface specification,” concluded Seidel.
Option 2 is attractive because of its commonality with the functional split in LTE Dual Connectivity, which would help operators deploy 5G cells attached to an LTE-DC master base station, to speed up deployment. Option 3 is favored because it allows two levels of error correction – HARQ (hybrid automatic repeat request) in the distributed unit to handle fast re-transmissions and link adaptation; with ARQ in the central unit to handle mobility or wireless relay errors.
It remains to be seen whether much progress was made on these issues at the recent RAN plenary meeting, but if there is one area of agreement, it is that there should not be too many options retained in the standard, since this runs the risk of over-complexity and lack of interoperability. Even with the greater flexibility afforded by NFV and software-defined networking (SDN), that flexibility can have its downside in terms of harmonization and openness. Seidel believes this is “one of the biggest risks for the success of 5G. We got such a huge community of people and companies involved in 5G research, standardization and commercialization. Everyone wants to get its share and compromises become more challenging using the consensus approach of 3GPP.”
Small Cell Forum’s nFAPI moves into open source
Small Cell Forum published its nFAPI interface last July, in the hopes that it would become the basis of industry standards for connecting physical small cells with virtual network functions in controllers or other central units. While, in the past, some of the Forum’s specifications have been taken up by 3GPP or other standards bodies, this time its technology is making the leap into open source, with the Open nFAPI project, sponsored by Cisco and distributed under the Apache Software Licence v2.
nFAPI, an extension of the FAPI (Functional API) for physical small cells, is the cornerstone of an interoperable framework designed to drive cost and complexity down dramatically in small cell deployments, and support operators’ current moves towards densification.
In November, the Forum warned that vRAN projects “have mainly been vendor proprietary, risking fragmentation and delaying progress”, whereas nFAPI provides a standard split between physical and virtual elements. It supports connections between the two over common packet Ethernet fronthaul links.
Not only will this support multivendor networks for conventional mobile operators, but it could also hasten the roll-out of multi-operator and neutral host small cell systems. In particular, there is the potential for a common set of distributed unit PNFs (physical network functions) to be shared between multiple operators, which are then responsible for their own central unit VNFs. When combined with shared spectrum LTE options such as MulteFire in the US’s 3.5 GHz CBRS band, there is real potential to open up the small cell platform to non-traditional mobile providers such as cablecos or vertical market specialists.
Mark Grayson, distinguished consulting engineer in Cisco’s Group CTO Office, announced Cisco’s championship of the open source nFAPI project, writing in a blog post: “Looking to leverage the experience of other wireless ecosystems that use a combination of published specifications, open source libraries and code to execute interoperability testing, Cisco is pleased to announce the establishment of the ‘open-nFAPI’ open source project, a set of libraries, simulators and associated Wireshark dissectors that is aimed at accelerating adoption of the Small Cell Forum’s multivendor nFAPI split architecture.”
Cisco and the Forum clearly have their eyes on support from what Grayson calls “other alternative open source ecosystems interested in RAN virtualization” such as Facebook’s Telecom Infrastructure Project (TIP), the Central Office Re-architected as a Data Center (CORD) initiative and the Open Air Interface (OAI) project.
Initiatives like Open-nFAPI provide such groups with a way to integrate with closed source and proprietary RAN equipment, which could significantly shake up the 5G landscape when it comes to standards, intellectual property and the main power brokers. Open, commoditized hardware; shared spectrum; and standardized VNFs and fronthaul could combine to make 5G a platform for a far wider variety of service providers, disrupting not just the traditional RAN vendors (clearly in Cisco’s sights) but the operators too.
This will mean multi-operator, as well as multivendor, interoperability. The ability to build specialized and optimized small cell subnets to serve specific business cases – especially in the IoT – should revolutionize the ability to support vertical industries with their own traffic patterns and performance criteria. These subnets will often be deployed by non-MNOs, accessing a ‘slice’ of capacity, and core network services, in a central cloud-based platform. That might be run by an MNO, but could just as easily be run by Amazon, Google or a vendor (think of Nokia’s recent upgrade of its Cloud Packet Core, with a focus on private networks). In turn, these trends will converge eventually to support the full concept of 5G network slicing.
“We think that the small cell industry has a great opportunity to drive the definition of multi-operator LTE and 5G systems using an open source approach,” wrote Grayson. “The establishment of the open-nFAPI open source project is the first step on this path.”