The race to Open RAN is a marathon, not a sprint
To meet 5G demands, RAN architecture will undergo its most dramatic changes yet – virtualization and cloudification.
Traditional RAN components in cellular networks are designed for optimal performance using proprietary technology.
In a typical distributed 3G network, a base station consists of a remote Radio Unit (RU) at the top of a tower connected to a Baseband Unit (BBU) located at the bottom of the tower with a fiber optic cable using the Common Public Radio Interface (CPRI) standard. To enable virtualization of the RAN, a different architecture is used in 4G/5G networks in which the BBUs are split into two parts: a Central Unit (CU) located on a centralized server and a Distributed Unit (DU) located either directly at the tower or at another location several kilometers away.
Mobile Network Operators (MNOs) have numerous options regarding where they locate DUs and CUs. In practice, the exact placement will depend on multiple variables such as business requirements, transport capabilities (for example, is fiber available?) and RAN coordination features.
Locating the DUs closer to the edge of the network reduces latency; conversely, centralizing the DUs can optimize the network by minimizing interference within a group of sites, but at the cost of higher latency.
The RAN market today is dominated by a small number of incumbent vendors, of which Huawei is one. In a bid to generate more competition and increased vendor diversity, some MNOs support the concept of open RAN in which proprietary RAN technologies are replaced by open standard alternatives. This involves two key initiatives:
A number of companies are developing an ecosystem of software and hardware based on the open RAN approach and collaborating in a number of standards organizations.
Two open RAN organizations actively promote Open RAN: O-RAN Alliance and the OpenRAN project, which is part of the Telecoms Infra Project. The members of each include MNOs and software and hardware vendors.
In February 2019, ORAN published the first open standard specification for the fronthaul interface between the RUs and BBUs. Other open interfaces standardised include the X2 interface, which interconnects BBUs.
OpenRAN is also defining an open white box device using standard server and accelerator components that meet the needs of operators for the speeds and densities required for 5G NR. The TIP projects use O-RAN interfaces and specifications.
Figure 1 shows the open RAN reference architecture being developed by the O-RAN alliance.
Figure 1 O-RAN Alliance Reference Architecture
A few years ago, MNOs were excited at the prospect of drastically reducing the cost of mobile infrastructure by replacing proprietary hardware with open RAN interfaces that would enable the mixing and matching of RUs and BBUs from different vendors. However, open RAN proved to be much more difficult to implement than initially envisaged, and several planned deployments were postponed. Today, the tide is beginning to turn.
At present, however, open RAN deployments consist of a handful of mostly greenfield commercial deployments and numerous legacy network trial deployments. With the majority of MNOs having already selected their initial 5G infrastructure partners, it’s unlikely that any of the new open RAN vendors will capture a significant share of any major MNO’s initial 5G build.
In the short term, Counterpoint Research believes that the most likely opportunities for open RAN are in new network builds or in emerging markets where 4G and 5G is still in the planning stages. Several MNOs are also looking to deploy open RAN in rural markets. Other opportunities include small cell deployments, particularly in-building deployments and 4G/5G private networks. However, other opportunities may open as 5G matures and MNOs start thinking about upgrading and enhancing their networks.
Companies that want to build their own enterprise networks are also potential customers. Many large-scale industrial conglomerates are interested in building private cellular networks, particularly with the increasing availability of unlicensed spectrum bands (such as the CBRS band in the US) and the imminent availability of 5G NR-U technologies specifically designed for unlicensed bands. Counterpoint Research believes that the private network market is probably one of the best short-term market opportunities for open RAN vendors once 5G NR becomes available.
Although a small number of major MNOs are believed to be starting limited commercial deployment, Counterpoint Research believes that numerous challenges and issues still need to be resolved before the widespread adoption of open RAN commences:
Open interfaces: These are essential to enable multi-vendor networks and a more competitive and vibrant supplier ecosystem and can be realized on both traditional proprietary and new open RAN platforms. The main fronthaul interface standards are the 3GPP-developed eCPRI and O-RAN 7.2x open standard from the O-RAN Alliance, both of which are in the initial phases of deployment. However, open versions of other RAN interface standards are required. These are still under development by the various open RAN industry groups.
Performance: The performance and cost differential between custom-built and generic open COTS hardware will be a critical deciding factor for MNOs. At present, proprietary ASIC processors are more power-efficient and less expensive than x86-based CPUs. The performance gap is significant for demanding workloads, for example, networks that operate 5G MIMO mmWave antennas. In addition, there are form factor requirements, as the processors may need to be fitted into standard industry enclosures designed to be attached to roof top fixtures or onto street light poles.
With the possible exception of rural networks that focus on coverage rather than serving densely populated areas, open RAN deployments are not being run on standard x86 COTS processors. For example, in the case of Rakuten, its partner Intel had to implement considerable hardware acceleration via FPGAs to run the baseband software stack supplied by another vendor.
The conclusion is that running heavily virtualized baseband stacks based on standard x86 CPUs isn’t feasible without additional bespoke accelerator platforms based on FPGAs, GPUs or possibly new chip architectures such as those being developed for AI applications, all of which have cost and power implications.
Scalability and coverage densification: Scaling to meet capacity and coverage demands, particularly in large macro networks or large indoor venues, such as stadiums, is unproven today. Coverage densification in cities, particularly at millimetre wave frequencies, is also a major challenge for open RAN. However, several vendors have recently initiated trials with major MNOs to test and improve performance in such environments.
Increased Operational Complexity: Perhaps the biggest concern of MNOs is the increased complexity of an open RAN network. Virtualization has a steep learning curve and MNOs have struggled with the sheer number of Virtual Network Functions (VNFs) that must be integrated into existing network architectures and linked to their operations systems.
In a conventional network, MNOs typically have access to a single vendor that’s available to resolve issues and problems. However, with an open RAN network, where software and hardware are sourced from many vendors, it might not be immediately clear which product is malfunctioning and hence which supplier should be contacted.
Total Cost of Operation (TCO): Although some MNOs claim significant cost savings with greenfield deployments of open RAN, MNOs in general remained unconvinced about the TCO benefits of open RAN for legacy networks. As a result, MNOs are looking for additional benefits such as the ability to deploy new applications that can generate new revenue streams, for example, edge compute and IoT. However, the business cases for many of these new services remain unproven.
No white box RU market: At present, Counterpoint believes that RU units supporting the O-RAN alliance 7.2 specifications are only available from two vendors, although several other vendors are believed to be engaged in developing such products.
In late February 2020, TIP announced the launch of the Evenstar RU initiative, which will develop reference designs aligned with 3GPP and O-RAN Alliance specifications. First RU products are believed to be Band 3 (1,800 MHz) units, which are expected to become commercially available towards the end of 2020 with an ultimate price target of US$1,000. Evenstar RUs will support the O-RAN Alliance 7.2 split architecture.
Interoperability: The ability to mix software and hardware from different vendors is often touted as one of the major benefits of virtualization. However, it’s also one of the greatest hurdles. At present, there are no well-defined standards or testing methodologies to guarantee interoperability and performance of software and hardware products from different vendors.
At this stage, vendors are just testing their own products to see if they comply with O-RAN standards rather than testing interoperability with each other’s products. However, the recent opening of two test and integration centers in Berlin and Beijing is a positive step that should ultimately resolve this issue.
Although there’s considerable momentum behind open RAN at present, the transition to fully interoperable, multi-vendor, cloud native RANs is a marathon, not a sprint. Counterpoint believes that there is still some significant research, lab testing, and trials, plus a few major milestones to be overcome before wide scale open RAN adoption happens.
While MNOs are clearly looking for vendor diversity, open RAN technology and standards need to mature before the technology can be deployed in commercial networks at scale as MNOs cannot risk exposing their customers to unreliable infrastructure.
Network performance is still a major challenge, as general-purpose x86-based hardware struggles to achieve performance and OPEX costs on a par with dedicated ASIC-based hardware. Open fronthaul interfaces will be critical to allow interoperability between RUs and BBUs and, although the first open RAN interface standard was released by the O-RAN Alliance in February 2019, an ecosystem of truly interoperable open RAN white box hardware will be required before open RAN can be deployed en-masse. Other important issues to be resolved include improvements in scaling and coverage densification, which will be required to meet capacity and coverage demands, particularly in large macro networks and in large indoor venues such as stadiums.
Nevertheless, the fact that major MNOs, such as Telefonica and Vodafone, are prepared to give smaller open RAN vendors even a small share of their business should be a signal to incumbent infrastructure vendors that MNOs are serious about open RAN. However, it won’t be easy to dislodge the incumbents that Counterpoint believes are well positioned to control the adoption of open RAN.
Major MNOs have already started selecting their initial 5G radio partners. There are millions of 5G NR-ready radios in the field that can be upgraded with minimal hardware, or in some cases remotely. This complicates the business case for new open RAN entrants. As a result, it may take three to five years for these players to gain a sizeable market share, and in that time, not all will survive. Furthermore, increasing network complexity, coupled with the application of new technologies such as AI-based automation, may favour incumbents as they have more technical resources to deal with the ever-increasing complexity of mobile cellular infrastructure.
As a result, Counterpoint believes that mainstream adoption by major MNOs will only happen when it’s clear that the operational benefits and flexibility offered by open RAN network solutions outweigh and compensate for the hardware, power, and system integration costs, while attaining the same high level of network performance and reliability. Although some limited commercial rollouts can be expected during 2020, Counterpoint Research therefore believes that large scale open RAN deployments are probably at least 18 to 36 months away.
More details, please refer to the website of Counterpoint.