Qualcomm has begun sampling its promised 5G RAN products, in what could be a significant boost to the scale of the Open RAN industry from mid-2023, when brownfield operators are expected to start deploying the technology in earnest, if mainly in small cell or rural environments rather than urban macrocells.
The products – the X100 5G RAN Accelerator Card and QRU100 5G RAN Platform – were first announced two years ago. They see Qualcomm returning to the macro network infrastructure market for the first time since 1999, when it sold its infra-side assets to Ericsson to concentrate on devices. It has a small cell system-on-chip family, FSM, but it has been clear that its roadmap for 5G RAN extends to macrocells including Massive MIMO, in time.
Qualcomm says its new offerings will be fully O-RAN compliant and will support cloud-native architectures with high power efficiency. They will support radio units (RUs), distributed units (DUs) and distributed radio units (DRUs – which integrate the RU and DU in a single box). The SoCs will incorporate Qualcomm’s 5G Modem-RF System with baseband, transceiver, RF front end and antenna panels, and there will be versions for sub-6 GHz and millimeter wave frequencies.
They appear to be roughly on schedule – at launch, sampling was promised for mid-2022, and Gerardo Giaretta, senior director of the 5G infrastructure business, told an analyst briefing that it had been a significant achievement to meet that timeline, given that this is an entirely new chip, developed from scratch for macrocell infrastructure. It is not related in design to the FSM range, which powers some small cells in commercial Open RANs, notably in some of Rakuten Mobile’s deployed units in Japan.
Giaretta was keen to emphasize that the QRU100 5G RAN Platform is a different beast from the small cell offering, and provided a laundry list of capabilities that it will support, including multi-operator RAN sharing (MORAN), dynamic spectrum sharing (DSS) and digital beamforming, as well as 64T64R and 32T32R Massive MIMO.
He has particularly focused on Massive MIMO as this is challenging to deploy and run and requires very advanced chipsets, and so is seen as the primary barrier to achieving equivalent performance to that of integrated RANs, affordably. Giaretta has said that Open RAN Massive MIMO is “really a silicon availability challenge” – if that were addressed, efforts like those of Mavenir or Symphony, which aim to create an ecosystem of homegrown and third party radios for Open RAN, would be easier.
“The problem right now is all of the Open RAN trials and early deployments that there’s been have been mostly based on generic SPUs,” Giaretta said. “And generic SPU is fine up to a point, but when you have to deploy this very, very high capacity type of scenario like Massive MIMO, the Layer 1 processing, the physical processing, in 5G is just too demanding.”
“Network operators understand the importance of deploying 5G network infrastructure that can support the demands of next-generation networks, enhance consumer experiences, and unlock new use cases. They require a scalable solution with high capacity and low latency,” said Durga Malladi, general manager of cellular modems and infrastructure at Qualcomm, in a statement.
Qualcomm expects commercial deployments to start in the second half of 2023, and needs to stick to this timeline to avoid losing the initiative to its competitors in the base station field. Competition is not just about Intel. Marvell is developing an end-to-end strategy combining processors and accelerators. Then there are well-established RAN players that address some or all of the Open RAN platform, such as Analog Devices, NXP and Xilinx (now part of AMD). And of course, the biggest challenge of all is to outdo the inhouse developments of the RAN vendors, notably Ericsson Silicon.
In the meantime, Qualcomm has been building an ecosystem around its platform. It has already started co-developing an Open RAN DU with NEC, a key 5G macrocell supplier to Rakuten Mobile and the broader Symphony platform. This vDU project, however, is under the auspices of the 5G Open RAN Ecosystem initiative, which was formed earlier this year under NTT Docomo’s leadership in Japan.
And at the Barcelona edition of MWC, in February, Qualcomm announced that HPE would make an Open RAN DU that would use the X100 for RAN acceleration. The X100 PCIe card offloads particularly low latency or computationally intensive 5G baseband tasks from the CPU to improve performance and efficiency. These tasks include Massive MIMO beamforming, DSS support, demodulation and channel coding, said HPE, whose system will run on its Proliant telco server and support up to four accelerator cards in a small-footprint 1-unit server.
The HPE example shows how Qualcomm aims to provide acceleration to surround vDU servers based on Intel processors – Marvell is working with Dell on a similar basis. Qualcomm therefore claims not to be competing with Intel, which has achieved a near-monopoly of trials and early deployments of Open RAN thanks to its FlexRAN reference architecture. It will focus on the PHY layer, Layer 1 accelerators and RF, the areas where it has differentiation. It did acknowledge “similarities between the different architecture approaches” between the FlexRAN PHY reference design and the Qualcomm proposal, but insisted “we are not targeting Layer 2. This is not a CPU play for us”.
But it is clear that Qualcomm certainly aims to limit the proportion of an Open RAN that Intel can target. It may have mothballed its own server processor development, but it will defend the network-specific chip territory aggressively. So where a network function, such as an Open RAN centralized unit (CU), can run on COTS processors, Intel will prevail (or a processor challenger such as Marvell). But where a function needs specialized hardware with significant acceleration – Layer 1 functions running on a virtualized distributed unit (DU) for instance – Qualcomm aims to be in pole position.
It will extend its reach from its heartland in the baseband and in the RF front end, into the open DU, leveraging the strengths it has honed in the device market, of power efficiency, very tight integration of multiple components, and high performance signal processing. None of these are Intel’s traditional strengths, and so far, even its DU development with Rakuten Symphony (see separate item) is not breeding confidence that it is challenging Qualcomm or other specialists in this area.
It has developed a 5G portfolio including accelerators, using FPGA and eASIC technology acquisitions, but its main advantage remains FlexRAN. But as other reference designs are developed, based on non-Intel chips but offering the same advantages of relatively simple, templated design, that differentiation will fade.
Already, Vodafone has selected Qualcomm to be its partner in creating an Open RAN reference design, which will be available next year, and will aim to make it easier and cheaper for a broad base of vendors to develop Open RAN radio units and distributed units that could target many cell sizes, all the way up to Massive MIMO.
The Vodafone partnership, announced in May 2021, is Qualcomm’s most significant Open RAN alliance. The pair will work to develop blueprints that aim to lower the barriers to smaller vendors entering the closed world of RAN equipment. Although Vodafone has also announced a broader initiative to drive an Open RAN chip ecosystem, working with about 20 silicon partners, Qualcomm remains its key partner. The initial reference designs will be based on Qualcomm’s RU and DU platforms including Massive MIMO and should follow hard on the heels of the new platform being commercially available.
The aim of the blueprints is to enable new entrants to the RAN market to focus on the performance and optimization of their VNFs, applications and protocol stacks, while not worrying about chips and hardware. In the meantime, they say that a common 5G RU and DU platform will drive critical mass in components so that equipment makers can reach scale more quickly.
Several operators, many of them leading lights in open RAN, have also welcomed Qualcomm’s plans, including AT&T, BT, Deutsche Telekom, LG Uplus, NTT Docomo, Rakuten, Reliance Jio, Telecom Italia, Telefónica, Verizon and Vodafone. Reliance Jio is developing its own 5G platform based on Qualcomm chips.
And among the O-RAN vendors, Mavenir has already announced that it is working with Qualcomm to integrate the 5G RAN Platform with its vDU software. It is also using Qualcomm chips to expand the capability of its Open Beam RU reference designs. Meanwhile, Fujitsu is developing a DRU using Qualcomm’s acceleration.