Open RAN and the mission to crack massive MIMO
Massive MIMO, as the name implies, is the big league of radio, a place of macho posturing about performance, where players flaunt their specs and records are set. While radios in widespread use might each include just a couple of transmitters and receivers, a high-end massive MIMO unit can pack 64 of them (so-called 64T64R). The elite can now churn these out in boxes that weigh about the same as a six-year-old child.
It is regarded by experts as one of the toughest challenges in 5G radio. "You are driving huge digital signal processing as well as radio processing on the mast, and that is a whole new game," says Alan Hynes, the director of engineering for Benetel, an Irish maker of radio equipment. As in other big-league sports, opportunity has tended to favor those with multi-billion-dollar budgets.
What chance, then, for smaller companies touting "open RAN," including Benetel? The disruptive technology promises new interfaces allowing components from one supplier to speak with products from another. Some operators hope it can vanquish the monolithic system, where everything is provided by a giant vendor like Ericsson or Huawei. Massive MIMO, they worry, is an Achilles heel – "a difficult thing for open RAN to crack," in the words of Yago Tenorio, Vodafone's head of network strategy.
Unfortunately, massive MIMO cannot simply be overlooked like a soccer team's predilection for scoring lots of goals rather than defending against a few. It will be needed to cope with soaring levels of data consumption on 5G networks. Even now, massive MIMO is understood to feature at 90% of Vodafone's 5G sites in the UK – a deployment, it should be noted, that has focused on urban areas initially.
"The number of locations that can benefit from massive MIMO is going up," says Gabriel Brown, a principal analyst with Heavy Reading, a sister company to Light Reading. "The addressable market is increasing because the price-performance ratio on massive MIMO base stations is improving."
On the plus side, this creates a bigger opportunity for open RAN players. Yet the challenge they face is daunting. Ericsson has recently been shouting about a 64T64R unit that weighs only 20 kilograms, about 45% less than its older model. That difference promises a huge reduction in deployment and energy costs for service providers. In the same week that Ericsson unveiled its 20-kilogram unit, Huawei boasted one that weighs just 19 kilograms, knocking 6 kilograms off its predecessor.
The key advantage these Tier 1 vendors hold is their custom silicon, and the volume of business to justify their investments in it, says Hynes. He is not the only one who recognizes that advantage. Ericsson itself hailed advances within Ericsson Silicon, its in-house chips division, when explaining its latest improvements. Commenting anonymously, one analyst said access to cutting-edge semiconductors has never been so important in radio.
ASICs versus FPGAs
Both Ericsson and Huawei have used what are known as application-specific integrated circuits (ASICs) in their radio designs. In high enough volume, it is these customized chips, designed for an individual purpose, that make the Tier 1 equipment vendors so proficient in massive MIMO. Yet ASICs are not the only option for companies building 5G radios.
The alternative is another type of semiconductor called a field programmable gate array (FPGA). In contrast to ASICs, these FPGAs can be tinkered with after their design to suit particular needs. That flexibility and the constraints faced by smaller vendors would seem to make FPGAs ideal for open RAN. "The fundamental decision maker on whether to go FPGA or ASIC is volume," says Hynes. Xilinx, one of the world's biggest manufacturers of FPGAs, clearly sees open RAN as a growth opportunity.
The trouble with FPGAs, when it comes to massive MIMO, is the energy expense. Volume considerations aside, FPGAs are generally seen as power hogs. Steve Papa, the CEO of an open RAN software developer called Parallel Wireless, describes them as the "friend of flexibility" but "enemy of energy efficiency." Hynes is just as critical.
"The things that operators are concerned about on masts are dimensions and weight and getting those down when there is so much digital signal processing going on is the number one challenge in massive MIMO," he says. "You need to be as power-efficient as possible. FPGAs for massive MIMO are not going to work."
Unsurprisingly, Xilinx begs to differ. "Time has changed since FPGAs were extremely power hungry," said Brendan Farley, Xilinx's vice president of wireless engineering, during a recent roundtable discussion with analysts and reporters. Xilinx claims to have made its FPGAs more power-efficient by "hardening" some features, including what is known as the RFSoC DFE (radiofrequency system-on-a-chip digital front end).
The risk of doing this is a loss of programmability, the very advantage FPGAs tout over ASICs. But Farley reckons Xilinx has been able to maintain programmability through investment in other technologies, including a processor branded AI Engines. What's more, in a recent paper, Xilinx said AI Engines could deliver up to eight times the computing power with half the energy consumption of traditional programmable components.
Farley says there is no reason why an open RAN-based massive MIMO product cannot match traditional gear on size and weight. He singles out KMW, a South Korean radio manufacturer and Xilinx partner that recently advertised a 64T64R unit weighing 25 kilograms, considerably less than Ericsson's lightest unit in 2020. Its readiness for open RAN seems in doubt, however. In its literature, KMW says it will be able to support open RAN hardware and software "on time," and that it is "developing platforms and product line-ups to support various needs of global customers."
The industry may need plenty of convincing about FPGAs. Negative publicity has dogged them as a result of a poor historical decision by Nokia, another Tier 1 vendor, to use FPGAs in its 5G radio units. A margin squeeze and loss of competitiveness ensued, and Nokia is now shifting to ASICs, a process it does not expect to complete until the end of next year.
And now for something completely different
For Benetel, the answer is not to use FPGAs at all with massive MIMO products. Instead, it has teamed up with two big semiconductor firms in the US to produce an integrated circuit with a difference. Where traditionally these chips would be developed by a Tier 1 vendor and incorporated into its overall radio access network offering, Benetel is pitching them as part of a standard, off-the-shelf unit.
The transceiver expertise – the equivalent of Ericsson Silicon's ASICs – comes from Boston-based Analog Devices. In the digital front end, it will also be responsible for carrying out the analog-to-digital and digital-to-analog conversions that need to happen. When it comes to the digital signal processing – a part of what the industry calls "baseband" – Marvell is supplying the smarts.
Baseband is usually associated with central and distributed units, entirely separate from the radio unit within the radio access network. Indeed, open RAN has momentum largely because of interest in using one vendor's baseband products in conjunction with another's radios – something difficult or impossible with the limitations of older interfaces.
With massive MIMO, however, it can be an advantage to include some of the "Layer 1" baseband processing in the radio unit because of the need for things like "really fast uplink calculations," says Heavy Reading's Brown. "One of the challenges for massive MIMO is that you need to co-optimize the radio unit and distributed unit to achieve the best performance," he says. As Hynes explains it, massive MIMO takes the digital signal processing and "sucks that up onto the mast."
Benetel, his own firm, is performing the crucial role of integrator, stitching together the Analog Devices and Marvell components and adding some of its own magic. Adrian O'Connor, Benetel's CEO, says that means "putting it together and addressing the system performance requirements so the overall end-to-end system works as a Tier 1 product would."
The plan is to have a prototype radio unit available for the second half of the year. An official release on the partnership suggests a 32T32R unit is the initial target. O'Connor also promises compliance with a fronthaul interface developed by the O-RAN Alliance, the open RAN specifications group. In theory, this would allow an operator to run Benetel's radio units alongside central and distributed units from companies such as Radisys or Accelleran, a Belgian developer that O'Connor singles out as an example of European capability.
A big question is whether these new open RAN interfaces can ever give operators the assurance that products from different vendors will work together as an instant mix. While confident, O'Connor concedes the industry has not arrived at this point. "That is why close partnerships with central and distributed unit vendors right now are critical," he says. "There is a lot of handshaking to do and some way to go before it is a very simple plug-and-play for everyone."
To succeed in massive MIMO, Benetel and others must also produce specs that do not look embarrassing on the technology catwalk alongside Ericsson and Huawei. "I don't think you can go head-to-head with a Tier 1 vendor this year or early next year in terms of size and weight – not with a reference design," says Hynes. "If you say here is a specific opportunity with specific targets, and you narrow down your frequency range, it becomes a very customized product, like the Tier 1s are offering, and that is when it gets interesting."
Detractors will continue to argue that smaller vendors stand little chance of replacing Ericsson, Huawei and Nokia, which may collectively have spent nearly $30 billion last year on research and development (not all on radio or even mobile, it should be noted). But competing in the main radio access network market will not be the priority for many challengers, O'Connor points out. "The requirements for an oil rig or smart construction site are very different than if you are supplying to Vodafone or Telefónica."
For a firm like Benetel, with only 50 employees, the sum of the different parts is also what matters. "The path to success is partnering with the right companies on that end-to-end solution," says O'Connor. It is worth remembering that Analog Devices and Marvell are especially muscular partners, together generating about $8.5 billion in annual sales.
Love it or hate it, open RAN may have one thing going for it that traditional vendors do not – the willingness of the world's biggest operators to forgive a slightly heavier or pricier radio unit in the interests of boosting competition. Telco enthusiasm for open RAN proves the traditional model "isn't liked," in the words of Hynes. That could turn out to be far more important than any weight difference.
- Verizon to start deploying open RAN gear this year
- Open RAN gains global momentum
- Airspan to go public with $166M in pursuit of open RAN, 5G
- Huawei spoils Ericsson party with 5G radio weighing 1kg less
- Ericsson now boasts superior 5G radios to Huawei
— Iain Morris, International Editor, Light Reading