Massive MIMO has been crucial to the performance of early 5G, and deployed at a far earlier stage and faster rate than was forecast a few years ago. The antenna arrays, featuring between 16 and 128 transmit and the same number of receive elements, allow operators to boost cell range, and so to use the same site grid that supported 2.5 GHz or even 1.8 GHz LTE, for 5G in higher bands around 3.5 GHz. The technology also supports sharp increases in capacity and cell edge performance. But it is at the start of its evolution. Already, 256T256R platforms are being tested. Huawei has applied a variant of the technology to an indoor architecture called Distributed Massive MIMO.
One of the next big challenges, which will be essential to operators’ goals for their next phase of 5G build-out, are to use Massive MIMO to improve 5G uplink to the same level as downlink, which will greatly improve user experience and enable a variety of new industrial applications that have more symmetrical data patterns than traditional apps based on Internet downloads.
Another key challenge is to develop Massive MIMO radios that are fully compliant with emerging Open RAN interfaces and architectures, enabling them to be implemented with basebands (centralized or distributed units) from a different supplier. This will also help to expand the performance of Open RAN platforms beyond their main home in smaller cells and into the high end macro network with its hugely demanding 5G requirements.
Both these challenges featured in vendor announcements last week. Ericsson claimed a record for 5G uplink performance, using Massive MIMO. Meanwhile, Mavenir and Xilinx provided details of a joint development of O-RAN-compliant Massive MIMO, and Samsung and Marvell have a similar co-development. In outlining progress so far, Mavenir also placed emphasis on the importance of uplink performance to many 5G applications.
Ericsson recently released Massive MIMO technology that can support traditional or virtualized RAN architectures, but of course, is not supporting O-RAN fronthaul links to third party basebands. Nor does it believe in the claims of the open network community that a high end Massive MIMO 5G network can be deployed on general purpose processors (GPP) or merchant accelerators, and has placed its homegrown Ericsson silicon at the heart of its differentiation story for its radios.
Last week, it claimed it had hit a record for single-user uplink rates using MIMO, achieving 15-20 times faster uplink speeds than are generally available today, at up to 315Mbps. It used single-user MIMO combined with 256QAM modulation and its Uplink Booster technology to reach these uplink rates, on a 5G Standalone network in a 100 MHz channel of spectrum in the 3.7 GHz band.
The speeds compare to common 5G commercial uplink rates of between 20Mbps and 30Mbps. A February report by OpenSignal said The Netherlands supported the fastest speeds, at around 32.5Mbps, but Ericsson said the average uplink rate in the USA is only 10Mbps.
Enhanced uplink is expected to be a key requirement as operators move to the next phase of 5G build-out and will often accompany a migration to the 5G core, to new spectrum bands such as the USA’s C-band, and to new antennas. Huawei and ZTE have also outlined uplink enhancement technologies that improve on the rates enabled by vanilla 5G specifications.
Uplink improvement is also one of the goals of Massive MIMO developments within the Open RAN community. But moving from customized and highly optimized inhouse silicon to merchant platforms is one of the biggest challenges facing Open RAN. The beamforming associated with Massive MIMO, combined with the Layer 1 performance needed to support real time processing, demand at least 30 times more processing power than LTE with 4T4R MIMO. This is hard to achieve with general purpose processors, especially while keeping cost and power consumption down.
Yago Tenorio, Vodafone’s head of group network architecture, told TelecomTV late last year that “5G Massive MIMO in particular is a complex problem to crack and will require not just a lot of software development but also hardware, in particular semiconductors… we will need new solutions to be developed in terms of chipsets that are powering those efficient, super-complex radios.” He speculated that a solution could be developed in 12-18 months.
However, several merchant chipmakers have identified Open RAN as an important way to break the dominance of proprietary silicon and gain a position in the cellular access network at last. Intel, Marvell, Qualcomm, Xilinx and others are bringing their engineering prowess to bear on the problem in different ways, and announcing their first platforms and partners.
The aim is to bring Massive MIMO within reach of a larger array of base station suppliers, not just the rarefied group that currently have these capabilities at scale. Some of those large vendors do support Open RAN platforms and have started to open interfaces to their next generation radio/antennas. NEC has developed such a product for Rakuten and will now offer it on a broad commercial basis; Nokia has pledged O-RAN interface support for some of its high end radios. But the promise of Open RAN economics and agility rests on a broad ecosystem in which many suppliers of radios and of baseband software can interwork.
Xilinx and Mavenir say their jointly developed Massive MIMO radio unit, which will support Open RAN specs, will be in field trials later this year. The product will support 64T64R array with bandwidths up to 400 MHz. It is based around a combination of Xilinx components – its RFSoC DFE (RF system-on-chip digital front end) and its processors for Layer 1 functions and beamforming. The DFE has a mixture of programmable components and single-purpose ASICs.
Mavenir contributes the Open RAN software as well as some integration, management and monitoring services. The company, which entered the Open RAN market as a software-only provider, has been spreading its wings lately and seeking to drive an ecosystem of radio/antenna vendors that support its software, and the broader O-RAN specs. It has acquired ip.access, a vendor of small cell hardware and software, and formed partnerships with Taiwanese radio manufacturers. But this joint development is the first to target the high end of the 5G macro RAN.
The Massive MIMO product will comply with the O-RAN Split 7.2, the O-RAN Alliance’s chosen open fronthaul interface between the radio unit and the baseband. That means the new platform will support O-RAN software from other suppliers as well as Mavenir, and in future there should be plugfests to prove that the interoperability between multivendor hardware and software does work at scale. Mavenir and Xilinx have already held lab trials in India involving six large operators.
Other companies are working on Massive MIMO for Open RAN, aware that this is a passport for the technology into the macro 5G network, and beyond its initial heartlands in rural extensions or small cell deployments. Marvell is working with Samsung, and also with Benetel – the latter has a development that combines elements from Marvell on the processing side and Analog Devices for the RF. Xilinx also has a partnership to power Fujitsu’s Open RAN 5G radios, which will be deployed by Dish Network in the USA, and Marvell has its own Fujitsu relationship too.
The hope will be that, with enough combined R&D effort and budget, the community can approach parity with the big vendors’ platforms within a reasonably short timeframe, and that could even bring forward the timeframe in which most operators believe Open RAN will be ready for the 5G macro big time (about 2025 is the current consensus among the main Open RAN operator supporters).
“For us to be competitive in the marketplace, we’ve got to be of equal size, weight, and power consumption,” Mavenir’s SVP of business development John Baker said in an interview. He said the partnership with Xilinx demonstrates an important claim that will be essential to the success of Open RAN – that high end performance could be achieved in the short to medium term with a high degree of general purpose silicon. “You don’t need to go to full custom ASIC any longer to get the cost savings and the power consumption numbers that are generated,” Baker said.
Of course, the proof of that pudding will be in the eating, and Mavenir is promising full details of the Massive MIMO platform, and the partners and operators involved in tests, in a few months’ time. The first target market is likely to be the USA and its upcoming deployments in the newly allocated C-band spectrum – the government, and challengers like Dish, are promoting the idea of a US-centric Open RAN ecosystem that would reduce reliance on Chinese firms, and build up an America-first industry. It is not clear how far the big three operators will support this dream – they are still mainly working with the largest OEMs, though they are testing various open architectures, and two of the vendors to AT&T and Verizon, Nokia and Samsung, will support O-RAN if the customer requires it.
Samsung will start shipping its latest Massive MIMO products, which will support O-RAN options, this summer. These will be based on a system-on-chip jointly developed with Marvell and will target midband deployments such as those by US operators in the C-band spectrum. The SoC approach will drive down size, weight and power consumption, said the firm – recently, Nokia and Ericsson have also launched Massive MIMO units with bold claims about size and power efficiency. “This is a single-chip solution and therefore it compresses quite a lot of functionality and algorithms into a single SoC, and it drives down power consumption and it improves the size of the radio,” Alok Shah, VP of networks strategy at Samsung Networks, told SDxCentral.
Samsung may be supporting Open RAN, but it is also keen to stress its own semiconductor credentials, and to dismiss the idea that 5G networks could run on truly general purpose processors in future. “One thing that’s become very clear in our industry is the importance of chips, the importance of silicon when it comes to 5G,” Shah said. “Samsung has an innate advantage by being an expert in semiconductors ourselves, so we have a level of inhouse expertise”. In addition, Marvell contributes “expertise in processors to bring the most effective beamforming chips that we could to market”.
The partners claim the single-chip design delivers a 70% reduction in power consumption on the radio compared to multi-chip designs or FPGAs.
“There are other SoCs that perhaps sit in the baseband unit, that do the modem functionality, this one actually sits within the Massive MIMO radio, and it runs all the beamforming algorithms that are so important to get the capacity that operators are looking for,” Shah explained.