Fitter, stronger, lighter…

Ericsson highlights role of custom silicon development and tight software integration as it slashes size and weight of its massive MIMO active antenna radio unit.

Ericsson’s first 64TR massive MIMO antenna-radio was the AIR 6468, released in 2018. Depending on spec, it weighed up to 60kg. Last year’s release, the AIR 6449, weighed 37kg. Progress.

This week Ericsson has announced its fourth generation of 64TR radio antenna units – the AIR 6419 family – and it says they weigh just 20kg each. That’s a 45% reduction in weight on the previous year’s version, given like-for-like specification options. The passively-cooled, 320W output power units also feature a 15-20% reduction in power consumption compared to the previous model, Ericsson said. The three variants, the 6419, 3219 and 3258, will be generally available in the fourth quarter of 2021.

The M-MIMO launch was of three new designs. The AIR 6419 is “performance optimised” – so everything is focussed on the top performance. Then 3219 has the same output power, the same number of antenna elements, but 32 branches rather than 64. So it has the same gain and coverage, but less digital degrees of freedom in elevation. If you have a sky-scraper type city, you need a wider vertical flexibility where you can steer per millisecond in a very wide vertical sector. Whereas with the 32 branch you have a more narrow vertical beam, and you can only tilt the antenna electronically. That optimises for a balance between TCO and performance, and is more suited to lower rise European cities.

Alongside the new antenna-radio units, Ericsson is launching a new line of its RAN Compute products, which is its name for the Ericsson Radio System baseband platform. The company says that the baseband units also have a 20% reduction in power consumption compared to the previous version.

Both the antenna radio units and the Baseband platforms benefit from Ericsson Silicon, the vendor’s custom range of System on a Chip (SoCs).

More performance and cost-optimised antenna units that reduce demands on space and power is something that operators have been asking for. Colin Bryce of antenna manufacturer Commscope told TMN in November 2020, “Operators…  are beginning to question the cost to performance benefits of 64T64R type massive MIMO”. Part of the issue Bryce identified was the size and weight of M-MIMO antenna units.

So how has it achieved these form factor and performance gains?

“If you want to understand how we are building our products, you need to understand what we are doing in our silicon R&D.”

Well, says, David Hammarwall, Head of Product Line Radio at Ericsson, it all comes down to the tight integration – “co-design” – between Ericsson’s System-on-Chip silicon, and its software. “If you want to understand how we are building our products, you need to understand what we are doing in our silicon R&D.”

He said that although Ericsson’s R&D in its own SoCs has increased significantly over a period of four to five years, the company is just now starting to talk much more about the benefits of a tight integration between silicon and software.

“We have more efficient silicon so the inherent power consumption goes down, so you can take out more cooling. [The antennas are all passively cooled, which is enabled by the smaller overall size of the unit.] You can have more efficient PA linearisation algorithms, so that makes the PAs more efficient and again takes out less add-on weight for cooling. And then you also have increasingly integrated design in the radios, with very tight co-design between software and hardware where we are able to process more efficiently, so we don’t encourage so much power.”

The issue of how and where to process the complex calculations needed to support M-MIMO beam forming is critical. An Ericsson blogpost from 2020 said:

“Ericsson’s Massive MIMO architecture has been designed to put as much as possible of the beamforming and MIMO processing in the antenna radio unit itself, gaining access to real-time and fine granular information about the radio channel. Therefore, Ericsson is able to do channel estimation and beamforming weight calculations that follow the extremely rapid changes that occur on the radio channel almost instantaneously.

“To generate ultra-precise beamforming, a massive set of complex calculations needs to be performed in real-time, scaling with the number of antennas, the bandwidth and number of users. This adds up to millions of mathematical calculations per second, which requires an extreme processing capability. In addition, it also requires our sophisticated software features and algorithms to make sure that we leverage that hardware in the best way. This can only be achieved with Ericsson custom silicon, system on a chip (SoC) solution.”

Hammarwall said that it is the ability to re-use IP blocks within its ASICs that gives Ericsson its processing advantage, and hence the ability to build small.

“When we work with custom silicon ASICs, we develop in-house a portfolio of IP blocks that we can re-use in different SoCs –  composing different chips where you can pull on your IP. We have a very rich portfolio of RAN-optimised IP blocks for our SoC family; we started to talk about one in London last year, the nanocore architecture. That’s an important block where we do a lot of the RAN optimised processing like L1 and beam forming. It’s a highly parallelised architecture where you can have hundreds of DSP cores, and we spent significant effort over many years to develop the SDKs to take full advantage of that parallel architecture.”

“Within the radios there are two different SoCs – one optimised for the Digital FE, all about linearising the power amplifiers, securing good bandwidth so can have 400MHz and beyond in bandwidth. So we have an optimised SoC for that. Then we have  a separate SoC optimised for L1 and beam forming. So it’s those two chips combined that allow us to really drive all of this computational processing. And it also allows to upgrade to a more performance-optimised RAN split than we see the industry embracing, where we can put more compute and processing in the radio.”

The tight integration of the ASIC with the software is the game of the moment in 5G architectures. Referencing Google Cloud and Tesla’s on-board AI, Hammarwall pointed out that other areas that are very computationally expensive are driving towards integrated, dedicated silicon solutions. Ericsson has said previously that its power amplifier engineers design new solutions that can handle more power and bandwidth and fit into a given space by integrating several discrete blocks into one single package.

Again, here’s previous messaging from Ericsson, from January 2021 this time: “This also applies for low level radio frequency and signal processing, which we also partly shifted it into the digital domain to utilise the latest semiconductor processing technologies.

“Additionally, with new advanced functionality like beamforming and interference suppression our solution will benefit from integrating RAN Compute and the radio harder. By moving some of the RAN Compute functionality into the radio, RAN transport requirements can be reduced and the radio can adapt quicker to the constantly changing radio channel conditions. This allows for significant improvements of performance, size and energy efficiency.”

In 2019 Intel did publicly state that Ericsson was a customer for its SnowRidge 10nm design, “for 5G basebands”. But Hammarwall would not talk about who Ericsson partners with. “When it comes to partners, we have very substantial R&D in this area, in both the US and Sweden, and we are also partnering with all the players in the industry for SoCs. We don’t talk about it publicly but we are partnering with all the big players for building ASICs.”

Cloud RAN?

Ericsson said in autumn last year that it was investigating Cloud RAN, and laid out some of its thinking on the topic. It said that by the end of 2021 it would have virtual Central and Distributed Units (vDU, vCU) that support low band 5G operation, as well as a Network Gateway that can interface between its vDU and already-deployed Ericsson remote radioheads. Eric Parsons, Head of Product Development Unit, Cloud RAN, told TMN that the vDU and vCU would be deployable either on Ericsson’s cloud platform or on any Intel x86-based COTS hardware without any further hardware optimisation or acceleration.

Is there a synergy, then, between the very tightly integrated radio units, and the cloud vDU-vCUs, which might be operating on COTS hardware?

Hammarwall says yes, “Actually I would say that they synergise really well with a Cloud RAN offering because we actually move more compute into the radios – so you actually offload the processing by doing more of it in the radio.

But Hammarwall stuck to the Ericsson view that, for now, an open Cloud RAN architecture is best suited for AI and management type applications. But he also reiterated that Ericsson is contributing to all of the Open RAN Alliance’s working groups, including those working on lower layer splits. “However, when it comes to these particular products, we see price and performance for integrated solutions are better than the O-RAN lower layer split in the short run.”