Since the early eighteenth century, visitors to St Paul's Cathedral have been able to whisper against the gallery walls inside its immense dome and be heard clearly on the opposite side – as if the speaker's lips were pressed against the listener's ear.
Not an intention of Christopher Wren's original design, this Whispering Gallery phenomenon remains a science tutorial for schoolchildren and adults alike.
"At the acoustic level, that is actually beamforming," says Ray Dolan, the CEO of Cohere Technologies.
Three hundred years since the famous London landmark first opened, beamforming has become a hotly discussed aspect of 5G, the latest mobile standard.
The idea is to harness the wireless signals that flow around any cell site, directing them toward the intended recipients like a whisper across a dome.
Done effectively, beamforming could massively improve coverage and capacity, lessening the need for mobile network equipment and boosting the value of spectrum, say experts. It is what makes Cohere such an exciting venture.
Now ten years old, the Silicon Valley-based company claims to have pioneered a version of beamforming that is far superior to any other product in the market.
It is taken seriously, too. Trials have led Vodafone to estimate that Cohere's technology could double the capacity of a 700MHz network, compared with the use of traditional technology. For a network based on 3.5GHz, more typically associated with 5G, it reckons a four- to fivefold improvement is possible.
That is a whisper of encouragement for Dolan, who became chairman and CEO in October 2018.
"It is a really cool aspect of our story," he tells Light Reading.
"We can make spectrum more valuable by doubling or quadrupling it."
The suggestion that multi-billion-dollar frequency investments could – with Cohere's technology – be worth several times that much seems to have lured investors, too.
"We just funded off that and now the company is extremely well capitalized," says Dolan.
While Cohere is not for sharing details, Dolan believes his company has enough funds to support growth next year and in 2023, when it is expected to be profitable.
The latest round comes after Cohere raised $35 million back in 2015 from companies that included Telstra Ventures, the investment arm of the Australian telco. Other service providers also look intrigued by its technology. Germany's Deutsche Telekom was involved in trials last year, and there is no shortage of interest in North America, says Cohere.
Out to beat the big guns
What makes Cohere so much better than anyone else? As Dolan tells it, the company's boffins have essentially borrowed techniques used in the radar industry and adapted these for wireless communications.
A system known as Delay Doppler assesses both time and frequency measurements to create a spatial map of the channels between a mobile tower and its users. Beams can then be adjusted for better performance.
"In that world, beamforming is like a lighting director who doesn't just have better pencil beams but also a script so that he can count on actors being where they are supposed to be," says Dolan.
"When people say you are just another beamforming company, I say no. We have literally tamed beamforming."
There is bound to be a degree of skepticism that such a small firm can outdo the likes of Ericsson and Nokia, which together spent almost $9 billion on research and development last year.
As a core capability in wireless communications, beamforming remains a priority for both Nordic vendors. Being outclassed by Cohere would be like a chess grandmaster's defeat by a child prodigy – embarrassing and game-changing in equal measure.
What Cohere does have, though, is pedigree. It was founded by Ronny Hadani and Shlomo Rakib, two Israeli scientists behind a modulation scheme called orthogonal time frequency and space (OTFS), conceived as a potential alternative to the OFDM system now used in 5G.
Cohere still sees this as a "next-generation waveform" and Gabriel Brown, a principal analyst with Heavy Reading (a Light Reading sister company), thinks Cohere may try to position OTFS as a 6G candidate.
Dolan's resume is similarly impressive. Something of a wireless industry veteran, he was the co-founder and CEO of Flarion Technologies, sold to Qualcomm for $805 million in 2005.
"That is basically 90% of the intellectual property that is in OFDM," he says. "It is 99% of what Qualcomm does and they are worth about $200 billion. I think they got a good deal."
Replicating OTFS or the Delay Doppler stuff would not be easy – partly, says Dolan, because Cohere has "patented the heck" out of its innovations. In the meantime, its rivals have been on different beamforming paths, he reckons.
Methods described by Nokia in a recent white paper on massive MIMO – an advanced antenna system that underpins beamforming – include grid of beams (GoB), codebook-based beamforming and Eigenbeamforming.
Yet in less advanced MIMO settings, Cohere has attracted praise where others have been criticized. Carlos Ubeda, a technical expert at Vodafone, takes aim at existing vendors of 4T4R (four-transmitter and four-receiver) units.
These companies, he said during a recent Light Reading webinar, have not been able to handle multi-user MIMO, allowing numerous devices to communicate via multiple antennas. Cohere's technology made that possible and led to the performance improvements that Vodafone touts.
"Today's capacity in 4T4R is typically 50% below the upper bond and this is because legacy suppliers don't support multi-user MIMO and so peak throughputs are achieved mainly by assigning two layers to the same user – single-user MIMO," he said.
"Thanks to Cohere and its channel-estimation procedures based on Delay Doppler, we can efficiently address the challenges of multi-user MIMO and multiplex two users at the same time. That means doubling capacity."
But if Cohere is so advanced and so hard to imitate (legally), why has it not already been acquired by Ericsson, Nokia or someone else in the wireless business?
James Crawshaw, a principal analyst at Omdia (another Light Reading sister company), speculates that Cohere would demand a higher price than either Nordic vendor would be willing to pay.
They may have surmised they can develop the technology for less in-house and then beat Cohere in the courtroom. Buying Cohere, moreover, could herald embarrassment for either's research-and-development unit.
Even a tie-up would be awkward for them, acknowledges Dolan.
"This could be incorporated into a traditional RAN, but it would require the existing guys to say please do this for me – please be my brain and decide what I am going to eat today," he says.
"That is the essence of who they are and so it will meet some commercial resistance, but technically it is quite simple."
The open RAN way
That sort of commercial resistance could be a serious obstacle, though. Cohere, much like other RAN specialists, instead hopes open RAN will unlock the gates.
If it boosts interoperability between suppliers – as proponents say – then Cohere could feasibly slot into a multivendor network built by numerous specialists. Operators including Vodafone and Deutsche Telekom are keen.
The Cohere pitch, accordingly, is that its technology can run entirely as a software application on the RAN intelligent controller (RIC), a new platform the open RAN community has conceived. Separated from RAN equipment and software, it could exist in the cloud and still work its magic, insists Dolan.
That is largely because the Delay Doppler system has been engineered to provide 100 milliseconds for the set-up of a real-time beam, ensuring latency is not a constraint.
"It's a myth that you can't reorganize into the cloud because the channel is changing so fast that you need to leave all the intelligence as close as you can to the handset," says Dolan.
"That serves the interests of people who want to keep the system closed."
If he is right, Cohere will be able to buddy up with RIC platform developers and integrate with various networks. It already has a partnership with VMware and is working with another player whose name has not yet been disclosed.
Not all RICs are equal, though. Near-real-time RIC, a successor to the distributed SON (self-organizing network) technology found in some traditional basestations, operates on time cycles of less than one second.
With longer time cycles, non-real-time RIC is more akin to the centralized SON technology that would, as the name implies, sit on servers in a central location. Where the near-real-time RIC comes with xApps, the equivalents for the non-real-time RIC are known as rApps. And Cohere does not appear to have one.
"We are not as focused on non-real-time," says Dolan. "There is not as much value there for us to act."
There are a couple of downsides. The first is that Ericsson, which launched a non-real-time RIC last month, looks resistant to the xApps concept.
"They think the value and the intelligence should sit in the nodes," said Crawshaw during a previous interview with Light Reading.
"They don't want the value being sucked out and stuck in a near-real-time RIC that VMware might sell."
The second issue is that interface specifications for the near-real-time RIC are still not ready.
"The non-real-time RIC is quite far ahead and there is a roadmap to commercial deployment," says Heavy Reading's Brown. "With the near-real-time RIC, it is still early."
Even if Cohere does perform no worse as an xApp than it would if integrated with nodes, this all means it will probably need to form partnerships with the suppliers of distributed units (DUs) – the parts of the RAN that handle baseband processing.
"The route to commercializing this is going to involve agreements with DU software suppliers," says Brown.
Live up to its billing and Cohere might turn the industry on its head. Massive spectral efficiency improvements realized entirely through software will be hard for anyone to ignore – whether operators that have spent billions on spectrum or kit vendors fearing commoditization.
The big questions are whether it works as well in the field as during Vodafone trials, how easy it is to deploy and if it really is a hard act for rivals to follow. By this time next year, the answers should be clearer.
- Cohere CEO: Open RAN ideal for enterprise applications
- Podcast: O-RAN discussion with CEO of Cohere Technologies
- Cohere CEO: New product, customer names coming this year
- The People's Front of Open RAN vs. the Open RAN People's Front
- Cohere crashes the open RAN party
— Iain Morris, International Editor, Light Reading