A tale of Open RAN innovation from Cohere Technologies

Cohere Technologies' shift is an example of how an Open RAN "ecosystem" can aid innovation.

Some governments and many operators are keen to see a wider choice of vendors supplying mobile network infrastructure.

One proposed mechanism is to “open” the RAN to more players, not just to create another massive vendor or two, but to create a hardware and software engineering environment in which all sorts of vendors could play. 

Openness means many things – for instance it can mean open interfaces between the disaggregated functions, so that an operator can use different vendors across the RAN. It can also mean openness vertically, so that an integrator or operator can easily use software applications from different vendors – say putting some SON or ML-based function from one vendor to work on the radio from a different vendor. 

One of the pressing issues for those developing the open, virtual RAN is how to do the sort of very sophisticated, fast processing that is required for Massive MIMO and 5G using commercial hardware platforms. Up until now, the main radio vendors have used their own customised silicon solutions within dedicated appliances and servers for radio processing. This has enabled them to optimise performance for the functions they need to support. 

Huawei, for example, has been explicit that that it does not think Intel-based COTS servers can meet performance and power requirements for 5G RAN. However, nobody has really suggested merely putting those functions into a general purpose environment and letting them do their best. Instead focus has fallen on hardware acellerators – FPGA, GPU or otherwise – to accelerate real time sensitive processing for the lowest layers of the radio baseband. When Altiostar deployed with Rakuten on Intel-based x86 there was a lot of work to do on the FPGA-based hardware acceleration, for example. And in Nokia’s announcement last week of its vDU (Distributed Unit) for 5G, it said it is using Intel FPGA-based hardware acceleration for near real time L1 processing. 

For the future, Ericsson and Nokia are also looking at GPU-based acceleration for some vRAN workloads, especially for 5G M-MIMO and for future AI and ML use cases. One challenger vendor, Parallel Wireless, has pointed to the need to invest more in different chip technologies: it says Huawei has taken a lead in Gallium Nitride solutions – which has mean it can build M-MIMO antenna units that are lighter and perform better than their competitors. 

So far, so much work on the chip and hardware end of things to enable the vRAN, and by implication a more open RAN as well. 

Using the open platform to improve the system

But you can come at this from the other end – using the open cloud platform to add software smarts, for instance to look at how you actually process the beam forming itself. One company with a different approach to processing beam management calculations is Cohere Technologies.

Cohere came to attention at the start of the year when it was revealed to have been part of a Deutsche Telekom trial of new O-RAN and vRAN technology. DT was testing virtual infrastructure from VMWare to host virtual instances of RAN software according to the Intel FlexRAN reference architecture. As well as Mavenir’s vRAN software, this trial included a near real time pre-standard RIC (RAN Intelligent Controller) –  the O-RAN defined element that acts as a host for radio control software. (See more on this here.)

By deploying on the open RIC, Cohere was providing its beam forming software algorithms as a functional adjunct to the Mavenir RAN – doing clever things like channel measurement and prediction in very quick time. And behind that sentence there is a story that illuminates the shift from battling through standards processes to exploiting open platforms, because this is not Cohere’s first foray into the 5G market.

From 5G candidate to smart MIMO adjunct

Cohere first came to attention in the middle of the decade with its proposal for a candidate waveform for 5G NR standards. In 2015, the company’s OTFS (Orthogonal Time Frequency Space) technology had attracted interest and investment from (amongst others) Telstra whose technical leadership spoke in strong support of the spectral efficiency gains the technology could offer. From 2016, Cohere Technologies was hoping that OTFS would be adopted as part of the 5G standards specifications. In late 2017, the company announced that C Spire would be looking at using the new modulation scheme for FWA trials, and it was still stating that OTFS had been “proposed as a part of 5G standards.” 

But that adoption by the SDOs never happened – 5G NR used a modulated version of LTE’s OFDM, with some more flexible numerology, frame rates and new coding. But Cohere travelled on. By 2018, research interest from Telefonica was translating into more trials of FWA applications of the technology. And trials conducted within the 5TONIC programme seemed to show that the non-OFDM technology – now labelled 5G turboConnect by Cohere – would give important efficiency gains. Next on the horizon were “high mobility” use cases. The technology, “leveraged standard architecture and components to implement a software-based cloud RAN solution,” according to publicity at the time.

By late 2018, the company’s new CEO – Raymond Dolan – had set about re-thinking the way the company might exploit the core smarts that underpinned OTFS. That saw a move away from producing an integrated hardware product based on proprietary modulation on the air interface, to using the technology as super-smart prediction software to make multi-user (MU) MIMO work better in any modulation scheme. 

To set up a beam, you need to know the radio channel conditions right now, and also make a prediction as to what that channel might look like in the very near future. Cohere’s technology makes different measurements from other beam forming management methods, and it says that as a result its predictions stay accurate for longer. The company takes uses UL reference symbols to determine and predict channels between transmitters and receivers. It describes this as “Delay Doppler-based channel detection, estimation and prediction, as well as pre-coding software to improve 4G and 5G wireless systems.” Cohere’s marketing claims that the Delay Doppler channel representation is predictable into the future given that its geometric nature is slow changing. That is why its predictions stay accurate for longer, slowing down “channel ageing”. 

That means that the beam forming calculations are introducing less latency into the overall system budget, and because of that, the software can be deployed at a cloud location rather than right out at the base station. This means that, in terms of aligning with the decoupling of functions in the O-RAN model, its software can be integrated as an app at the RIC (Radio Intelligent Controller) node.

Ronny Haraldsvik, Cohere’s marketing lead, says that OTFS-based channel estimation software “can make a material improvement for 4G and 5G, and make Massive MIMO possible in both TDD and FDD spectrum.”

Haraldsvik said, “By taking time channel frequency measurements and seeing a multipath representation of that channel, we can translate that so we can predict what the channel looks like 30-50ms or more later. So it’s not static beam forming; our software is dynamic, and works all the time looking for the best beam in the cell at any point in time. We form a space dimensional view in time and frequency that has not been done before.”

Cohere, clearly, wants to commercialise its solution as a possible answer to thorny issues of where and how to deploy M-MIMO processing. What’s of wider importance is that Cohere’s own shift is an example of how the industry can bring innovation into the market as a result of adopting open vRAN platforms.