A group of companies, some very well-known and some less well known in the mobile network space, have joined up to develop a commercial AI-native 6G wireless platform that can be deployed at scale in mobile operator networks.
The group is led by Nvidia, in partnerships with T-Mobile, MITRE, Cisco, ODC, and Booz Allen Hamilton
The group will jointly develop “an AI-native wireless network stack based on the NVIDIA AI Aerial platform”, which is Nvidia’s CUDA-accelerated software RAN platform.
The stated goal is to develop more than a blueprint or reference model, or to create an interesting lab project. Nvidia told TMN, “The group of companies will work together on research and development to build, test, integrate and deliver a minimum viable product (MVP), with an end goal of commercial platform for global scale deployment with X-factor gains.”
NVIDIA and telecom industry leaders to collaborate on the research and development of
AI-native wireless networks for 6G.
Don’t call it an Alliance
Ronnie Vashishta, who heads up Nvidia’s telecoms business, told reporters and analysts on a call that the partnerships do not constitute an alliance or a consortium, rather a “group of likeminded companies coming together” – given T-Mobile’s presence perhaps he should have called it an un-Alliance. However, legally-cleared semantics aside, a looser federation of partnerships would also offer scope for the companies involved to make individual decisions as they go along.
Vashista told reporters the companies are intending to define a next-gen platform that is AI-native, “with AI embedded into hardware and software from day one”, accelerated and multi-purpose, meaning it has the ability to support multiple workloads.
AI-native signal processing will be necessary, Vashishta said, to unlock gains in spectral efficiency that will be required to connected “billions of smart devices” and support AI traffic in the network.
New capabilities such as Integrated Sensing and Communications (ISAC) require a new wireless stack built from the ground-up using AI, Vashista said. “That’s not possible with an incremental approach, bespoke hardware or using conventional algorithms,” he added.
It seems the goal is also to expand the range of companies that could get involved in the AI-native wireless network.
Nvidia told TMN, “The companies also aim to democratise the AI-native wireless technology to create a vibrant ecosystem and innovation.”
“USA-USA”
Although not called out by Nvidia, it also appears an additional aim might well be to create a US-based IP capability, with all of the partner companies based in the USA.
While T-Mobile is already a partner of Nvidia’s developing its AI-RAN vision, and Cisco is a huge network software entity, MITRE and Booz Allen Hamilton add to the US feel around the AI element.
And Nvidia is not working with an established AI-RAN partner for the Cloud RAN software. Instead it is partnering with the little-known O-RAN Development Company (ODC) for software at Layer 2 and above.
The O-RAN Development Company is a bit of an unknown entity in the public sphere. It sits within the portfolio of Cerberus Capital, whose co-founder and CEO has just been confirmed as Deputy Secretary of Defense of the United States.
In a previous release announcing work with African and Middle East operator MTN, the company named Matthew Johnson as its head. Johnson is a former Altiostar and Rakuten Symphony executive who works for Cerberus. The Nvidia press release carried a quote from former Juniper Networks CEO Shaygan Kheradpir, naming him as “chairman of the advisory board of ODC.”
The ODC’s newly-updated website claims to develop next-gen cloud-native radio baseband (DU-CU) software that is AI-native and 6G-ready.
Its website says it is also developing Radio Units (O-RUs) using radio silicon from Analog Devices, building a flexible platform that can support radio bands that will “meet the needs of our two major operator partners.”
The company is clearly leaning heavily into the “US ecosystem”, marketing itself as a US-based play to develop a wireless communications business.
The presence of MITRE, which operates federally funded research and development centers, and Booz Allen Hamilton also point to the seeming motivation of developing US-led AI technology within the 6G wireless platform.
Nvidia’s expanding telco ambitions
The driving motivation for the grouping clearly comes from Nvidia’s desire to foster an ecosystem that will build wireless solutions upon its accelerated signal processing capabilities. If that ecosystem happens to include the established wireless vendors, then all well and good, but it can also incorporate newer entrants and specialist AI developers.
As well as the formation of the un-alliance, Nvidia also announced new tools aimed at enabling companies to develop AI-RAN and 6G solutions.
It announced an updated Aerial Commercial Test Bed, which is a full-stack system that includes its CUDA-accelerated RAN software, L2 and L3 OpenAirInterface software, hardware from SuperMicro and radio units from LITEON and WNC.
And at its GTC event it is demonstrating a trial of Neural Network for Channel Estimation that has been trained and developed on NVIDIA AI Aerial, which shows that AI-enabled channel estimation could improved real-world spectral efficiency by 40%.
As well as that, the company has released the Sionna Research Kit. With the kit, researchers can connect 5G equipment and begin prototyping AI-RAN algorithms for next-generation wireless networks.
Sionna 1.0, released at GTC 2022, is a link layer simulator that has had 150,000 downloads and been cited in 300 publications. The Sionna Research Kit is powered by Nvdia’s Jetson platform and integrates acceleration for AI and Machine Learning techniques with the Open AirInterface RAN stack.
It is available for researchers to try out in hardware over the air, with Sionna and the Sionaa Research Kit playing together as a link layer simulator and as hardware platform to try out.
Plus, Nvidia announced that its Aerial Omniverse Digital Twin (AODT) simulation platform is now available on NVIDIA DGX Cloud.
