- Federated Wireless is helping operators get more bang for their buck in shared spectrum
- Spectrum AI is a physical AI platform built for shared spectrum planning and coordination
- The initial release is designed for large and mid-sized operators
Amplex Internet, a wireless ISP based in Ohio, is getting so much interference from higher-powered Canadian wireless carriers that its internet customers are losing connectivity – sometimes for hours at a time. But if Amplex had access to Federated Wireless’ newly launched Spectrum AI and related tools, it would have avoided all that disruption, according to Federated Wireless CEO Iyad Tarazi.
“That is exactly what Spectrum AI does,” he told Fierce. “In the past, we would just assign a channel and say, let's see what happens, we think there's a good probability you're not going to have any issues.”
Now, “we simulate what's going to happen before they deploy to the point they know what price they’re going to charge the customer, what speed they’re going to have,” he said.
Federated Wireless today announced general availability of Spectrum AI, which follows on the Adaptive Network Planner (ANP) the company unveiled earlier this year. Together, they’re providing network operators with the tools they need to make better informed decisions.
With a 3D modeling tool, network engineers can look at rooftops to see where they want to put an antenna and whether they would get a better signal if they moved it another meter or raised it a couple of feet.
Federated’s AI for CBRS, 6 GHz
Federated describes Spectrum AI as the first physical AI platform purpose-built for shared spectrum planning and coordination. It’s applicable for CBRS Priority Access License (PAL) and General authorized Access (GAA) users as well as 6 GHz operators and networks constrained by incumbent-protection or coordination requirements.
Specifically, Federated is using an open source model from Nvidia called Sionna RT that’s implemented inside an Amazon cloud using Bedrock technology.
“The biggest impact we've seen so far is from doing this 3D simulation of the network because that's what the Nvidia model does. It allows us to do literally billions of simulations for every cell site, trace every single ray of electromagnetic wave, and where it's going to go,” he said. “You can actually simulate in every home what's going to happen … The prediction is very accurate, equivalent to doing a drive test yourself.”
Spectrum AI for mobile, cable operators
Tarazi said these new tools should be well received by mobile operators because they’ll see better spectrum efficiency and performance.
Spectrum AI was inspired by discussions with big mobile and cable operators and regional WISPs that complained about their inability to offer full speed, full capacity services due to overly protective guard bands.
“In the old world we couldn't predict and therefore we had to build just in case, so we don't end up with that [Amplex] issue on the Canadian border very often,” he said.
About 25% of sites for these big operators have some sort of restrictions on them. “They're not able to do everything at full speed because we're trying to protect them from each other,” he said.
Once they came up with a way to do these simulations, “we were able to remove 99% of these restrictions,” he said. “We monitor interference and resolve it before they have to deal with it, which we weren’t doing before. We were just assigning channels.”
Tarazi explained that most AI in the wireless industry operates above the physical layer, automating workflows or optimizing software. Spectrum AI works at the radio and propagation level in the physical world where coverage, interference and capacity are actually determined.
Field deployments show up to five times improvement in network capacity and up to 50% more usable spectrum capacity. They also demonstrated up to ~20 dB improvement in interference coordination accuracy, validated across live shared-spectrum environments.
From weeks to seconds
Tarazi has been at the helm of Federated Wireless, a company he co-founded, for more than 10 years, but he spent almost nine years of his career in the network side of the house at Sprint.
Fierce asked how today’s network tools compare. “I’ve never seen anything like this. When I used to run a network plan like this, it would take me two weeks and a server farm to run it. Now I could do this in seconds. I've never really been able to simulate at this level.”
One operator called him to say they’re not going to do drive testing anymore in their state – they’re just going to use the Federated Wireless tool.
That’s not to say it was easy to get to this point. Federated Wireless spent two years pulling all the data together so that it has details about every building, structure and tree out here – the same kind of data used by self-driving vehicles, for example. On top of that, they’re constantly collecting more data from the field and putting it in the AI engine.
“We’ll never be done,” he said. “You have to constantly update your data and constantly run your models and they keep learning and getting better. It’s really a new world.”