To listen to On DoD on your phone or mobile device, subscribe on PodcastOne or Apple Podcasts. The best listening experience on desktop can be found using Chrome, Firefox or Safari.
When the Army’s new Rapid Capabilities Office set about one of its first tasks — providing electronic warfare gear to soldiers in Europe — it became clear fairly quickly that artificial intelligence and machine learning technologies offered a lot of promise for helping EW officers identify signals in an environment where the electromagnetic spectrum is jam-packed with transmissions of all sorts.
But even though AI algorithms are already widely used in commercial industry to classify images, doing roughly the same thing with radio signals was more uncharted territory. So instead of issuing a traditional request for information, the RCO decided to use a “challenge” process, giving industry and academia access to a sample of real-world data from Army sensors and seeing who could produce the most accurate results.
“It’s a different approach, and I think it’s more of a challenge for the community to go and decompose the data, to understand it, and build the algorithms and compete against each other versus just writing a white paper and submitting it,” Rob Monto, the RCO’s director for emerging technologies said in an interview for Federal News Radio’s On DoD. “This way, we’re putting a subset of the problem we want to look at and we’re getting the metrics on how well they’re performing. It’s a way for us to say mathematically they are better than someone else. It’s not just a thought. It may be at a lower technical readiness level, but it’s something that is running today.”
The RCO announced the results of the 90-day signal classification challenge in late August. Out of 150 teams, it picked Aerospace Corporation, a federally-funded research and development center; an Australian data science team called TeamAU; and Motorola Solutions.
They earned prizes of $100,000, $30,000 and $20,000, respectively, after showing that their algorithms were able to deal with the problem the Army is trying to solve, namely identifying signals in the airwaves quickly and reducing the “cognitive burden” on EW soldiers.
“The data science community, at least in this challenge, they were getting better and better at understanding how you would apply artificial intelligence and machine learning to this specific problem space,” Monto said. “There was a rising of the tide, and I think we definitely got to the point of advancing the area in the spectrum space. It was really good to see.”
Where the Army goes from here is still uncertain. The RCO plans to announce a second phase of the competition later this year. It’s still a long way off from actually procuring any of the technologies the challenge participants demonstrated, but Monto said there are definitely pathways in the acquisition system for doing so.
“It could be a prototyping activity where we bring some of these winners’ algorithms into an actual system and integrate it, or we could put out another challenge to advance the dataset that we put out there,” he said. “Initially, it was just a set of modulations that we put out for the community to work on, so we may layer that with different types of signals, like Wi-Fi signals or cell phone signals. I hope the community that was involved in the challenge got some lessons from the environment that we provided. The Army and the RCO learned a lot from the process, and I hope to do more of these.”