DARPA’s autonomous radios outperform humans in spectrum allocation.
Originally posted by: Signal
A U.S. military-funded artificial intelligence (AI) contest that wraps up later this year may result in radio devices capable of autonomously and collaboratively sharing radio frequency spectrum for the next generation of mobile devices.
Fifth-generation (5G) cellular services are widely expected to hail a new era of greater speed, reduced latency and the ability to connect many more devices—think smart cities and the Internet of Things—and move vastly more data. The wireless revolution is fueling a voracious global demand for access to the radio frequency spectrum, but managing that increasing demand in a way that avoids interference is a challenge.
Researchers with the Defense Advanced Research Project Agency (DARPA) are sponsoring a contest to develop radio devices capable of deciding how best to share scarce spectrum. The Spectrum Collaboration Challenge (SC2) is the world’s first collaborative machine intelligence competition to tackle the spectrum sharing issue. The program will end later this year with a competition between the remaining teams from the United States, Europe and Asia vying for a $2 million prize.
Paul Tilghman, the SC2 program manager within DARPA’s Microsystems Technology Office, discussed the program at the agency’s AI Colloquium in Alexandria, Virginia.
“Today, humans around the world, policy makers, decide on human timescales how to take this very precious resource, the wireless spectrum, divvy it up into little slices, and assign those individual slices to certain technologies and specific operators,” Tilghman said. “This has served us well for the better part of a century, but the real challenge here is the exponential growth in demand for wireless spectrum. Increasingly, we like to think of ourselves as a wireless society. It’s a connected society, but that connection happens in the wireless world.”
At current growth rates, experts believe wireless networks could be transmitting zettabytes of data by 2030, Tilghman noted. A zettabyte is one sextillion bytes. A sextillion is represented by a one followed by 21 zeroes. “This [increased demand] really challenges this brittle, static ecosystem that we have today,” Tilghman said.
The program manager indicated that DARPA’s smart radio approach could prove “foundational” in the approaching 5G era. “A key tenant that’s on the 5G roadmap is this idea of spectrum sharing. One of the challenges, though, is that there’s no defined path or structure,” Tilghman explained. “We’re looking at SC2 as one of the technologies that really perform quite well in helping different radio networks use the spectrum simultaneously.”
In a demonstration held in December, the AI-enabled radios outperformed human spectrum managers by 50 percent, but there is plenty of room for the machines to improve. “We have another 150 percent to go in order to fully utilize the spectrum, so these radios are on the right track. Closing this gap and actually getting every opportunity to use the spectrum is really our final goal. Moving into our championship event that’s what we hope to see,” he said.
Because of the burgeoning demand for spectrum, human approaches and timescales simply are not efficient enough. “If we want to eek all of the utility out of the spectrum, we don’t want to make decisions about how it’s used on a timescale of hours or days. We want to make the decisions in seconds and milliseconds,” Tilgman asserted. “We want to make machine-speed decisions on how we how we use the wireless spectrum.”
The best way to do that, he added, is to hand the responsibility over to the devices that need to share the spectrum. “We need to actually take that decision making and push it out to the radio networks that are trying to use the spectrum. They are the ones that best know how it needs to be used second by second,” Tilghman said.
During the question-and-answer session following his presentation, Tilghman said he cannot estimate how long it will take for the military to field the smart radio technology, but he expressed confidence the systems will perform better at the final event in October. “I don’t know how long it will take before we find this in a military application, but I say firmly by the end of this year, at our October event, I do believe we’ll see that paradigm shift. We’ll see that breakthrough and realize there is an opportunity for autonomy and collaboration to play a role in spectrum.”
He also reported that the radios performed well in December when the DARPA team surprised competitors with intentional jamming during the demonstration. “What was really encouraging was that we actually saw our radio networks work autonomously, not to individually avoid the jammer, but to avoid the jammer without violating any spectrum-sharing strategies they had devised. I think we’re on the right path there with how to address not just collaborative actors, but also malicious actors,” Tilghman stated.