Darpa Grand Challenge Finale Reveals Which AI-Managed Radio System Shares Spectrum Best
Erebus (Erebus Solutions Inc.)-A three-person company created specifically to tackle SC2.
MarmotE (Vanderbilt University, University of Szeged, Budapest University of Technology and Economics)-Team leveraging their radio expertise to create AI "advisors" that shift their radio systems' strategies on the fly.
SCATTER (IMEC-IDLAB, Ghent University, University of Antwerp, Rutgers University)-Team tweaks 4G and 5G waveforms to slip their signals into any unused spectrum and squeeze their data through.
Sprite (Northeastern University)-Treats the competition's challenges as games of "spectrum Tetris" where the best strategy changes based on what's just happened.
GatorWings (University of Florida)-Using a technique called "foundational reinforcement learning AI" to optimize usage of the spectrum its systems can bite into.
Sodium-24 (Independent)-A one-man team with no prior wireless experience who's made it to the final ten teams.
Dragon Radio (Drexel University)-Operating under the motto of "mostly brawn and a modicum of brains," the Drexel team is combining expertise from all areas of the software defined radio world.
Zylinium (Zylinium)-This team's approach combines interference-resistant waveforms and AI to adapt to messy spectrum situations.
Andersons (Independent)-The winners of the previous 2014 DARPA Spectrum Challenge treat their radios as self-driving cars navigating congested spectrum traffic.
How Make Radio (Agitator LLC)-Two independent radio hackers have embraced "designing for failure" by allowing each of their radios to optimize its own performance.
It all started when Grant Imahara, electrical engineer and former host of MythBusters, walked onto the stage. A celebrity guest appearance is quite a way to kick off a U.S. Defense Advanced Research Projects Agency (DARPA) event, right?
The event in question was the finale of DARPA's Spectrum Collaboration Challenge, or SC2. SC2 has been one in the long list of DARPA's grand challenges-starting with 2004's eponymous Grand Challenge for self-driving cars. Over the duration of the three-year SC2 competition, teams from around the world have worked to prove that it is possible to create AI-managed radio systems that can manage wireless spectrum better than traditional allocation strategies.
DARPA's motivation in running the challenge-and the hope of Paul Tilghman, the program manager who's overseen it-is to develop AI-managed radio systems to help avoid a looming spectrum crunch. Wireless spectrum is a finite and exceedingly valuable resource. And an ever-growing number of applications and industries that want a slice of spectrum (with 5G being just the latest newcomer in a long line of technologies) means there soon won't be any available spectrum left. That is, if the wireless ecosystem sticks to its century-old method of pre-allocating spectrum for different uses in rigid bands.
At stake in the finale (beyond bragging rights, of course): nearly US $4 million in prize money. The winner: the team whose AI-managed radio system collaborated best when matched up with the diverse lineup of systems other teams had built and brought to the finale. The winning team walked away with a $2 million first place prize, while second and third left with $1 million and $750,000, respectively.
The finale was presented live by Imahara, Tilghman, and Ben Hilburn, president of the GNU Radio Foundation. There to witness it were attendees of MWC Los Angeles, a mobile industry convention. Though it was the culmination of three years of competition, the finale allowed the teams' radio systems to duke it out in real time for the first time. The head-to-head matchup was made possible by DARPA's efforts in relocating Colosseum, the massive radio frequency emulator at the heart of the competition, to the Los Angeles Convention Center floor just before the event.
In the end, GatorWings, the team representing the University of Florida's Electrical and Computer Engineering Department, walked away with the first-place prize. MarmotE came in second, and rounding out the podium was Zylinium.
The format of the finale hewed closely to that of last December's Preliminary Event 2, but with a few twists. All ten finalists competed in the first set of round-robin match-ups in a scenario called "Alleys of Austin." Each of the radio systems played the role of a squad of soldiers in a concentrated effort to sweep an urban neighborhood; by the end of the round, the lowest performing team would be eliminated.
The first five rounds of match-ups, run and analyzed while Colosseum was still in its home at Johns Hopkins University Applied Physics Lab, weeded out teams as DARPA tested them on different qualities necessary for AI-managed radio systems to succeed in the real world. In each match-up, teams were awarded points based on how many radio connections their systems could make during the scenario. The catch, however, is that if a competitor stifled one of the other teams and hogged too much of the spectrum, everyone would be penalized and no one would earn points. Awarding points based on collaborative performance avoided what Tilghman calls "spectrum anarchy," where teams would incentivize their own spectrum usage at the expense of everyone else.
The second round event, called "Off the Cuff," saw the systems learning to negotiate ad-hoc communications as though the teams were disaster relief agencies setting up operations just after a natural disaster.
"Wildfire," the third-round challenge, measured the teams' ability to prioritize traffic, by placing them in a wildfire response scenario where radio communications with the important air tankers had to take precedence over everything else.
"Slice of Life," the fourth-round scenario, imagined the remaining teams in a shopping mall setting, where each team's radio acted as a Wi-Fi hotspot for a different shop or restaurant-the idea being that they'd have to collaborate as different stores experienced a surge in traffic at different points during the day. (A coffee shop, for example, would require more spectrum in the morning, and a bar would need more at night).
The last round for the live finals was called "Trash Compactor." That setup forced the radio systems to share an increasingly narrow slice of spectrum. Over the course of the match, the band narrowed from 20 megahertz to 10 MHz and again to only 5 MHz by the end.
Then, with Colosseum in place outside the portioned room in which MWC Los Angeles attendees had gathered to see the spectacle, the competitors did it again, and results were churned out in real time. The five remaining teams-MarmotE, Erebus, Zylinium, Andersons, and GatorWings-found themselves duking it out again in versions of the previous scenarios that came with added twists that Tilghman described as "fun surprises."
The amped-up "Slice of Life" scenario, for example, dropped the teams back into the shopping mall. But this time, there was a nearby legacy incumbent system, like a satellite base station. The incumbent had priority over a portion of the spectrum band, and would trip an alert if the collaborating teams drowned out its own communications. Suddenly, the teams were having to manage random surges in demand for spectrum via their hotspots while also not taking up too much and disrupting the legacy operator.
With a few exceptions, the radio systems showed they could roll with the punches SC2 threw at them. The fifth round of match-ups, "Trash Compactor," proved to be perhaps the most challenging, as teams seemed to struggle to share the spectrum effectively as the available amount got squeezed. And yet, situations like working alongside an incumbent operator-something that stymied many of the teams last December-seemed much easier this time around.
The teams' systems also established radio connections more quickly, and made more effective use of the available spectrum to complete more connections than they had less than a year ago. In one of the "Slice of Life" match-ups, teams collaborated to achieve almost 340 percent spectrum occupancy. In essence, their radio systems crammed almost 70 MHz of traditional spectrum applications into 20 MHz of bandwidth.
It's hard not to be impressed by that level of spectral efficiency. But after the awards were handed out and the photographs taken, Tilghman stressed the fact that the technology has a long way to go to be fully developed. Still, he thinks it will be developed. "I think the paradigm of spectrum collaboration is here to stay," Tilghman said to conclude the event, "and it will propel us from spectrum scarcity to spectrum abundance."