Radio Wrestlers Fight It Out at the DARPA Spectrum Challenge

Clash of the software-defined radio algorithms leaves two winners


The words engineering and sports aren’t usually used in the same sentence, and the two activities usually don’t happen in the same room. But they were, and they did, last week at the DARPA Spectrum Challenge, held at the Defense Advanced Research Projects Agency headquarters in Arlington, Va.

The goal of the contest, the first ever DARPA challenge on the use of spectrum, was to demonstrate how a software-defined radio can use a given communication channel in the presence of other users and interfering signals. Over the course of two days, normally taciturn techies oohed, aahed, and cheered as 18 teams competed in a series of head-to-head matches to see who had the best algorithms.

A nearly yearlong process winnowed an initial field of 90 teams down to the 18 that competed last month. While most of the teams were composed of academics, there were also amateur hobbyists, including a medical doctor whose second love is electronics and a 28-year-old contract programmer with no formal education beyond high school.

The challenge was actually two separate competitions. In one tournament, two teams in each match competed to send and receive 15 000 packets on a 5-megahertz-wide channel centered at 600 MHz within the UHF band. This was an adversarial contest, where one team was allowed, even encouraged, to try to interfere with the other team’s broadcast. This type of adaptive radio technology is appropriate for military applications—on a battlefield, for example—where the enemy is trying to jam transmissions. In this contest, the first team to successfully transmit and receive all their packets won and moved on to the next round. A team called Efficient Spectrum—made up of two engineers employed at a large defense contractor but competing as individuals—won this tournament and its US $50 000 grand prize. Team GTRI, a group of research faculty members from the Georgia Tech Research Institute, came in second place and won $25 000.

In the other tournament, three teams per match were judged on how well their algorithms shared spectrum. This type of technology is more germane to commercial applications in unlicensed bands, where there is no prior coordination or even definitive rules. In particular, this is how companies hope to use the unlicensed TV white spaces freed up by the Federal Communications Commission. In these matches, the team that transmitted the most packets won, but it also got the number of packets transmitted by the second-highest scorer added to its total, which helped it in later rounds. This was intended to encourage cooperation, because a win depended not only on how many packets a team successfully sent and received but also on how successful the other teams were. A team made up of an assistant professor and an undergraduate from Tennessee Technological University won this competition, receiving $50 000, with Efficient Spectrum placing second for a $25 000 prize.

The running of the algorithms and broadcasting of the signals was performed remotely, at the Open-Access Research Testbed for Next-Generation Wireless Networks, known as Orbit, a radio research lab at Rutgers University, in Piscataway, N.J. That ensured that the algorithms were run independently, using the same hardware, so the software could be judged clearly on its own merits.

For each match, the Orbit staff programmed the system with the teams’ algorithms, and the race was on. A display of each team’s signals was live-streamed to the audience at DARPA, which lent the competition a spectator-sport atmosphere. At the top of the display were spectrograms showing the frequencies over which each team’s source and destination nodes were transmitting. In real time, the teams and audience watched as one team or another pulled ahead or, in a couple of instances, didn’t even make it out of the gate.

Yiftach Eisenberg, program manager of DARPA’s Information Innovation Office and organizer of the challenge, provided a play-by-play for each match, commenting on how each team was doing, noting when a team was pulling ahead, and explaining what some of the waveforms indicated. In the adversarial tournament, some teams actively tried to interfere with the others’ signals, but that approach wasn’t as prevalent as it was in the preliminary competition, held last September, when the propensity for interference was so high that sometimes neither team was able to transmit or receive anything. As a consequence, DARPA made some adjustments to reduce the interference levels, says Eisenberg.

Eisenberg notes that many of the teams had dramatically improved their designs since September. “I’m amazed at how much innovation took place between the preliminary tournament and this one,” he says. Specifically, in September teams used designs without feedback loops, spectrum sensing, or adaptation. “Today, what we saw was that many of the teams made this huge leap to radios that can sense and adapt their transmissions in real time.”

The final competition proved “that spectrum sharing among peers is feasible. We can take different systems in a complex environment, and they can share the spectrum without human intervention and spectrum preplanning,” says Eisenberg.

Among the most valuable lessons of the competition were the things that didn’t work. “We saw the unintended consequences and unexpected behaviors when you put these adaptive systems together,” he says. In one match, for example, one team started its transmission by displaying its logo in the spectrogram, which somehow prevented its competitor from registering its program and starting transmission. That was a surprise to both teams and to Eisenberg. “These are the kinds of things none of us could have ever predicted or tested for,” he says. “This is what happens in real life, when dynamic complex systems are thrown into dynamic complex environments.”

Video: DARPA
Radio Race: Results were streamed live to DARPA from the Orbit lab test site during the DARPA Spectrum Challenge. Each team’s signal appears in a different color. The top part of the picture shows spectrograms as the signals are sent. On the bottom of the screen are waterfall plots that show how many packets have been broadcast by each team’s source radios and how many have been received by the team’s destination radios. (Each square in the plot represents 1000 packets.)

About the Author

Tam Harbert is a freelance journalist based in Washington, D.C., who specializes in technology and business. Her November 2013 IEEE Spectrum feature on artificial intelligence for patent law won an American Society of Journalists and Authors Periodicals writing award.