DARPA's Spectrum Collaboration Challenge: finally some progress towards a "Cognitive Radio" future

For 17 years, I've been writing about the possibilities of "cognitive radio", in which radios sense which spectrum is available from moment to moment and collaborate to frequency-hop (and perform other tricks) to maximize the efficiency of wireless communications.

It's hard to overstate how revolutionary this would be; today, most radio communication takes place through dedicated spectrum allocations (for example, a radio station will have exclusive rights to a given band in a given territory) that prohibit others from using the spectrum, even if it's not in use at a given moment.

With cognitive radio, spectrum becomes a commons that everyone shares, with computation, software-defined radios, and phased-array antennas subbing in for the blunt instrument of exclusive spectrum allocation.

Writing in IEEE Spectrum, DARPA's Paul Tilghman reports on the Agency's Spectrum Collaboration Challenge, in which teams competed to design algorithms that found efficient models for collaboration with one another. The Challenge has a $4m prize for the winning team, and the championship will be held in LA in October.

The Challenge runs in Colosseum, a simulated environment hosted on a supercomputing cluster at the Johns Hopkins University Applied Physics Laboratory.

As Tilghman writes, the approaches taken by the teams in previous rounds are a kind of recapitulation of the history of AI, starting with first-wave-style expert systems, then evolving into second-wave-style Big Data/machine learning approaches.

I'm not thrilled about this stuff being associated with the DoD, but it's very exciting to see progress towards delivering on the long-anticipated promise of cognitive radio. Read the rest