The other day I found this short item in the ACM Tech News:
DARPA Pop-Up Testbed Takes On Spectrum Management
Government Computer News
George Leopold
April 7, 2017The U.S. Defense Advanced Research Projects Agency (DARPA) sponsored a recent event to test new modulation recognition strategies for navigating the radio frequency spectrum. The desired objectives included demonstrating new modulation recognition methods for identifying signal origins and types, and finding better techniques for sending and receiving information over the least trafficked spectral bands. “Modulation recognition is that first step towards getting beyond just describing ‘presence’ or ‘absence’ [and] actually describing what is present,” says Paul Tilghman with DARPA’s Microsystems Technology Office. The event employed about 30 modulation schemes, with hand-coded expert systems matched against newer machine-learning platforms. The former did better in identifying signal characteristics, but DARPA program manager Tom Rondeau expects the machine-learning method to catch up. DARPA says modulation recognition is vital for achieving “wireless situational awareness” that could help extract more capacity from congested electromagnetic spectrum, and then predict spectrum use and increase throughput.
After reading this, I said to myself, “How cool is this? what a great new use for artificial intelligence.” When I did a little Googling on the topic, though, I found out that modulation recognition has been around for at least 20 years. Now, I’m saying to myself, “How did I miss this?”
I’m not an AI expert, but it seems to me that whoever can crack this nut will be years ahead of the competition. Instead of setting up your radio for CW, SSB, or one of the digital, just let the radio do it. Instead of you or me extracting the information from a noisy signal, we’ll let the computer do it instead. And you know what, chances are the radio will do a lot better job than we can. The future’s going to look a lot more different than we can imagine.