In the May 2020 issue of IEEE Spectrum, there’s an intriguing article by four engineers who claim, “The era of communications systems designed solely by humans is coming to an end.” In the article, “In the Future, AIs—Not Humans—Will Design Our Wireless Signals,” they say, “The reason is simple: rapidly escalating complexity.” Of course, three of the four are associated with a company, DeepSig, that sell the AI tools to do this development, but I found this to be an interesting concept.
In the article, they describe their design approach and how they applied it to the signals sent via NASA’s Tracking Data Relay Satellite System (TDRSS). They claim that using their AI-designed signals, they were able to improve the bit error rate by 42% from 5% to just under 3%. They do admit, however, that in both tests, they ran the system without any error correction, which would have reduced the bit error rate of the system.
My question would be is the effort (and the cost of the software) worth it? The article doesn’t really give much detail, so it’s hard to say. Also, the authors note that this work was just a proof of concept, meaning that they do intend on making improvement in the future.
So, is AI going to take over ham radio, and if so, when? I’ll probably still be pounding brass myself in 20 years, but I certainly think AI is going to play a bigger part in amateur radio in the future. At least I hope so.
You gotta admit it’s interesting to consider the ramifications.. For example, will an AI get the Ham of the Year award at Dayton in 2040? And, when will hear hear the FT8 guys complain about AI being the death of ham radio?
Ed Jones K8MEJ says
Hi Dan. I think a more important use of AI in radio will be the use of electromagnetic spectrum. There is more and more demand for scarce spectrum and AI will be useful for finding available bandwidth at any given time for a particular use. There was a talk three or four years at the TAPR/DCC conference about “cognitive radio” or something like that. I would encourage you to go find that talk and listen to it. It was one of the most interesting TAPR presentations I have listened to.
Rick Barnich K8BMA says
Dan,
When I worked with the 8080 microprocessor in 1973, it was comprised of 2800 transistors and was laid out by hand. Today, Apples A12 processor encorporates 6.9 billion transistors and is designed by computers. In 2012, I worked with PGA software that could take C program code and convert it to gate array logic. No reason to believe communications should be any different.
Arne K5ARN says
It won’t be long before AI software will start coding software without human interaction and eventuality take over amateur radio digital modes. It will be called AI-mateur Radio and humans will not understand a thing, except 73 or 73s will always confuse AI computers!
Phil W1PJE says
In response to K8MEJ’s comment, yes, AI is used already in cognitive radio to find empty band slots. But doing this across all users of radio poses a tremendous danger to science using the spectrum for passive observations – e.g. radio astronomy, or Earth microwave remote sensing of water from orbit, etc. If it’s a passive application, it radiates nothing that can be detected as occupying the band – so AI has no way to avoid that spectrum slice for transmissions, other than by having pre-defined rules that say no one should transmit there. This is harder than you think – consider out-of-band emissions for example. How close can you let a TX get so it doesn’t leak into the protected band? There’s another rule that has to be hard-coded.
When you put it all together, the bookkeeping is so overwhelming (and the potential for interference so great) that it is a really hard and as yet unsolved problem. Bears watching for future developments.