What can I do with AI? - Let's Build a digital mode for HAM Radio!

Alright, first things first, I am not any kind of programmer or code genius.  It was decades ago when I was heavily into writing any kind of code and today that only gives me the fundamentals of the concepts.  The languages have changed, the hardware has changed...everything.  This was more of an idea to see what I could do using AI to write a new mode and UI for a digital mode.

This is one iteration of the UI showing the actual signal in the waterfall from another computer and radio transmitting on a dummy load.  This is a simple test using two radios on dummy loads.  I have logging working, or at least as far as I can tell it is.   

 

I have focused this on POTA/SOTA/FD type operations, but obviously it could be used for everyday digital ragchewing.

The design goals of this mode are to create a mode that keeps the ragchew ability of PSK31, better speed than the OLIVIA modes, and improve weak signal and noise resistance over them both.  I am keeping this software as light as I can with the features that I feel are important for the activities mentioned.  The idea is one laptop, Raspberry Pi, or other portable compact computer of ones choice can run it in the field. All that with incorporated log book so you don't have to switch back and forth between a log program and the UI for the digital mode.  The key piece I have working so far is the log export that automatically exports an .adig that is properly formatted for upload to the POTA and SOTA sites.  Once the operator sets up the station information, the software recognizes the activity or activities and will generate the proper logs for them.  If you are doing a multi reference POTA, it will export a log formatted for each reference automatically.

 

The audio routing has worked flawless from the get go.  VAC from Thetis is working as it should and the USB Codec with the TS590SG is working great. 

 

 I have TCI rig control working with Thetis and my Hermes Lite 2+, the rigctld for my TS590SG still is not working. That is on the TO DO list.

 

One feature I have built in is an automatic report system.  The software generates the received signal report from an algorithm that analyzes the signal during decoding and gives a signal report based on that score. 

 

If and when this gets operational it should have:

-Strong FEC for improved weak signal ability.

-Strong resistance to noise.

-Built in logging

-Automatic POTA/SOTA formatted log exports

-Speed that exceeds OLIVIA

-Free text chat to overcome the useless nature of FT8's inability to send actual messages.(EmComm version may be coming after this is functional that should allow the most commonly used forms to be fillable and sent with a macro.) 

-A data rate that equals ~94wpm typing. (That signal burst on the waterfall was a full CQ in ~3.5 seconds)

  

As of 6/30/2026 I have not been able to get the decoder functioning properly.  The transmit side is generating the outgoing message great, but getting the demodulator, FFT, buffer, decode, FEC all working together to decode the message is the most complex part of the code.  The AI's keep wanting to focus on the signal isn't correct or not present, which I find amuzing since the waterfall clearly shows the tones spread across the waterfall in the proper passband.  It's been a learning experience with AI to see how it responds to prompts.  It is common for the AI to misinterpret the intent of a prompt.  I have had to learn to write prompts as I see how it construes the meaning of english language in ways I never would have thought.

More to come on this project!