Shower thought: XARA had the merit to rise the debate on the need for a decent alternative to VARA. While that modem has stood out because of being a decent contender of the latest PACTOR revisions, and beyond the fact it was proprietary and required a license, mandating a graphical session for a modem (and dependency on Wine) was a steep constraint, especially if you wanted to operate a dirt-cheap gateway (hardware wise).

In the end, the discourse has made me noticing github.com/arodland/Data2G, which, while being young, looks quite promising, especially in the weak signal scenarios, thanks to Selective rejects, instead of plain ARQ, and to a few elegant refinements.

f4ivy reshared this.

in reply to isithran

"When it comes to DSP and error-correcting codes, I know the principles, but I'm not an expert on the details, especially the details of doing them efficiently. I set the goals and the direction, Claude turned the spec into tests and the tests into code, and created something that would have taken me years otherwise."

Writing magic software that you don't even know how it does what it does, and paying for that privilege until you get bored of the project sounds sustainable to me.

in reply to Scott VO1BAY 🚩

Amateur radio needs more people that can write complex signal processing code, totally agree with you. And there are lots of standards and approaches that are well documented to build off of.

But short-circuiting the toil required to create understandable, documented, _teachable_ implementations, even if they lack some of the whiz-bang features, is damaging to continuity in the digital data domain for amateur radio.

This entry was edited (today, 1:49 PM)
in reply to Scott VO1BAY 🚩

In that regard, I really like the Direwolf source code - it's not just a Bell 103 modem doing APRS and AX.25 out the digital side - it's well written code, and _extensively_ documented. There are parts of the code where the comments read like a small article or mini-essay on the topic at hand.

It's so good that _I_ am able to learn and start to understand from it, and what I don't understand, it's at least signposts to where to go to learn more.

Unknown parent

mastodon - Link to source

Scott VO1BAY 🚩

@n6cta I don't feel like I'm making any assumptions at all - the LLM generated RF modes that are not understood are substantially almost literally carving out the bootstrap loop.

I'm in no way suggesting that coding be a purity challenge, and that if you can't write it from first principles you shouldn't write it at all - that would exclude basically everyone.

But if the community can only bootstrap via LLM, vs a pile of research papers, the community is captured.

in reply to Scott VO1BAY 🚩

@n6cta And given the choice between reverse engineering a stack of LLM code and adding comments and documentation post to untangle whatever its generated, vs sitting down and doing some planning, design, and a forward explanation of what you are about to write in code, I know which I'd prefer to spend my time on, either alone or in community working on building something.
in reply to Scott VO1BAY 🚩

@n6cta "a future that hasn't been written" is fair, but it cuts both ways. The initial indications are not looking good, and possibly worse in the amateur radio space, than I'd have expected.

Self-reflecting, just even the AI summaries on top of Google searches are an example of how terribly slippery the slope is right now. Going from glancing at them, to "just once", to reliance is exceptionally easy (and is being done by design, to be sure. It's barely anyone's fault that they use)

⇧