Some Fun with Claude

I've had some downtime, recently, courtesy of a couple nasty health issues (no, not cancer).  In order to relieve some of the boredom while recovering, I started sorting through some of the electronic "stuff" I've accumulated for projects which, in turn, never seemed to get done.  I can across a ESP32-S2 feather - the one with the built-in TFT display - and thought it might be fun to turn it into a cw decoder.  (Yes I know such things are no substitute for the ability to copy by ear, but remember, I was bored.)  I'm also not a big fan of "vibe coding" but for this exercise I though it'd be interesting to see what Claude could do.  Some I told it:
 
1. Target device is Adafruit ESP32-S2 TFT Feather board.
2. Will be using the current Arduino IDE for editing, compilation and uploading.
3. Input will be raw audio on analog input A5.
   a. Audio input will be line level (2.00 volts - high impedance).
4. Decoded text will be displayed (in real time) on the integral tft display.
5. Incoming signal will not be perfect
   a. program must deal with "hiss" (high frequency noise) that
      accompanies the desired audio.  May need some sort of low
      pass filtering.  Desired audio will be in the 300 Hz to
      1000 Hz range, typically.
   b. Character/word spacing may be erratic and vary. Program must
      deal with this.
   c. Sending speeds from 5 words/minute (wpm) to 50 wpm must be
      accommodated adaptively.  Would like to have a real-time wpm display.
   d. Signal level can change.  Program must provide some sort
      of AGC to deal with this. Fading could be rapid or slow. 
      Decoder must not lock up on very strong or very weak
      signals. 
   e. Do not use look-up table for decoding.  Suggest binary
      tree algorithm or other more modern approach.  Feel free
      to search on-line for current state of the art.
   f. Must be able to decode letters, numbers, and punctuation.
 
Well it took Clause about 5 minutes to give me a decoder sketch which didn't compile.  After not so politely asking Claude to fix the compile errors, I got a sketch that actually worked pretty well (or as least as well as any software decoder works).   One of the days I'll make a short clip of it decoding a real cw signal (not W1AW code practice, which it gets perfectly).  The sketch as produced by Claude, follows (I claim no credit):
 
You can find it here:   https://github.com/dave-metzler/AI-Morse-Decoder-.git

Comments

Post a Comment

Popular posts from this blog

A Half-baked Fix for Si5351 Quadrature Issues

Image
In a previous post, I briefly described issues I had preserving quadrature output from an Si5351 board.  Allegedly, once a 5351 is programmed to provide a fixed phase offset on a second output,  it should maintain that phase shift across frequency excursions, provided the "even divisor" remained unchanged - No PLL reset needed until the a large enough frequency excursion requires a change in the "even divisor".  PLL resets caused annoying "pops" in the output of direct conversion front ends that I typically use for phasing-type receivers.  I was loathe to include a PLL reset on every frequency change because of that.  So, I (and many others) only do a PLL reset when the required integer "even divisor" was changed.  Many fewer "pops", but semi-random loss of LO quadrature within a relatively small frequency range.  I added a polled front-panel switch to my hardware that I use to manually reset the  PLL when my "ear detector" de...
Read more

Playing with Phasing - part 1

Image
      Every so often it seems that some folks are lured into the less-populated territory of SSB reception and transmission using the phasing method.  Maybe it's the promise of simpler (?) circuitry, eliminating the need for crystal/mechanical filters, or perhaps as a gateway drug into SDR-land.   Rick Campbell's (KK7B) articles describing a single-signal direct conversion receiver (QST - Jan 1993), followed by a phasing transmitter (QST - April 1993) are seminal works that detail the implementation of the phasing method with modern (for the 1990's, at least) circuitry.  These papers also discuss design tradeoffs in terms that are easily understood.  Quite a few years ago, I actually built a R2/T2 transceiver, shamelessly copying those designs and adding a Si5351 quadrature LO along the lines of that described by ZL2CTM in his blog.  It worked OK, but the opamp-based, all-pass phase shifters required hand matching of the RC components fo...
Read more

Revisiting the LT5517

Image
Some time ago (just before Linear Tech & Analog Devices merged), I requested a few samples of LT's 5517 quadrature demodulator and 5598 quadrature modulator.   They happily supplied a few samples of each, but the local LT rep also called me at work and asked if I'd also like to have a dc678a evaluation board ( https://www.analog.com/en/resources/evaluation-hardware-and-software/evaluation-boards-kits/dc678a.html#eb-overview ).  Of course, I said, "Hell yes!" and it appeared in the mail soon after (no charge).  But then it sat unloved and unused for years until I recently decided to see how well it worked at HF. The 5517 board is the green one in the upper left corner.  A 2-channel baseband preamp is the Manhattan-style board next to it.  The board below the 5517 board is a split supply boost converter that powers a 100x gain, 2-channel baseband preamp using THAT1510 instrumentation amps.  These devices are very similar to the widely used INA217....
Read more