FreeDV Sunday net - a new record

A very well attended Sunday FreeDV net on 7045 this morning from 10:30 eastern time. Stations on frequency were: VK2AWJ, VK2BX, VK2UMZ, VK3BAL, VK3CDH, VK3DJB, VK3FC, VK3GTP, VK3ID, VK3KEZ, VK3PCC, VK3PTR, VK3SRC, VK5ABE, VK5AG, VK5KHZ, VK5KVA, VK5MH, VK7DMH, VK2KO, VK2KNC, VK2YW, VK2CJB and me VK3TPM. 24 stations!

 

Generating sideband to transmit on a HackRF

While working on my C port of FreeDV RADE1 I wanted a reliable radio signal to test with. While I have off air recordings from the FreeDV GUI app I thought it would be nice to generate my own files to transmit with the HackRF Portapack Mayhem which can record to C16 files and play them back. There's a TXT file with each recording that specifies the sample rate and output frequency.

I've made a python utility that can read a WAV file recording and produce C16 sideband. Here's how it sounds.

The code is pretty short so I've published it as a gist on GitHub.

The HackRF is a wonderful tool for radio experimentation. I'm using the standalone Mayhem but it works just as well from a computer via USB with hackrf_transfer.

FreeDV Sunday net - 23 stations

An excellent FreeDV RADE1 net here in South East Australia with 23 stations on frequency. Some for the first time. Stations: VK5KHZ, VK5KFG, VK5AG, VK3ZUM, VK5KVA, VK3UBK, VK3SRC, VK3KEZ, VK3JCO, VK3GTP, VK3FC, VK3DJB, VK3CDH, VK3BQO, VK3BAL, VK2YW, VK2XOR, VK2UMZ, VK2KNC, VK2DWG, VK2AWJ, VK5HM and me VK3TPM. 

Normally I call in stations in alphabetical order but today I thought I'd trick Jack VK5KVA, by going in reverse.

 

C port of FreeDV RADE going well - transmit working today

Today marked a milestone in my investigation into using Claude Code with a complex codebase. I had a FreeDV RADE contact with another station using code it wrote under my instruction in just one week - very much part time. Here's a receive demo:

My thanks as always to Joe, VK3SRC, for assistance in testing and feedback. Also thanks to Britt, VK3AOB who responded to my first call which I really expected to not be decodable.

Here's how Claude Code looks while working on a bug in the code (click to enlarge):

 I'm developing on Linux Mint. The app is C++ using GTK and Claude seems very knowledgable in this scenario. I suspect that the large amount of open source software around is a great advantage for this sort of thing.

The code is here: https://github.com/peterbmarks/radae_decoder As you can see the name of the project is no longer correct as it encodes as well.

Tech talk on ABC Radio - Is OpenClaw the future?

Have you heard of a tool called OpenClaw? Apparently, you install it on your computer and it acts like your personal assistant. it's generating a lot of buzz and users are saying it's like they're 'finally getting a glimpse of the future'. 

Tech Talk takes a look at this and how the big four tech companies are faring after they've spent big on AI. 

Peter Marks, mobile software developer and technology commentator from Access Informatics, joined Philip Clark and listeners to Nightlife with the latest tech news. Listen here: https://www.abc.net.au/listen/programs/nightlife/nightlife-tech-talk-with-peter-marks/106328266 

FreeDV Sunday net - excellent conditions

An excellent Sunday FreeDV net this morning with VK2, VK3 and VK5 stations all with good SNRs. 

Some interest was expressed in the C port of the RADAE python code which is encouraging. On the frequency, (not everyone transmitted), VK2AMF, VK2AWJ, VK2DWG, VK2GMH, VK2YE, VK3DJB, VK3FC, VK3GTP, VK3JCO, VK3KEZ, VK3XCI, VK3YV, VK3ZUM, VK4TEW, VK5ABE, VK5KHZ, VK5KVA, VK5LN, VK5ST, VK6YR, VK3BAL (running three copies of FreeDV!), VK3AOB, and me VK3TPM.

Experimental C port of the FreeDV RADEV1

One of the objectives of the work on RADEV2 is to replace the currently required Python runtime from the distributed FreeDV app. This would make it easier to install, smaller, and hopefully more efficient. (Other objectives are improvements in how it works). Python is an important part of the project and is used for prototyping and testing - that won't change.

Re-writing all the python in C is a daunting, tedious, expert task, estimated to take many months that would result in an app that works pretty much the same as it does today.

I've been experimenting with Anthropic's Claude Code recently and it occurred to me that one of the things LLMs are particularly good at is translating between languages, human or programming. The result of just a few days of guiding Claude Code is an experimental RADEV1 decoder app.

The intial work, which includes command line tools to decode and encode from and to WAV files is on github here.

The proof of concept Linux application shown above is in a Github repo here. 

While it seems to work and does pass important tests, this is not code supported by the project at this time.

Claude Code enabled me to attempt something I wouldn't dream of until now. It was quite a big job for it (I hit my quota on two separate days and had to wait to resume).

Working with an LLM to write software does still require enough knowledge to know what to ask for and to guide the work.

I hope this work will encourage others to develop innovative FreeDV RADE software along side the official FreeDV App which remains the reference release. My sincere thanks to Dr David Rowe for his fantastic work and his help and encouragement with this little experiment.

A well used laptop keyboard

Many years ago I handed down a MacBook air to our youngest daughter. She used it until I handed down a later machine. Last week she returned it and I'm amazed to see how heavily she had used it over the years without complaint.

It still works although the battery is dead now. 

Some people complain that Apple is expensive but I think it's value for money.

This kind of keyboard wear reminds me of an HP25C calculator I had many years ago. It was also a piece of quality engineering and, as I recall, one feature was that the symbol on the buttons was plastic that extended right through the key so it would never wear off. 

Xiegu X6200 review

Keen to do more field operating (and overnight camping). My rig of choice has been the Elecraft KX3 with it's excellent antenna tuner and easy to read LCD display. Since using rigs with a waterfall display of the band I find it hard to going back to tuning up and down trying to find stations. 

I purchased a Xiegu X6200 from AliExpress for AU$1200. 

It has some terrific features for portable operation:

• Decent flat battery pack that clips on the back
• Wonderful sharp and bright display
• Excellent built-in antenna tuner
• USB-C socket with interfaces for CAT control and audio
• Handles to protect the knobs

The front facing speaker is small but quite effective. Surprisingly it is NOT a touch screen (although if you connect a mouse you get a mouse pointer). This means that to go through menus you must spin the outer lower left knob to move between settings and spin the inner knob to change a setting.

A waterfall with a touch screen is a wonderful thing and on a 7300 being able to tap a signal to tune there is missing here. Tuning with the knob is a little laggy and seems to have some inertia for some reason. (Unfortunately clicking on the waterfall with a mouse isn't supported).

The rig is certainly a quality build. All the controls feel excellent and the battery pack, sadly proprietary, clips on to the back. It's remarkably compact and solid feeling. The microphone has a full set of buttons including three configurable buttons for things like NR or NB. You can also direct enter frequencies. There is also a built-in microphone and PTT button so you could use it like a hand-held.

Because of my focus on digital modes, I wouldn't buy a radio without a direct USB connection and this radio does it all. I have found that it's rather sensitive to the cable being used and I've experienced the serial devices not appearing even though I was using a USB-C to USB-C cable that clearly carries data as it works with hard disks. The supplied USB-C to USB-A cable.

When plugged into a computer two serial devices appear. The second one is the CAT port (the other is a serial terminal).

Here's the settings that work with fldigi:

Here's the settings that work with FreeDV:

Rig: Xiegu X6200

Serial Port: /dev/ttyACM1

Baud Rate: 19200

Data bits, stop bits, handshake: Default

PTT Method: CAT

Mode: Data/Pkt

Split Operation: None.

Audio Input: alsa_input.usb-C-Media_Electronics_inc.USB_Audio_Device-00.mono-fallback

Audio Output: also_output.usb-C-Media_Electronics_inc._USB_Audio_Device-00.analog-stereo

The radio should switch to mode: U-DIG or L-DIG depending on band with FreeDV

I note that the audio level to WSJT-X is too high, 88dB. I needed to reduce "mic" gain in Linux sound settings.

Bugs

Xiegu has a history of software bugs in their radios. It seems some never get fixed. The most disappointing one for me is that I'm on the latest firmware 1.0.7 and I cannot get it to connect to Wifi which is a pity as there's a WFView server (Icom radio networking server) built in and it would be great to remote access the radio.

There is a discussion group at https://xiegu-x6200.groups.io/g/main/topics that looks to be a great resource.

Conclusion

I think this is quality radio for the money. I joined our club 80m net this week and got good reports from other stations. The radio puts out about 4W on the attached battery and about 9W with external power.

Received a rather lovely QSL Card

This just in from an FT8 contact. Very retro.

Thanks Jim.

Very good morse decoder from 101 things

Prolific ham radio hacker Jonathan P Dawson has just shared a remarkable project he calls Hamfist that is a morse decode device running in the Arduino runtime for Pi PICO processors (including the original). 

Go and read his description but it's quite a sophisticated decoder, rather like a CW skimmer it can decode several morse signals in the audio passband. The PICO does the audio sampling with one of its A/D pins with just a simple circuit on the input.

"It combines adaptive signal processing, automatic speed estimation, probabilistic decoding, dictionary-based correction, and multi-channel decoding, all while keeping memory and CPU usage firmly under control."

I grabbed the source and built it in the Arduino IDE targeting the PICO. All very smooth. Unfortunately I don't have the display he uses, the ili9341, so I've ordered some. The code seems to run but of course I can't see the morse.

Update: A display board arrived.

So. I pointed Claude Code at the Arduino files and simply asked it to create a C++ command line program that could read a WAV file with morse audio in it and print out the decoded morse. It wrote wav_decoder.cpp and a Makefile. Jonathan includes some sample WAV files so I tried some of these. Here's how that goes:

cw_decoder % ./cw_wav_decoder examples/cw2.wav

WAV file: examples/cw2.wav

  Sample rate: 12000 Hz

  Channels: 1

  Bits per sample: 16

  Target sample rate: 15000 Hz

CQ SP3VT SP3VT TEST E4EQ SP3VE E3VT TEST PA3A IN PA3AT 5NN E 27 MIKEEE EAT/ HI TU S E3VT CQ SP3VT SE 3VT TEST CZ G3VT SPSE UT TEST CQ SW3VT SP EVE TEST CQ SP3VT SAME3VT TEST DK3T DA3T 5NT 

I would add that the decode all happens in a fraction of a second.

I'm not sure if I'm more impressed with Jonathan's contribution or Claude Code's capabilities. Here's the file it produced and the Makefile.

#include <cstdio>

#include <cstdlib>

#include <cstdint>

#include <cstring>

#include <string>

#include "cw_dsp.h"

#include "fft.h"

// WAV file header structure

struct WavHeader {

char riff[4]; // "RIFF"

uint32_t file_size; // File size - 8

char wave[4]; // "WAVE"

char fmt[4]; // "fmt "

uint32_t fmt_size; // Format chunk size

uint16_t audio_format; // 1 = PCM

uint16_t num_channels; // Number of channels

uint32_t sample_rate; // Sample rate

uint32_t byte_rate; // Bytes per second

uint16_t block_align; // Bytes per sample * channels

uint16_t bits_per_sample; // Bits per sample

};

// Derived DSP class that outputs decoded text to stdout

class c_wav_decoder : public c_cw_dsp {

private:

std::string last_text[NUM_CHANNELS];

protected:

void decode(uint16_t channel, std::string text, std::string partial) override {

if (!text.empty() && text != last_text[channel]) {

printf("%s", text.c_str());

fflush(stdout);

last_text[channel] = text;

}

}

public:

void print_final() {

// Flush any remaining text

flush();

printf("\n");

}

};

// Simple linear interpolation resampler

class Resampler {

private:

double ratio;

double position;

int16_t last_sample;

bool first_sample;

public:

Resampler(uint32_t input_rate, uint32_t output_rate)

: ratio((double)input_rate / output_rate)

, position(0.0)

, last_sample(0)

, first_sample(true) {}

// Process one input sample, may produce 0 or more output samples

// Returns number of output samples produced

int process(int16_t input, int16_t* output, int max_output) {

int count = 0;

if (first_sample) {

last_sample = input;

first_sample = false;

}

while (position < 1.0 && count < max_output) {

// Linear interpolation

double frac = position;

int32_t interpolated = (int32_t)((1.0 - frac) * last_sample + frac * input);

output[count++] = (int16_t)interpolated;

position += ratio;

}

position -= 1.0;

last_sample = input;

return count;

}

};

void print_usage(const char* program_name) {

fprintf(stderr, "Usage: %s <wav_file> [channel]\n", program_name);

fprintf(stderr, "\n");

fprintf(stderr, "Decodes Morse code (CW) from a WAV audio file.\n");

fprintf(stderr, "\n");

fprintf(stderr, "Arguments:\n");

fprintf(stderr, " wav_file Path to the input WAV file (mono or stereo, any sample rate)\n");

fprintf(stderr, " channel Optional: frequency channel 0-%d (default: all channels)\n", NUM_CHANNELS - 1);

fprintf(stderr, "\n");

fprintf(stderr, "The decoder uses %d frequency channels spanning 0-%.0f Hz.\n",

NUM_CHANNELS, NUM_CHANNELS * CHANNEL_SIZE * (SAMPLE_FREQUENCY / 2.0) / (FRAME_SIZE / 2));

}

int main(int argc, char* argv[]) {

if (argc < 2 || argc > 3) {

print_usage(argv[0]);

return 1;

}

const char* filename = argv[1];

int selected_channel = -1; // -1 means all channels

if (argc == 3) {

selected_channel = atoi(argv[2]);

if (selected_channel < 0 || selected_channel >= NUM_CHANNELS) {

fprintf(stderr, "Error: channel must be between 0 and %d\n", NUM_CHANNELS - 1);

return 1;

}

}

// Open the WAV file

FILE* file = fopen(filename, "rb");

if (!file) {

fprintf(stderr, "Error: Cannot open file '%s'\n", filename);

return 1;

}

// Read WAV header

WavHeader header;

if (fread(&header, sizeof(WavHeader), 1, file) != 1) {

fprintf(stderr, "Error: Cannot read WAV header\n");

fclose(file);

return 1;

}

// Validate WAV format

if (strncmp(header.riff, "RIFF", 4) != 0 || strncmp(header.wave, "WAVE", 4) != 0) {

fprintf(stderr, "Error: Not a valid WAV file\n");

fclose(file);

return 1;

}

if (header.audio_format != 1) {

fprintf(stderr, "Error: Only PCM WAV files are supported (format=%d)\n", header.audio_format);

fclose(file);

return 1;

}

if (header.bits_per_sample != 16 && header.bits_per_sample != 8) {

fprintf(stderr, "Error: Only 8-bit or 16-bit WAV files are supported\n");

fclose(file);

return 1;

}

fprintf(stderr, "WAV file: %s\n", filename);

fprintf(stderr, " Sample rate: %u Hz\n", header.sample_rate);

fprintf(stderr, " Channels: %u\n", header.num_channels);

fprintf(stderr, " Bits per sample: %u\n", header.bits_per_sample);

fprintf(stderr, " Target sample rate: %.0f Hz\n", SAMPLE_FREQUENCY);

fprintf(stderr, "\n");

// Skip to data chunk

// The fmt chunk might be larger than our struct, and there might be other chunks

fseek(file, 12, SEEK_SET); // Skip RIFF header

char chunk_id[4];

uint32_t chunk_size;

while (fread(chunk_id, 4, 1, file) == 1) {

if (fread(&chunk_size, 4, 1, file) != 1) {

fprintf(stderr, "Error: Malformed WAV file\n");

fclose(file);

return 1;

}

if (strncmp(chunk_id, "data", 4) == 0) {

break; // Found data chunk

}

// Skip this chunk

fseek(file, chunk_size, SEEK_CUR);

}

if (strncmp(chunk_id, "data", 4) != 0) {

fprintf(stderr, "Error: Cannot find data chunk in WAV file\n");

fclose(file);

return 1;

}

// Initialize FFT and DSP

fft_initialise();

c_wav_decoder decoder;

// Create resampler

Resampler resampler(header.sample_rate, (uint32_t)SAMPLE_FREQUENCY);

// Process audio data

const int BUFFER_SIZE = 1024;

uint8_t buffer[BUFFER_SIZE * 4]; // Max size for stereo 16-bit

int16_t output_buffer[16]; // Resampler output buffer

int bytes_per_sample = header.bits_per_sample / 8;

int bytes_per_frame = bytes_per_sample * header.num_channels;

int samples_per_read = BUFFER_SIZE;

uint32_t total_samples = chunk_size / bytes_per_frame;

uint32_t samples_processed = 0;

while (samples_processed < total_samples) {

int samples_to_read = samples_per_read;

if (samples_processed + samples_to_read > total_samples) {

samples_to_read = total_samples - samples_processed;

}

size_t bytes_read = fread(buffer, bytes_per_frame, samples_to_read, file);

if (bytes_read == 0) break;

for (size_t i = 0; i < bytes_read; i++) {

int16_t sample;

if (header.bits_per_sample == 16) {

// 16-bit sample

int16_t* samples = (int16_t*)(buffer + i * bytes_per_frame);

if (header.num_channels == 1) {

sample = samples[0];

} else {

// Mix stereo to mono

sample = (samples[0] + samples[1]) / 2;

}

} else {

// 8-bit sample (unsigned)

uint8_t* samples = buffer + i * bytes_per_frame;

if (header.num_channels == 1) {

sample = ((int16_t)samples[0] - 128) * 256;

} else {

// Mix stereo to mono

sample = (((int16_t)samples[0] - 128) + ((int16_t)samples[1] - 128)) * 128;

}

}

// Resample to target rate

int num_outputs = resampler.process(sample, output_buffer, 16);

// Feed resampled samples to decoder

for (int j = 0; j < num_outputs; j++) {

decoder.process_sample(output_buffer[j]);

}

}

samples_processed += bytes_read;

}

// Flush remaining decoded text

decoder.print_final();

fclose(file);

return 0;

}

Makefile:

CXX = g++

CXXFLAGS = -std=c++17 -O2 -Wall

# Source files

SOURCES = wav_decoder.cpp \

cw_dsp.cpp \

cw_decode.cpp \

cw_classifier.cpp \

cw_data.cpp \

dictionary.cpp \

fft.cpp \

utils.cpp

# Object files

OBJECTS = $(SOURCES:.cpp=.o)

# Output binary

TARGET = cw_wav_decoder

.PHONY: all clean

all: $(TARGET)

$(TARGET): $(OBJECTS)

$(CXX) $(CXXFLAGS) -o $@ $^

%.o: %.cpp

$(CXX) $(CXXFLAGS) -c -o $@ $<

clean:

rm -f $(OBJECTS) $(TARGET)

# Dependencies

wav_decoder.o: wav_decoder.cpp cw_dsp.h fft.h

cw_dsp.o: cw_dsp.cpp cw_dsp.h cw_decode.h fft.h utils.h

cw_decode.o: cw_decode.cpp cw_decode.h cw_classifier.h cw_data.h dictionary.h

cw_classifier.o: cw_classifier.cpp cw_classifier.h

cw_data.o: cw_data.cpp cw_data.h

dictionary.o: dictionary.cpp cw_data.h

fft.o: fft.cpp fft.h

utils.o: utils.cpp utils.h