CW stands for Continuous Wave. In CW communications, a continuous carrier is transmitted as long as the key contacts are closed. By making and breaking the contacts into specific patterns, characters and entire messages may be sent.
Here is some work about Ham CW, Morse code, by using Nerual Network, Machine Learning method help us learn, recognize the code.
nn_morse is implemented by pd0wm almost 3 years ago, more details can be found at https://github.com/pd0wm. It works well now, with pyton3.9, pytorch 1.13 + cuda 11.7. It's very useful to broaden vision on the data generating and neural network traning.
morse-dataset is another open source project by souryadey, almost 5 year ago, which is very useful, too. He suggests a method to distinguish 0 and 1 in morse code, by using stochastical method. More detailed introduction can be found at https://github.com/souryadey/morse-dataset.
Also, I'd like to add some features on code generating, then use it with matlab, the new code will be collected in CWlab. Here, ml, nn and dsp are to be used, different technical roadmap. In general, I will try to implement realtime audio input, and with realtime text message as output.
Some notes:
-
The directory models_lib is used to store some pre-trained models, the name would be useful to indicate where it come from.
-
The directory of sounds_lib is used to store some example. Real QSO seems somewaht difficult to the Models.
Release 1.0
-
Change the morse dictionary, add some useful procedural signals, also called ProSigns. At last, the dictionary has 59 chars.
-
Adjust the project structure, and prepare to add more functions. Change decoder, adjust the main function for batching traing work.
-
The decoder.py, for the Dense-LSTM-Dense(DLD) network structure is kept unchanged, keeping for further comparing. It seems pd0wm's nn-morse is strong enough, faster training, smaller network. I have to say that pd0wm's work is very GOOD. In future release, I will try other neural network structure.
Just for fun, keep it simple.
By BFcat @ 2023.01.20