I have made some progress in implementing PNN classifier for an experimental
version of the FLDIGI Morse decoder.
The experimental decoder creates a symbol stream of ("mark", "space") pairs and
uses Probabilistic Neural Network (PNN) algorithm to calculate best symbol
classification candidate based on a set of class examples created by analyzing
collected data set from real CW traffic.
PNN accomodates errors and significant variance in timing for "dit"/"dah" and
space ratios. Another benefit is that PNN is "one-pass" learning neural network,
whereas many other types require hundreds or thousands of learning cycles to
This software needs still more work and is not ready for alpha release yet. If
you are interested in this experimental Morse decoder you can read more on this
PNN work from here:http://ag1le.blogspot.com/2013/02/probabilistic-neural-network-classifier.html
My goal is to use the symbol stream coming from the PNN classifier and use
Viterbi as the decoder for characters. Based on literature Viterbi should
provide few dB of gain compared to other type decoders.
If you happen to know Viterbi algorithm well I would really like to get some
insights & help in this area.