Call Search
     

New to Ham Radio?
My Profile

Community
Articles
Forums
News
Reviews
Friends Remembered
Strays
Survey Question

Operating
Contesting
DX Cluster Spots
Propagation

Resources
Calendar
Classifieds
Ham Exams
Ham Links
List Archives
News Articles
Product Reviews
QSL Managers

Site Info
eHam Help (FAQ)
Support the site
The eHam Team
Advertising Info
Vision Statement
About eHam.net

   Home   Help Search  
Pages: [1]   Go Down
  Print  
Author Topic: Ultimate Morse decoder?  (Read 3982 times)
AG1LE
Member

Posts: 137


WWW

Ignore
« on: June 23, 2012, 09:24:38 PM »

Has anybody looked at wavelet transformation to de-noise signals before Morse code detection?

Over the last couple weeks I have read many papers on using wavelets in signal processing and also done some experimentation.  This area looks really promising - some teams have reported detecting and localizing RF pulses down to  -24dB  SNR level.

See details in my blog:
http://ag1le.blogspot.com/2012/06/ultimate-morse-code-decoder.html

73 
Mauri AG1LE
Logged
AG1LE
Member

Posts: 137


WWW

Ignore
« Reply #1 on: June 28, 2012, 04:23:58 PM »

I created 3 new test audio files with Morse code at different signal-to-noise ratios.  

See the new visualizations using Morlet wavelet transformation with Morse code at -9dB and -12dB  SNR level in the original signals
in here: http://ag1le.blogspot.com/2012/06/ultimate-morse-code-decoder.html

Morlet wavelets seem to provide pretty good time AND frequency resolution at the same time.
This has been a problem with traditional short time windowed Fast Fourier Transformations (SFFT).  
With SFFT you can get good accuracy in either frequency OR time domain, but not in both simultaneously.
See http://en.wikipedia.org/wiki/Short-time_Fourier_transform for more scientific explanation.
  

I other words this means that wavelet transformation could enable much improved real time CW decoding even for very noisy signals.
What would you do if you could get 12 dB signal-to-noise gain without investing over $5000 to a large multiband Yagi and a tower?  


Would you work CW stations that are barely audible or below noise level if this feature would be available in software like FLDIGI?  


73  
Mauri AG1LE  
« Last Edit: June 28, 2012, 04:32:28 PM by AG1LE » Logged
AG1LE
Member

Posts: 137


WWW

Ignore
« Reply #2 on: June 28, 2012, 10:48:16 PM »


In other words this means that wavelet transformation could enable much improved real time CW decoding even for very noisy signals.
What would you do if you could get 12 dB signal-to-noise gain without investing over $5000 to a large multiband Yagi and a tower?  


Some more work done tonight writing modified Morlet wavelet algorithm using Octave.

Achieving Morse decoding at -12 dB SNR level seems possible based on these latest results:  
http://ag1le.blogspot.com/2012/06/morse-code-detection-using-modified.html

73
Mauri AG1LE
« Last Edit: June 28, 2012, 11:14:49 PM by AG1LE » Logged
AA4PB
Member

Posts: 12892




Ignore
« Reply #3 on: June 29, 2012, 06:05:34 AM »

Timing is the other side of the Morse decoder problem. While it is possible to decode perfectly times Morse at low SNR, it can be difficult to copy some hand sent Morse even with good SNR. IMHO both sides need to be considered for the untimate (general purpose) decoder.

It looks like you are doing some very interesting work.
Logged
AG1LE
Member

Posts: 137


WWW

Ignore
« Reply #4 on: June 29, 2012, 06:51:33 AM »

Timing is the other side of the Morse decoder problem. While it is possible to decode perfectly times Morse at low SNR, it can be difficult to copy some hand sent Morse even with good SNR. IMHO both sides need to be considered for the untimate (general purpose) decoder.

It looks like you are doing some very interesting work.

Thanks Bob

Fully agree with you on the timing problem. I took a bit different approach on that dimension - I implemented an algorithm (best matching unit)  from well known neural network called Self Organizing Maps (SOM) . The BMU algorithm does a fuzzy matching of incoming pulse pattern against a known  codebook.  See this http://ag1le.blogspot.com/2012/05/fldigi-adding-matched-filter-feature-to.html and this http://ag1le.blogspot.com/2012/05/fldigi-matched-filter-and-som-decoder.html.  This algorithm uses Euclidian distance measure to calculate best match and it seems pretty robust in decoding rhytm variations.

Dave W1HKJ  was kind enough to help getting this SOM decoder code into FLDIGI  version 3.21.43AJ.  I have received some positive comments from early test users.

If I get some help to get this modified Morlet transform implemented in C++ for FLDIGI  we should see significant improvements in handling low SNR signals. At least  I was pretty impressed to get  -12dB SNR so well visible with this simple  Octave simulation.

73
Mauri AG1LE

Logged
Pages: [1]   Go Up
  Print  
 
Jump to:  

Powered by MySQL Powered by PHP Powered by SMF 1.1.11 | SMF © 2006-2009, Simple Machines LLC Valid XHTML 1.0! Valid CSS!