New Bayesian Morse Decoder
In my previous post I shared initial results of the new Bayesian Morse Decoder. I have worked a bit more on this topic to integrate the decoder software to FLDIGI v3.21.75. Dave W1HKJ provided me some space on his web server so I posted full source code including all changes and new source files related to Bayesian Morse decoder in here.
To demonstrate the Bayesian decoding capability I tuned my KX3 on W1AW frequency 3.58105 Mhz at 8:00 PM EST today and let two identical FLDIGI copies of the same v3.21.75 software running from the same audio source. I have a SignaLink USB connected between KX3 and my ThinkPad laptop running Linux Mint OS.
On Fig 1. below I have the legacy SOM decoder on the left side and the new Bayesian decoder on the right . The legacy decoder makes a lot of errors and CW bulletin is not really readable. On the right side I have enabled Bayesian decoder and you can actually read the bulletin, though the lack of proper word spaces make this still a bit difficult to read. In any case Bayesian decoder provides significant accuracy improvement with real life noisy CW signals.
 |
| Figure 1. FLDIGI running legacy and Bayesian Morse decoder |
If you choose to compile FLDIGI from the sources yourself please see Figure 2. below. When you go to menu item Configure / Modems / CW / General and click the tick box "Bayes decoding" that will enable the new Bayesian decoder (and disables legacy decoder). You will also notice that CW RX speed indicator (Fig 1. bottom left corner of FLDIGI window) will change more rapidly - this is a feature of the new decoder. It evaluates the instantaneous speed continuously and uses this information as part of the decoding process.
 |
| Figure 2. Enabling Bayesian decoder. |
I would be interested in getting your feedback and suggestions how to improve this software. Current version is buggy and it does crash occasionally. Also, I still have not found the root cause of the problem that causes approximately 3% base error rate.
There are number of improvement areas that need work. For example noise estimator ("noise.c") does not seem to be able to keep track of the real noise level. This noise power information is later used by Kalman filters ("kalfil.c") in estimating likelihood of keystate given the observed signals. Bayesian inference is then used in updating posterior conditional probabilities for each new path in ("probp.c"). I am getting intermittent overflows in this section of the code. Any help in this area would be very welcome.
However, given the visible improvement in the decoding accuracy at least during certain conditions (when you have the signal level correct, not too much noise etc.) I am now posting this for alpha testing with hopes that I could get some more eyes to look at the code and find those bugs.
Happy Holidays!
73
Mauri AG1LE
Hi,
ReplyDeleteI've built the software. I have no experience in amateur radio but do have some in C and signal processing. I'm interested in seeing if I can help with the decoder. Could you please provide some sample wav files and instructions for how to run the them through the decoder via the UI?
Cheers,
Roy
Hi Roy
DeleteHere is an example WAV file https://www.dropbox.com/sh/2j9chxsizxjeqgp/4C36_4qby6/rand16db20wpm.wav that you can use. If you are able to compile FLDIGI from sources (they are here http://www.w1hkj.com/ag1le/) you can install the software and open the WAV file by "File / Audio / Playback" menu. Once you have the audio running enable tickbox "Bayesian decoder" in Fig2. above. This menu is in "Configure / Modems / CW / General" section.
Hi, I can't find the "File / Audio / Playback" menu. Under "File" I only have "Folders", "Macros", "Text Capture" and "Exit".
DeleteHi Anonymous
Deletehave you installed sndfile library on your system? In many systems you can do this by "sudo apt-get install libsndfile" or "sudo apt-get install libsndfile-dev" if you want sources. You need to run ./configure again and compile sources.
73
Mauri
IS there a language model here? Language models are typically used in speech recognition, and I bet could provide a huge improvement to the readout. If you don't have one, a model would substantially change your inference procedures, but it would provide some "top down" knowledge about likely sequences of characters that I bet would help you a lot!
ReplyDeleteThis software does not include a language model as ham radio operators use many different kinds of "languages". In contest situation they use short acronyms, while in "rag chewing" more normal language but peppered with shorthands etc. I have thought about language models but would like to get this layer working properly before starting another project.
DeleteMerry Christmas and a Hapy New Year to you and yours.
ReplyDeleteI am using the Windows version of Fldigi (3.21.77) is there any method to include Bayes decoding? Maybe in a futuer version?
73
Once I get this working in reliable manner then Dave W1HKJ can make a decision to include it in mainline FLDIGI releases. However, I need more help to finish alpha testing and fix the bugs in the code.
Delete73
Mauri
Thank you, unfortunately I can n't help you in this testings. Good luck.
ReplyDelete73
Your Bayesian decoder does not detect WORD SPACE of 7 bits and makes text almost unreadable to humans. Compare FLDIGI with good MRP40 CW decoder.
ReplyDeleteI am glad you continued your research on this noncommercial subject. I stoped mine as I enjoy ionospheric peak for few more years.
73 de Mario, S56A, N1YU - HF CW DSP PhD in 1990s and again in 2010
Hi Marijan
DeleteIn the initial alpha release there was a bug in transl.cxx file that caused word space to be decoded but not printed on scree. This bug was fixed in the subsequent release. The latest release is in here: http://www.w1hkj.com/ag1le/fldigi-3.21.77.bayes.tar.gz
73
Mauri AG1LE
Great! I got that latest Fldigi realease on this X60S notebook so I would test your decoder against Morse Runner using stereo loop. Mine works fine until QRM is added :-)
DeleteFew comments after quick test: O always shown as 0(zero). First char detection poor. Level sensitivity high, sporadic E detected. No tracking (AFC) within RX filter bandwidth. Low speed decoding puts spaces. Test done with Morse Runner in single calls mode without QRM.
DeleteNow is a time to walk a dog and later talk with two quantum mechanics docs from UK :-)
Hi Mario
DeleteThanks for test results. Are you using version http://www.w1hkj.com/ag1le/fldigi-3.21.77.bayes.tar.gz ?
Let me comment found issues one by one:
1) O always shown as 0(zero).
The translation table is in file transl.cxx - see lines 44 and 87 (copied below)
{29,30,"O"}, // ---
{00,00,"0"}, // -----
Could this be some localization issue? I have US English locale in my development laptop, running Linux Mint 14 with KDE desktop.
2) First char detection poor.
This is a known issue. Speed estimation algorithm converges too slowly.
3) Level sensitivity high, sporadic E detected.
Are you using Squelch function? SQL button on bottom right corner should be on and squelch level adjustment bar on right bottom side should be adjusted so that noise won't trigger decoding.
4) No tracking (AFC) within RX filter bandwidth.
This is correct, currently tracking is manual, you need to click on waterfall to center the RX filter to signal. I have been testing some algorithms to automatically find and track multiple signals but haven't integrated that work to FLDIGI yet.
5) Low speed decoding puts spaces.
What speed in WPM are you testing?
The lower and upper WPM limits are set in spdtr.cxx - lines 75 and 78. I have lower limit set at 8 WPM and upper limit to 80 WPM.
Thanks again for your feedback and test results.
73
Mauri AG1LE