[aprssig] "Best" packet decoder solution

[Matti Aarnio]

On Mon, Dec 27, 2010 at 03:10:53PM -0800, Stephen H. Smith wrote:
> On 12/27/2010 12:02 PM, Lynn W. Deffenbaugh (Mr) wrote:
>> Experimenting with the CD would hopefully get you reproducible
>> results, but IMHO the actual performance of a set of TNCs in your
>> "airspace" as it were, is really what you need to know if you're
>> trying to choose a TNC.  One thing I have learned in my
>> experiences with APRS over RF is that YMMV (Your Mileage May Vary)
>> is an absolute truism.
>> 
> 
> This is what I created the test CD to address!
> <http://wa8lmf.net/TNCtest>

I do thank you for it.

Over xmas I spend couple days trying different algorithms at different
sample rates, pre-/post-emphasis processing, phasing errors, etc..
I have the CD tracks pre-converted to raw sample data that I could feed
to test algorithms and observe their quality.

Each about 26 minutes long recording was processed by my test harness
in 4 to 10 second compute heavy test run.  It supplied a very quick way
to learn that tested algorithm is crappy.

As to the best algorithm - I am not yet sure.
I had time to test a bunch of crappy ideas, but not to implement ones
that are good at high sample rates, 5% tone offsets, presense of
tone-squelsh tones or other low frequency qrm, etc.

One of my references is Thomas Sailer's multimon program from 1996.
It decodes more than his soundmodem, and decodes multiple modulations
in parallel, which soundmodem can't.  This is of course due to their
different architectures.

However I do think that mere 950 packets decoded out of the test track 1
is not yet enough..  Most of my tests fell below 900, best of Sailer's
algorithms decoded more, but still several packets did just result in
CRC error or something alike.

In all probability the "DSP" is the answer, but what should the DSP do?
Float math, integer math, FIRs, IIRs, DPLLs, BPFs, LPFs, HPFs, ...
The tool chest has lots of things, but finding a most reliably working
combination will not be simple.


> --
> Stephen H. Smith    wa8lmf (at) aol.com
> Home Page:          http://wa8lmf.net

73 de Matti, OH2MQK