Hi Steve

Thanks - yes the trap of small sample size caught me :-( 

I also fixed the other problems I reported earlier,  and learned a lesson
(do NOT use MS Powerpoint to edit and copy command lines from).  This is why
I could not get -G and -F to work.  Also, it often throws up slanted
inverted commas, which also don't work. Since moving to Notepad all is now
fine.  

Charlie



-----Original Message-----
From: Steven Franke [mailto:[email protected]] 
Sent: 02 January 2018 14:50
To: WSJT software development
Subject: Re: [wsjt-devel] Inconsistent JT65b reporting

Hi Charlie,  Bill, and all,

> I also tried 40Hz spreading at -23 with 11025 with WSJT-X and found AP and
> no DS did not decode anything, whereas it gets 2/10 with 12000 sample
> rate.  It does OK however  with DS with these -23/40Hz 11025 signals. 
> Perhaps the resampling on very marginal signals like this is responsible?

I think that you need to look at a larger sample to decide whether or not
the decode probability is different between the two sample rates. 

I used the following to generate 100 JT65C files at SNR=-23 dB with 40 Hz
Doppler spread, at sample rate 12000 /s:

./jt65sim -d 40.0 -m C -n 1 -F 1200.0 -t 2.5 -f 100 -s \\-23 -G 10 -M "K9AN
G3WDG RRR"

With AP decoding enabled, G3WDG in the Dx Call box, and Tx5 selected, I got
34/100 decodes. All decodes were AP decodes with aptype=3 or 4. Deep Search
was not enabled.

Then, I repeated the experiment with -S added to the command line so that
the files were generated with 11025 /s rate. When WSJT-X detects a wav file
with 11025/s rate, it calls Joe's wav12.f90 resampler to resample the data
to 12000/s. In this case I got 38/100 decodes. 

It's worth noting that we expect the number of good decodes to have a
binomial distribution. The standard deviation of the binomial distribution
is sigma = sqrt( n*p*(1-p) ) where n is the sample size and p is the decode
probability. My results with 100 simulated wav files suggest that the decode
probability for the case that we are considering is somewhere in the
neighborhood of 0.36. With a sample size of 100, and p=0.36, the standard
deviation of the binomial distribution is sqrt(100*0.36(1-0.36))=4.8. 

Thus, the difference between my results with sample rate 12000 (34/100
decodes) and sample rate 11025 (38/100) is not statistically significant,
i.e. the observed differences are of the same order as the differences that
we'd expect if we just processed two different groups of files, with the
same underlying decode probability for each group. 

In summary, there is no evidence (so far) that resampling in WSJT-X causes
any degradation in decoding performance. It would take many more trials
(>1000) to be able to detect any differences in WSJT-X's decode probability
at the two sample rates. Of course, this experiment does *not* address what
happens when the operating system does the resampling, as would be the case
when processing audio data in real time.

Steve k9an


----------------------------------------------------------------------------
--
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot
_______________________________________________
wsjt-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/wsjt-devel


------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot
_______________________________________________
wsjt-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/wsjt-devel

Reply via email to