For those testing wspr mode in wsjt-x ver 1.6, I’ve just committed r5644 which 
makes decoder #5 from Joe’s table, below, the default decoder in wsjt-x ver 
1.6. It is no longer necessary to separately compile wsprd_exp to get the 
benefits of two-pass decoding. 
Steve k9an

> On Jun 29, 2015, at 6:25 PM, Joe Taylor <j...@princeton.edu> wrote:
> 
> Hi Steve,
> 
>>> The test runs all used the same set of 386 *.wav files, processed with
>>> the following decoders:
>>> 
>>> 1. wsprd, from WSPR-X (baseline decoder)
>>> 2. wspr4
>>> 3. wsprd, as built for WSJT-X v1.6.0 r5636
>>> 4. wsprd_exp, signal subtraction using symbol-by-symbol coherence
>>> 5. wsprd_exp, subtraction with full coherence and test for local maxima
>>> 6. wsprd_exp, subtraction with full coherence and snr>  min_snr
>>> 
>>> For each run the following table gives the number of decodes, the
>>> wall-clock running time, the average time per wav file, and the
>>> "improvement factor" for number of decodes and speed.
>>> 
>>>    Decodes Time1  AvgTime    Improvement   Decoder
>>>              (s)    (s)    Decodes  Speed
>>> ----------------------------------------------------------------
>>> 1.  1451   2111     5.5     1.00    1.00   baseline
>>> 2.  1693   1599     4.1     1.17    1.32   wspr4
>>> 3.  2208    335     0.9     1.52    6.30   WSJT-X v1.6.0 r5636
>>> 4.  2464    413     1.1     1.70    5.11   partial coherence
>>> 5.  2567    431     1.1     1.77    4.90   full coherence
>>> 6.  2839   2136     5.5     1.96    0.99   more candidates
>> 
>> Thanks for running these tests. These agree with my results, although
>> I see somewhat more improvement as you go down the list, probably
>> because my test files are all from 20m under crowded band conditions.
> 
> Agreed: I think my test files were not quite so homogeneous as yours.
> 
>> At some point, we should look at the coherent subtraction lowpass
>> filter. The length (nfilt) and impulse response were chosen without
>> much thought...
> 
> Yes, this may need further tweaking.  In addition, I suspect we can find 
> a criterion better than simply "smspec[j] > min_snr" for choosing 
> candidate frequencies, and likely some other speedups, as well.
> 
>       -- Joe
> 
> ------------------------------------------------------------------------------
> Don't Limit Your Business. Reach for the Cloud.
> GigeNET's Cloud Solutions provide you with the tools and support that
> you need to offload your IT needs and focus on growing your business.
> Configured For All Businesses. Start Your Cloud Today.
> https://www.gigenetcloud.com/
> _______________________________________________
> wsjt-devel mailing list
> wsjt-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/wsjt-devel


------------------------------------------------------------------------------
Don't Limit Your Business. Reach for the Cloud.
GigeNET's Cloud Solutions provide you with the tools and support that
you need to offload your IT needs and focus on growing your business.
Configured For All Businesses. Start Your Cloud Today.
https://www.gigenetcloud.com/
_______________________________________________
wsjt-devel mailing list
wsjt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/wsjt-devel

Reply via email to