Re: [wsjt-devel] FreqCal Mode

[Jay Hainline]

Thanks for confirming it is on the air Joe. I am hearing it this afternoon
on 3330 KHz.

 

73 Jay KA9CFD

 

From: Joe [mailto:n...@mwt.net] 
Sent: January 10, 2017 15:00
To: wsjt-devel@lists.sourceforge.net
Subject: Re: [wsjt-devel] FreqCal Mode

 

At 15:00Z here in SW Wisconsin they are S-5,
But I'm not sure what mode.

On AM Barely Understandable,
On LSB forget it, just a mess.
On USB perfect audio, just like I remember from my SWL days on the 7.3xx
freq. AM

Joe WB9SBD


The Original Rolling Ball Clock
Idle Tyme
Idle-Tyme.com
 <http://www.idle-tyme.com> http://www.idle-tyme.com

On 1/10/2017 8:53 AM, Joe Taylor wrote:

Hi Jay,
 
On 1/10/2017 8:44 AM, Jay Hainline wrote:

Can anyone confirm CHU at 3330 is actually on the air? I do not hear it here
and suspect it no longer is.
 
73 Jay KA9CFD

 
I'd like to know the answer to your question, too!  I find no indication 
on the NRC web site that 3330 is no longer used; but I haven't heard the 
signal here, recently.
 
  -- Joe, K1JT
 

--
Developer Access Program for Intel Xeon Phi Processors
Access to Intel Xeon Phi processor-based developer platforms.
With one year of Intel Parallel Studio XE.
Training and support from Colfax.
Order your platform today. http://sdm.link/xeonphi
___
wsjt-devel mailing list
wsjt-devel@lists.sourceforge.net <mailto:wsjt-devel@lists.sourceforge.net> 
https://lists.sourceforge.net/lists/listinfo/wsjt-devel
 
 
 

 

--
Developer Access Program for Intel Xeon Phi Processors
Access to Intel Xeon Phi processor-based developer platforms.
With one year of Intel Parallel Studio XE.
Training and support from Colfax.
Order your platform today. http://sdm.link/xeonphi___
wsjt-devel mailing list
wsjt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/wsjt-devel

Re: [wsjt-devel] FreqCal Mode

[Joe Taylor]

Hi Edson,

On 1/10/2017 12:39 PM, Edson W. R. Pereira wrote:

> I got across this paper some time ago. The algorithm suggested could
> perhaps be applied to FreqCal.
>
> https://mgasior.web.cern.ch/mgasior/pap/FFT_resol_note.pdf
>
> 73, Edson PY2SDR

Thanks for the pointer to an excellent paper!

The parabolic interpolation described by Gasior and Gonzalez is exactly 
what is done in peakup.f90.  It easily gets us easily down to better 
than 0.1 Hz, with a stable signal.

The even better performance they get using Gaussian interpolation would 
not be useful in our FreqCal procedure.  Owing to ionospheric 
propagation effects, our HF test signals are not stable in frequency to 
a tiny fraction of one Hz.

-- Joe, K1JT

--
Developer Access Program for Intel Xeon Phi Processors
Access to Intel Xeon Phi processor-based developer platforms.
With one year of Intel Parallel Studio XE.
Training and support from Colfax.
Order your platform today. http://sdm.link/xeonphi
___
wsjt-devel mailing list
wsjt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/wsjt-devel

Re: [wsjt-devel] FreqCal Mode

[Bill Somerville]

On 10/01/2017 18:42, Bill Somerville wrote:
> So for example if a window function
> that caused broadening with a Gaussian shape then a Gaussian curve
> fitting would be the best candidate for estimating inter-bin frequency
> peaks.

Hi Hi!

thanks for the paper link Edson, clearly I have read something similar.

73
Bill
G4WJS.


--
Developer Access Program for Intel Xeon Phi Processors
Access to Intel Xeon Phi processor-based developer platforms.
With one year of Intel Parallel Studio XE.
Training and support from Colfax.
Order your platform today. http://sdm.link/xeonphi
___
wsjt-devel mailing list
wsjt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/wsjt-devel

Re: [wsjt-devel] FreqCal Mode

[Bill Somerville]

On 10/01/2017 18:23, Bill Somerville wrote:
> I suspect that MacLeod's estimator is not a simple replacement for 
> Joe's simple three ... point parabolic fit

Hi Mike,

I should also point out, as I understand it, that the choice of window 
function and interpolating estimator for frequency finding are closely 
related. The window function will have a bearing on the curve shape of 
peaks, this is because the window function trades off DFT end point 
discontinuity with peak broadening. So for example if a window function 
that caused broadening with a Gaussian shape then a Gaussian curve 
fitting would be the best candidate for estimating inter-bin frequency 
peaks.

It may be that a rectangular window i.e. no window function at all, has 
some matching inter-bin frequency estimator that performs best although 
I doubt it if there are other signals present in the spectrum of the 
DFT. Note that the FTol only applies to the search and does not 
eliminate artefacts caused by other signals.

I can't make a suggestion for improvement but it is probably not 
productive to take it further with radio signals.

73
Bill
G4WJS.


--
Developer Access Program for Intel Xeon Phi Processors
Access to Intel Xeon Phi processor-based developer platforms.
With one year of Intel Parallel Studio XE.
Training and support from Colfax.
Order your platform today. http://sdm.link/xeonphi
___
wsjt-devel mailing list
wsjt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/wsjt-devel

Re: [wsjt-devel] FreqCal Mode

[Bill Somerville]

On 10/01/2017 18:02, Black Michael wrote:
Hmmm...gotta' wonder though...which is more "real" ??  Without knowing 
the frequency stability of your sound card can't say which is actually 
more accurate, can we?


McLeod's is supposed to have pretty good behavior.


Hi Mike,

nope, the sound card is not involved - as stated I am using a digital 
loopback. I believe my test is valid and the fact that selected expected 
frequencies (e.g. 1500Hz) give exact results even when the signal 
amplitude is varied to extremes and with both a rectangular and sin^2 
windowing pre-DFT seems to back that up. Of course there are always 
other potential sources of errors such as a faulty upstream filter 
algorithms, off by one sample handling errors, and so on but these 
components are used elsewhere and behave as advertised.


I suspect that MacLeod's estimator is not a simple replacement for Joe's 
simple three, adjacent magnitude point, parabolic curve fit. I believe 
it requires the DFT coefficients to be something other than the 
magnitude only information we currently use. If it used phase 
information in some way it could improve estimation.


As Joe has stated, LF and HF signals suffer spreading and shifting that 
we cannot compensate for and the current frequency estimation resolution 
is probably already one order of magnitude better than the best estimate 
that is possible.


73
Bill
G4WJS.

--
Developer Access Program for Intel Xeon Phi Processors
Access to Intel Xeon Phi processor-based developer platforms.
With one year of Intel Parallel Studio XE.
Training and support from Colfax.
Order your platform today. http://sdm.link/xeonphi___
wsjt-devel mailing list
wsjt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/wsjt-devel

Re: [wsjt-devel] FreqCal Mode

[Black Michael]

Hmmm...gotta' wonder though...which is more "real" ??  Without knowing the 
frequency stability of your sound card can't say which is actually more 
accurate, can we?
McLeod's is supposed to have pretty good behavior.
http://www.ericjacobsen.org/fe2/fe2.htm


One more if you're up for itEric's quadratic which has a pretty flat 
error...this looks very similar to what's in peakup.f90 now...but simpler
subroutine peakup(ym,y0,yp,dx)! Eric Jacobsen's estimator  
dx=(ym-yp)/((2*y0)-ym-yp)  returnend subroutine peakup

de Mike W9MDB

  From: Bill Somerville <g4...@classdesign.com>
 To: wsjt-devel@lists.sourceforge.net 
 Sent: Tuesday, January 10, 2017 11:15 AM
 Subject: Re: [wsjt-devel] FreqCal Mode
   
 On 10/01/2017 16:52, Black Michael wrote:
  
 try your loopback test using McLeod's estimator. Just replace peakup.f90 with 
this: 
  
  subroutine peakup(ym,y0,yp,dx) ! McLeod's estimator   delta = 
(ym-yp)/(2*y0+ym+yp)   dx=(sqrt(1+8*delta*delta)-1)/4*delta;   return end 
subroutine peakup 
 Hi Mike, here you go: 16:58:29660  1  1500  1500.000 0.000  -39.4   
78.9
16:58:43660  1  1501  1501.088 0.088  -39.5   78.1
16:58:57660  1  1502  1501.953-0.047  -30.2   68.2
16:59:16880  1  1503  1503.039 0.039  -27.5   63.0
16:59:39   1210  1  1504  1503.902-0.098  -39.5   77.9
16:59:50   1210  1  1505  1504.991-0.009  -39.5   79.0
17:00:11   2500  1  1506  1506.078 0.078  -39.6   78.3
17:00:23   2500  1  1507  1506.944-0.056  -35.8   74.8
17:00:41   3330  1  1508  1508.030 0.030  -39.4   78.8
17:00:53   3330  1  1509  1508.893-0.107  -39.6   77.8
17:01:11   5000  1  1510  1509.983-0.017  -39.5   79.0

 Not an improvement, actually worse than the orginal non-windowed DFT plus and 
peakup function.
 
 73
 Bill
 G4WJS.
  
--
Developer Access Program for Intel Xeon Phi Processors
Access to Intel Xeon Phi processor-based developer platforms.
With one year of Intel Parallel Studio XE.
Training and support from Colfax.
Order your platform today. http://sdm.link/xeonphi
___
wsjt-devel mailing list
wsjt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/wsjt-devel


   --
Developer Access Program for Intel Xeon Phi Processors
Access to Intel Xeon Phi processor-based developer platforms.
With one year of Intel Parallel Studio XE.
Training and support from Colfax.
Order your platform today. http://sdm.link/xeonphi___
wsjt-devel mailing list
wsjt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/wsjt-devel

Re: [wsjt-devel] FreqCal Mode

[Edson W. R. Pereira]

Hello everyone:

I got across this paper some time ago. The algorithm suggested could
perhaps be applied to FreqCal.

https://mgasior.web.cern.ch/mgasior/pap/FFT_resol_note.pdf

73, Edson PY2SDR


---
- We humans have the capability to do amazing things if we work together.
- Nós seres humanos temos a capacidade de fazer coisas incríveis se
trabalharmos juntos.

On Tue, Jan 10, 2017 at 2:08 PM, Bill Somerville 
wrote:

> On 09/01/2017 23:57, Joe Taylor wrote:
>
> Still, it will probably be best to put a window on the FFT.
>
> Hi Joe,
>
> the new sin^2 window function improves accuracy most of the time, here
> are some sample measurements using identical tones synthesised by another
> WSJT-X instance:
>
> 15:23:57   5000  1  1500  1500.000 0.000  -39.4   78.8
> 15:24:06   7850  1  1501  1501.022 0.022  -39.5   78.0
>
> 15:24:41  1  1  1502  1501.979-0.021  -39.3   78.5
>
> 15:25:06  14670  1  1503  1503.018 0.018  -39.4   78.6
>
> 15:25:34  15000  1  1505  1504.976-0.024  -22.2   59.9
>
> 15:26:06  2  1  1506  1506.025 0.025  -39.4   78.2
>
> 15:26:46660  1  1507  1506.977-0.023  -39.4   78.5
>
> 15:27:09880  1  1508  1508.014 0.014  -39.4   78.8
>
> 15:27:36   1210  1  1509  1508.991-0.009  -39.5   77.7
>
> 15:28:04   2500  1  1510  1509.991-0.009  -39.1   78.5
>
> without the window:
>
> 15:48:55660  1  1500  1500.000 0.0002.8   77.5
>
> 15:49:09660  1  1501  1501.010 0.010   12.6   64.1
>
> 15:49:20660  1  1502  1501.956-0.044   19.5   60.1
>
> 15:49:43880  1  1503  1503.037 0.037   18.0   61.9
>
> 15:49:55880  1  1504  1503.946-0.054   23.4   54.1
>
> 15:50:06   1210  1  1505  1504.991-0.0096.5   73.8
>
> 15:50:18   1210  1  1506  1506.061 0.061   22.6   55.9
>
> 15:50:50   2500  1  1507  1506.949-0.051   20.9   58.5
>
> 15:51:09   3330  1  1508  1508.029 0.029   16.0   64.1
>
> 15:51:20   3330  1  1509  1508.966-0.034   23.5   53.3
>
> 15:51:39   5000  1  1510  1509.983-0.017   11.7   68.5
>
> I note that the SNR figures are roughly the same but the "Level" column
> has changed dramatically.
>
> 73
> Bill
> G4WJS.
>
> 
> --
> Developer Access Program for Intel Xeon Phi Processors
> Access to Intel Xeon Phi processor-based developer platforms.
> With one year of Intel Parallel Studio XE.
> Training and support from Colfax.
> Order your platform today. http://sdm.link/xeonphi
> ___
> wsjt-devel mailing list
> wsjt-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/wsjt-devel
>
>
--
Developer Access Program for Intel Xeon Phi Processors
Access to Intel Xeon Phi processor-based developer platforms.
With one year of Intel Parallel Studio XE.
Training and support from Colfax.
Order your platform today. http://sdm.link/xeonphi___
wsjt-devel mailing list
wsjt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/wsjt-devel

Re: [wsjt-devel] FreqCal Mode

[Rich Griffiths]

CHU transmits in USB with a reduced carrier.

... Rich


On 01/10/2017 10:00 AM, Joe wrote:

At 15:00Z here in SW Wisconsin they are S-5,
But I'm not sure what mode.

On AM Barely Understandable,
On LSB forget it, just a mess.
On USB perfect audio, just like I remember from my SWL days on the 
7.3xx freq. AM


Joe WB9SBD
Sig
The Original Rolling Ball Clock
Idle Tyme
Idle-Tyme.com
http://www.idle-tyme.com
On 1/10/2017 8:53 AM, Joe Taylor wrote:

Hi Jay,

On 1/10/2017 8:44 AM, Jay Hainline wrote:

Can anyone confirm CHU at 3330 is actually on the air? I do not hear it here
and suspect it no longer is.

73 Jay KA9CFD

I'd like to know the answer to your question, too!  I find no indication
on the NRC web site that 3330 is no longer used; but I haven't heard the
signal here, recently.

-- Joe, K1JT

--
Developer Access Program for Intel Xeon Phi Processors
Access to Intel Xeon Phi processor-based developer platforms.
With one year of Intel Parallel Studio XE.
Training and support from Colfax.
Order your platform today.http://sdm.link/xeonphi
___
wsjt-devel mailing list
wsjt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/wsjt-devel







--
Developer Access Program for Intel Xeon Phi Processors
Access to Intel Xeon Phi processor-based developer platforms.
With one year of Intel Parallel Studio XE.
Training and support from Colfax.
Order your platform today. http://sdm.link/xeonphi


___
wsjt-devel mailing list
wsjt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/wsjt-devel


--
Developer Access Program for Intel Xeon Phi Processors
Access to Intel Xeon Phi processor-based developer platforms.
With one year of Intel Parallel Studio XE.
Training and support from Colfax.
Order your platform today. http://sdm.link/xeonphi___
wsjt-devel mailing list
wsjt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/wsjt-devel

Re: [wsjt-devel] FreqCal Mode

[Bill Somerville]

On 10/01/2017 15:05, Bill Somerville wrote:
> Table entries may be directly edited by
> double clicking on fields and using the usual editing keystrokes.

Hi Claude & all,

I should also have added that the table column headings can be clicked 
to select sort column and sort order, so for example you might click the 
"Mode" column heading to bring all the "FreqCal" frequency entries 
together for easy review and maintenance.

I also forgot to add that the "Working Frequencies" table contents are, 
like all settings, persisted between operating sessions. Note they are 
only committed once the "Ok" button is clicked so do not leave the 
settings window with "Cancel" button, "ESC" key or, the system window 
close button (the X or red dot in the corner) unless you wish to discard 
your changes.

73
Bill
G4WJS.


--
Developer Access Program for Intel Xeon Phi Processors
Access to Intel Xeon Phi processor-based developer platforms.
With one year of Intel Parallel Studio XE.
Training and support from Colfax.
Order your platform today. http://sdm.link/xeonphi
___
wsjt-devel mailing list
wsjt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/wsjt-devel

Re: [wsjt-devel] FreqCal Mode

[Steven Franke]

Mike, 

I’ll have to disagree with your statement that “it’s not the discontinuity…”. 

A more careful and correct statement could have been something along these 
lines:  "instead of explaining the observed phenomenon in terms of an endpoint 
discontinuity, one could also note that the phenomenon will be observed 
whenever the tone frequency is not an integral multiple of the FFT bin width, 
i.e. whenever the tone does not fall in the exact center of a bin.” 

I’ll leave it as an exercise for you to convince yourself that these two 
explanations are two ends of the same stick. 

Sometimes it’s better to think before you respond.

Steve k9an

> On Jan 10, 2017, at 7:21 AM, Michael Black  wrote:
> 
> It's not the discontinuity -- it's the bin values.
> If the desired frequency is not an EXACT multiple of the bin value you'll get 
> a reduced peak and leakageone of the downfalls of DFTs.  As Joe noted a 
> larger FFT would have more resolution but is likely not going to improve 
> things.
> A rectangular window preserves minimum main lobe width at the expense of 
> higher leakage.  Other windows will increase peak value and have reduced 
> leakage but also have less accuracy due to spreading of the main peak.  The 
> leakage should not be much of a factor for this frequency calibration unless 
> you are borderline on the peak SNR where a Triangular window would be best 
> but picks up less then 1dB over a rectangular window -- see 
> http://www.bores.com/courses/advanced/windows/files/windows.pdf 
> 
> 
> There's already a fit function in the code which is giving better than the 
> bin width in accuracy.  Could be improved on slightly perhaps with a cubic 
> spline but looks good enough as-is to me and I doubt a spline would really 
> improve much at all.
> 
> On my freq cal I see numerous RMS entries < 0.1Hz -- Average RMS is 0.27 -- 
> so the peak fit being done is doing pretty well.
> 
> de Mike W9MDB
> 
> On Mon, Jan 9, 2017 at 5:23 PM, Steven Franke  > wrote:
> Since Joe is probably eating dinner at the moment…
> 
> With 55296 point FFT, a 1600 Hz tone (7.5 sa/period) has 55296/7.5=7372.8 
> cycles in a record - so there is an endpoint discontinuity (i.e. a 
> discontinuity between the first and last points of the record).
> 
> A 1500 Hz tone (8 samples per period) has 6912 cycles in a record, so there’s 
> no endpoint discontinuity.
> 
> I think that what you’re seeing is the result of using a non-windowed FFT. 
> It’s easily addressed, if necessary.
> 
> Steve
> 
> 
> > On Jan 9, 2017, at 11:07 PM, Bill Somerville  > > wrote:
> >
> > On 09/01/2017 22:28, Bill Somerville wrote:
> >> ... for sure the three
> >> magnitudes going into peakup are much flatter when the error is greater ...
> > Hi Joe,
> >
> > Here are some examples, these are DFT magnitudes (s vector) for five
> > bins around a 1500Hz tone:
> >
> >   0.107814305  0.175234795   9.38052759E-02   17124122.0   
> > 8.95743668E-02   8.78201425E-02   4.62990999E-02
> >7.97733366E-02  0.128112748  0.307582766   17126434.0  
> > 0.167245105   1.00381924E-02   1.45341307E-02
> >5.32349795E-02   7.02278912E-02  0.130553111   17125166.0  
> > 0.135780737   6.56466857E-02  0.104219861
> >
> > and here is the same for a 1600Hz tone:
> >
> >79235.5156   191998.594   973516.500   15560843.0   
> > 432274.188   128688.836   60769.3555
> >79641.3359   192280.516   972233.375   15562208.0   
> > 432276.500   128727.242   60738.6406
> >79411.3125   192080.391   972825.562   15561087.0   
> > 431534.812   128623.734   60596.7227
> >
> > These numbers are from print *, s(ipk-2:ipk+2) with DF set to the tone
> > and FTol at 1000Hz.
> >
> > 73
> > Bill
> > G4WJS.
> >
> >
> > --
> > Developer Access Program for Intel Xeon Phi Processors
> > Access to Intel Xeon Phi processor-based developer platforms.
> > With one year of Intel Parallel Studio XE.
> > Training and support from Colfax.
> > Order your platform today. http://sdm.link/xeonphi 
> > ___
> > wsjt-devel mailing list
> > wsjt-devel@lists.sourceforge.net 
> > https://lists.sourceforge.net/lists/listinfo/wsjt-devel 
> > 
> 
> 
> --
> Developer Access Program for Intel Xeon Phi Processors
> Access to Intel Xeon Phi processor-based developer platforms.
> With one year of Intel Parallel Studio XE.
> Training and support from Colfax.
> Order your platform today. 

Re: [wsjt-devel] FreqCal Mode

[Michael Black]

It's not the discontinuity -- it's the bin values.
If the desired frequency is not an EXACT multiple of the bin value you'll
get a reduced peak and leakageone of the downfalls of DFTs.  As Joe
noted a larger FFT would have more resolution but is likely not going to
improve things.
A rectangular window preserves minimum main lobe width at the expense of
higher leakage.  Other windows will increase peak value and have reduced
leakage but also have less accuracy due to spreading of the main peak.  The
leakage should not be much of a factor for this frequency calibration
unless you are borderline on the peak SNR where a Triangular window would
be best but picks up less then 1dB over a rectangular window -- see
http://www.bores.com/courses/advanced/windows/files/windows.pdf

There's already a fit function in the code which is giving better than the
bin width in accuracy.  Could be improved on slightly perhaps with a cubic
spline but looks good enough as-is to me and I doubt a spline would really
improve much at all.

On my freq cal I see numerous RMS entries < 0.1Hz -- Average RMS is 0.27 --
so the peak fit being done is doing pretty well.

de Mike W9MDB

On Mon, Jan 9, 2017 at 5:23 PM, Steven Franke  wrote:

> Since Joe is probably eating dinner at the moment…
>
> With 55296 point FFT, a 1600 Hz tone (7.5 sa/period) has 55296/7.5=7372.8
> cycles in a record - so there is an endpoint discontinuity (i.e. a
> discontinuity between the first and last points of the record).
>
> A 1500 Hz tone (8 samples per period) has 6912 cycles in a record, so
> there’s no endpoint discontinuity.
>
> I think that what you’re seeing is the result of using a non-windowed FFT.
> It’s easily addressed, if necessary.
>
> Steve
>
>
> > On Jan 9, 2017, at 11:07 PM, Bill Somerville 
> wrote:
> >
> > On 09/01/2017 22:28, Bill Somerville wrote:
> >> ... for sure the three
> >> magnitudes going into peakup are much flatter when the error is greater
> ...
> > Hi Joe,
> >
> > Here are some examples, these are DFT magnitudes (s vector) for five
> > bins around a 1500Hz tone:
> >
> >   0.107814305  0.175234795   9.38052759E-02   17124122.0
>  8.95743668E-02   8.78201425E-02   4.62990999E-02
> >7.97733366E-02  0.128112748  0.307582766   17126434.0
> 0.167245105   1.00381924E-02   1.45341307E-02
> >5.32349795E-02   7.02278912E-02  0.130553111   17125166.0
> 0.135780737   6.56466857E-02  0.104219861
> >
> > and here is the same for a 1600Hz tone:
> >
> >79235.5156   191998.594   973516.500   15560843.0
>  432274.188   128688.836   60769.3555
> >79641.3359   192280.516   972233.375   15562208.0
>  432276.500   128727.242   60738.6406
> >79411.3125   192080.391   972825.562   15561087.0
>  431534.812   128623.734   60596.7227
> >
> > These numbers are from print *, s(ipk-2:ipk+2) with DF set to the tone
> > and FTol at 1000Hz.
> >
> > 73
> > Bill
> > G4WJS.
> >
> >
> > 
> --
> > Developer Access Program for Intel Xeon Phi Processors
> > Access to Intel Xeon Phi processor-based developer platforms.
> > With one year of Intel Parallel Studio XE.
> > Training and support from Colfax.
> > Order your platform today. http://sdm.link/xeonphi
> > ___
> > wsjt-devel mailing list
> > wsjt-devel@lists.sourceforge.net
> > https://lists.sourceforge.net/lists/listinfo/wsjt-devel
>
>
> 
> --
> Developer Access Program for Intel Xeon Phi Processors
> Access to Intel Xeon Phi processor-based developer platforms.
> With one year of Intel Parallel Studio XE.
> Training and support from Colfax.
> Order your platform today. http://sdm.link/xeonphi
> ___
> wsjt-devel mailing list
> wsjt-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/wsjt-devel
>
--
Developer Access Program for Intel Xeon Phi Processors
Access to Intel Xeon Phi processor-based developer platforms.
With one year of Intel Parallel Studio XE.
Training and support from Colfax.
Order your platform today. http://sdm.link/xeonphi___
wsjt-devel mailing list
wsjt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/wsjt-devel

Re: [wsjt-devel] FreqCal Mode

[Bill Somerville]

On 09/01/2017 23:57, Joe Taylor wrote:
> ... so there's no
> way we're going to get accuracies much better than 0.2 (or maybe 0.1)
> Hz.  This is by design; since we're using HF-propagated signals, which
> can have Doppler shifts or spreads greater than 0.2 Hz, there is little
> point in doing longer FFTs.  Much better to do as FreqCal does it now:
> do many FFTs each of length several seconds, and see how consistent the
> results may be. ...

Hi Joe,

I assumed this inaccuracy may well be swamped. It was just so enticing 
to see those three zeros after the DP hi hi. I did start trying a window 
function but was already bogged down with just how long the sample 
vector is. Does that matter or can the window function be applied over 
the zeroed tail? I suspect not.

I assume the multiple smaller FFTs make it easy to average out noise as 
well, should the peak finding be done after averaging rather than once 
per FFT?

I will divert my attention to the other issue with the synchronization 
of frequency changes, I suspect I may have over simplified something 
when I switch to using the frequencies list.

73
Bill
G4WJS.


--
Developer Access Program for Intel Xeon Phi Processors
Access to Intel Xeon Phi processor-based developer platforms.
With one year of Intel Parallel Studio XE.
Training and support from Colfax.
Order your platform today. http://sdm.link/xeonphi
___
wsjt-devel mailing list
wsjt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/wsjt-devel

Re: [wsjt-devel] FreqCal Mode

[Joe Taylor]

Hi Bill,

On 1/9/2017 5:28 PM, Bill Somerville wrote:

> I have just been looking at a problem with FeqCal when FTol>DF, I have a
> fix for that but I also noticed that, when I connect another WSJT-X
> instance as a tone generator back to back (digital loopback) with an
> instance running FreqCal, I see slightly broader peaks from the DFT for
> all but a few frequencies. These cause the peakup routine be less
> accurate. For the following tone frequencies, and probably others:
>
> 500Hz 750Hz 1000Hz 1250Hz 1500Hz 1750Hz 2000Hz 2250Hz ... 4500Hz
>
> I see accuracy>3DP but most other frequencies I see accuracy to only
> ~1.5DP.

This is one of the ways in which my initial implementation of FreqCal 
mode uses quick-and-dirty code.  Steve's explanation is surely correct; 
a windowed FFT would not show these (potentially undesirable) effects. 
But you should not be fooled by seeing a frequency "measurement" 
displayed to 3 decimal places.  The bin width for a 55296-point DFT (or 
FFT) with 12000 Hz sample rate is 12000/55296 = 0.217 Hz, so there's no 
way we're going to get accuracies much better than 0.2 (or maybe 0.1) 
Hz.  This is by design; since we're using HF-propagated signals, which 
can have Doppler shifts or spreads greater than 0.2 Hz, there is little 
point in doing longer FFTs.  Much better to do as FreqCal does it now: 
do many FFTs each of length several seconds, and see how consistent the 
results may be.

Still, it will probably be best to put a window on the FFT.

-- Joe

--
Developer Access Program for Intel Xeon Phi Processors
Access to Intel Xeon Phi processor-based developer platforms.
With one year of Intel Parallel Studio XE.
Training and support from Colfax.
Order your platform today. http://sdm.link/xeonphi
___
wsjt-devel mailing list
wsjt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/wsjt-devel

Re: [wsjt-devel] FreqCal Mode

[Bill Somerville]

On 09/01/2017 23:23, Steven Franke wrote:
> With 55296 point FFT, a 1600 Hz tone (7.5 sa/period) has 55296/7.5=7372.8 
> cycles in a record - so there is an endpoint discontinuity (i.e. a 
> discontinuity between the first and last points of the record).
>
> A 1500 Hz tone (8 samples per period) has 6912 cycles in a record, so there’s 
> no endpoint discontinuity.
>
> I think that what you’re seeing is the result of using a non-windowed FFT. 
> It’s easily addressed, if necessary.

Hi Steve,

Ah ok, thanks for doing the math. I was deliberately saying DFT rather 
than FFT because I knew in the back of my mind somewhere that the cyclic 
characteristics had something to do with it but it didn't click with me. 
I need to do more of this stuff so it comes a bit more naturally and I 
don't have to go back to basics for every new situation.

I suppose it depends if other factors in real world on air conditions, 
rather than my near perfect tone closed set up, are going to swamp the 
small inaccuracy introduced.

73
Bill
G4WJS.


--
Developer Access Program for Intel Xeon Phi Processors
Access to Intel Xeon Phi processor-based developer platforms.
With one year of Intel Parallel Studio XE.
Training and support from Colfax.
Order your platform today. http://sdm.link/xeonphi
___
wsjt-devel mailing list
wsjt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/wsjt-devel

Re: [wsjt-devel] FreqCal Mode

[Steven Franke]

Since Joe is probably eating dinner at the moment…

With 55296 point FFT, a 1600 Hz tone (7.5 sa/period) has 55296/7.5=7372.8 
cycles in a record - so there is an endpoint discontinuity (i.e. a 
discontinuity between the first and last points of the record).

A 1500 Hz tone (8 samples per period) has 6912 cycles in a record, so there’s 
no endpoint discontinuity.

I think that what you’re seeing is the result of using a non-windowed FFT. It’s 
easily addressed, if necessary.

Steve


> On Jan 9, 2017, at 11:07 PM, Bill Somerville  wrote:
> 
> On 09/01/2017 22:28, Bill Somerville wrote:
>> ... for sure the three
>> magnitudes going into peakup are much flatter when the error is greater ...
> Hi Joe,
> 
> Here are some examples, these are DFT magnitudes (s vector) for five 
> bins around a 1500Hz tone:
> 
>   0.107814305  0.175234795   9.38052759E-02   17124122.0   
> 8.95743668E-02   8.78201425E-02   4.62990999E-02
>7.97733366E-02  0.128112748  0.307582766   17126434.0  
> 0.167245105   1.00381924E-02   1.45341307E-02
>5.32349795E-02   7.02278912E-02  0.130553111   17125166.0  
> 0.135780737   6.56466857E-02  0.104219861
> 
> and here is the same for a 1600Hz tone:
> 
>79235.5156   191998.594   973516.500   15560843.0   
> 432274.188   128688.836   60769.3555
>79641.3359   192280.516   972233.375   15562208.0   
> 432276.500   128727.242   60738.6406
>79411.3125   192080.391   972825.562   15561087.0   
> 431534.812   128623.734   60596.7227
> 
> These numbers are from print *, s(ipk-2:ipk+2) with DF set to the tone 
> and FTol at 1000Hz.
> 
> 73
> Bill
> G4WJS.
> 
> 
> --
> Developer Access Program for Intel Xeon Phi Processors
> Access to Intel Xeon Phi processor-based developer platforms.
> With one year of Intel Parallel Studio XE.
> Training and support from Colfax.
> Order your platform today. http://sdm.link/xeonphi
> ___
> wsjt-devel mailing list
> wsjt-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/wsjt-devel


--
Developer Access Program for Intel Xeon Phi Processors
Access to Intel Xeon Phi processor-based developer platforms.
With one year of Intel Parallel Studio XE.
Training and support from Colfax.
Order your platform today. http://sdm.link/xeonphi
___
wsjt-devel mailing list
wsjt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/wsjt-devel

Re: [wsjt-devel] FreqCal Mode

[Bill Somerville]

On 09/01/2017 22:28, Bill Somerville wrote:
> ... for sure the three
> magnitudes going into peakup are much flatter when the error is greater ...
Hi Joe,

Here are some examples, these are DFT magnitudes (s vector) for five 
bins around a 1500Hz tone:

   0.107814305  0.175234795   9.38052759E-02   17124122.0   
8.95743668E-02   8.78201425E-02   4.62990999E-02
7.97733366E-02  0.128112748  0.307582766   17126434.0  
0.167245105   1.00381924E-02   1.45341307E-02
5.32349795E-02   7.02278912E-02  0.130553111   17125166.0  
0.135780737   6.56466857E-02  0.104219861

and here is the same for a 1600Hz tone:

79235.5156   191998.594   973516.500   15560843.0   
432274.188   128688.836   60769.3555
79641.3359   192280.516   972233.375   15562208.0   
432276.500   128727.242   60738.6406
79411.3125   192080.391   972825.562   15561087.0   
431534.812   128623.734   60596.7227

These numbers are from print *, s(ipk-2:ipk+2) with DF set to the tone 
and FTol at 1000Hz.

73
Bill
G4WJS.


--
Developer Access Program for Intel Xeon Phi Processors
Access to Intel Xeon Phi processor-based developer platforms.
With one year of Intel Parallel Studio XE.
Training and support from Colfax.
Order your platform today. http://sdm.link/xeonphi
___
wsjt-devel mailing list
wsjt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/wsjt-devel

Re: [wsjt-devel] FreqCal Mode

[Bill Somerville]

On 09/01/2017 21:45, Joe Taylor wrote:
> ... about FreqCal mode in WSJT-X ...

Hi Joe,

I have just been looking at a problem with FeqCal when FTol>DF, I have a 
fix for that but I also noticed that, when I connect another WSJT-X 
instance as a tone generator back to back (digital loopback) with an 
instance running FreqCal, I see slightly broader peaks from the DFT for 
all but a few frequencies. These cause the peakup routine be less 
accurate. For the following tone frequencies, and probably others:

500Hz 750Hz 1000Hz 1250Hz 1500Hz 1750Hz 2000Hz 2250Hz ... 4500Hz

I see accuracy >3DP but most other frequencies I see accuracy to only 
~1.5DP.

I can't work out if the issue not real and is in the synthesis, maybe at 
some integer fractions of 48000 accuracy is better. Or is the issue 
inherent in the DFT or maybe at the peak detection, for sure the three 
magnitudes going into peakup are much flatter when the error is greater.

Perhaps I am expecting too much and 1/10Hz accuracy is all that is 
possible with this sampling period and rate but the comb of more 
accurate values at the tone frequencies above makes me think there may 
be better accuracy possible for any tone.

73
Bill
G4WJS.


--
Developer Access Program for Intel Xeon Phi Processors
Access to Intel Xeon Phi processor-based developer platforms.
With one year of Intel Parallel Studio XE.
Training and support from Colfax.
Order your platform today. http://sdm.link/xeonphi
___
wsjt-devel mailing list
wsjt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/wsjt-devel

Re: [wsjt-devel] FreqCal Mode

[Greg Beam]

Hi Joe,

This is great news. I tried to convert the FMT Fortran files to C++, but 
failed in an epic way, e. This will be a very welcome addition to 
WSJT-X.

Thanks!

Greg.



On 1/9/2017 2:45 PM, Joe Taylor wrote:
> Hi all,
>
> A brief follow-up to my message about FreqCal mode in WSJT-X.  The
> attached screen shot shows what my screen looked like after the program
> has cycled through the 11 test frequencies in my list.
>
> Counting up from the bottom in the waterfall, the frequencies are 660,
> 880, 1210, 2500, 3300, 5000, 7850, 1, 14670, 15000, and 2 kHz.
> The signals at 2500, 3330, and 7850 were too weak to use atthe time of
> this test.  The sequence starts over again after the 20 MHz signal from
> WWV.
>
> -- Joe, K1JT
>
>
> --
> Developer Access Program for Intel Xeon Phi Processors
> Access to Intel Xeon Phi processor-based developer platforms.
> With one year of Intel Parallel Studio XE.
> Training and support from Colfax.
> Order your platform today. http://sdm.link/xeonphi
>
>
>
> ___
> wsjt-devel mailing list
> wsjt-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/wsjt-devel
>

--
Developer Access Program for Intel Xeon Phi Processors
Access to Intel Xeon Phi processor-based developer platforms.
With one year of Intel Parallel Studio XE.
Training and support from Colfax.
Order your platform today. http://sdm.link/xeonphi
___
wsjt-devel mailing list
wsjt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/wsjt-devel

Re: [wsjt-devel] FreqCal Mode

[Joe Taylor]

With WSJT-X code revision r7467 the tools are in place to allow you to 
determine calibration constants (Intercept A and Slope B) for your radio.


Builds using cmake will put the programs fmtave[.exe], fcal[.exe] and 
fmeasure[.exe] in the .../install/bin directory along with wsjtx[.exe] 
and the other executables.


To use this capability you should enter known frequencies of your 
calibration stations under FreqCal mode.  I use these frequencies: 0.660 
0.880 1.210 2.500 3.330 5.000 7.850 10.000 14.670 15.000 20.000 MHz. 
(The first three are clear-channel AM broadcast stations within 100 km 
of my location, and the others are WWV and CHU.)  In other parts of the 
world you will use other calibration stations.


With *Measure* active, select *Frequency calibtation* from the *Tools* 
menu.  The program will cycle through your FreqCal frequencies, 
measuring the effective Dial Error approximately every 3 seconds for a 
total of 30 s and writing the results to file fmt.ave in the WSJT-X log 
directory.


As things stand now (r7467), to complete the calibration procedure you 
need to copy fmt.all to a convenient location and then edit and process 
it as described here:

http://physics.princeton.edu/pulsar/k1jt/FMT_User.pdf

A calibration solution for the FT-2000 at W2PU is shown in the attached 
plot.


-- 73, Joe, K1JT


FT-2000_cal.pdf
Description: Adobe PDF document
--
Developer Access Program for Intel Xeon Phi Processors
Access to Intel Xeon Phi processor-based developer platforms.
With one year of Intel Parallel Studio XE.
Training and support from Colfax.
Order your platform today. http://sdm.link/xeonphi___
wsjt-devel mailing list
wsjt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/wsjt-devel

Re: [wsjt-devel] FreqCal Mode

[Bill Somerville]

On 09/01/2017 13:38, Joe Taylor wrote:
> Thanks for your additions to FreqCal mode committed in r7463.  They are
> just about what I had been planning, but you were one step ahead of me!
> Before long I will provide some instructions for using this mode to
> calibrate one's radio.

Hi Joe,

The frequency list data model has had capabilities for something like 
this since inception so I thought it best to make use of it.

I have just added a few more tweaks. The Rx DF spin box is now 
"real-time" which may help when using a narrow filter to home in on a 
time signal on a noisy band. I have also sorted out some minor issues 
with switching through the various time signals configured.

73
Bill
G4WJS.


--
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
wsjt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/wsjt-devel

[wsjt-devel] FreqCal Mode

[Joe Taylor]

Hi Bill,

Thanks for your additions to FreqCal mode committed in r7463.  They are 
just about what I had been planning, but you were one step ahead of me! 
Before long I will provide some instructions for using this mode to 
calibrate one's radio.

-- Joe, K1JT

--
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
wsjt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/wsjt-devel