Ray Gralak
Vendor (PEMPro)
Reged: 04/19/08

Re: PEPrep help
[Re: korborh]
#5599493  01/01/13 12:20 PM

Edit

Reply

Quote 
Quick Reply



Quote:
Quote:
I think to measure guiding quality you need to take several very short exposures (or measure many stars in a single short exposure) to get an average "shortexposure FWHM". This shortexposure FWHM is the baseline against which you can measure autoguider performance by comparing it to the average FWHM you get in a longer duration autoguided exposure. How little the FWHM increases I think would be the most accurate determination of autoguiding quality. But even in this case seeing can be an unknown factor so it's important to average many stars.
Ray Gralak
The complication is that in short (focus) exposures, the FWHM estimates can be inaccurate as the stars can be quite nonGaussian. Most astrosoftware rely on Gaussian fit to get FWHM and centroid. Also during a longexposure, the local seeing could actually improve compared to the (earlier) short exposure. So I am not so sure if the shortexposure is a good baseline to quantify guiding quality.
Does it really make sense to measure autoguiding with 35 arcsecond/pixel image scale? Of course you want to have an image scale at least 1 arcsec/pixel for doing these types of meanurements.
Ray

cn register 5
scholastic sledgehammer
Reged: 12/26/12

Re: PEPrep help
[Re: Ray Gralak]
#5599501  01/01/13 12:23 PM

Edit

Reply

Quote 
Quick Reply



If the 8/3 term repeats every three cycles then it's separated by 120 degrees in each cycle and averaging batches of three will give zero contribution of the 8/3 fundamental and its harmonics. The actual shape may not be a sine wave but if it's periodic it can be approximated by a series of sine waves with this as the fundamental.
So it seems to me to be best to take multiples of three consecutive PEC cycles for averaging  3 or 6 or 9 or 12. not 5 or 8 or 10. And don't take nonconsecutive cycles.
There will still be noise but noise doesn't have a frequency.
Chris

orlyandico
Postmaster
Reged: 08/10/09
Loc: Singapore

Re: PEPrep help
[Re: cn register 5]
#5599512  01/01/13 12:29 PM

Edit

Reply

Quote 
Quick Reply



imho just use the fundamental....
although some CGEM's (like mine) have a huge 8/3  larger than the fundamental. for these mounts, PEC will have limited usefulness (although PEC would still have usefulness).

korborh
scholastic sledgehammer
Reged: 01/29/11
Loc: Arizona

Re: PEPrep help
[Re: Ray Gralak]
#5599529  01/01/13 12:37 PM

Edit

Reply

Quote 
Quick Reply



Quote:
Does it really make sense to measure autoguiding with 35 arcsecond/pixel image scale? Of course you want to have an image scale at least 1 arcsec/pixel for doing these types of meanurements.
Ray
Huh? How did I imply that? I am not saying the stars are nonGuassian because of under sampling. Individual shortexposures can still be significantly nonGuassian at my pixel scale of 0.5"/px because of dispersion. examples here

Ray Gralak
Vendor (PEMPro)
Reged: 04/19/08

Re: PEPrep help
[Re: korborh]
#5599582  01/01/13 01:03 PM

Edit

Reply

Quote 
Quick Reply



Quote:
Quote:
Does it really make sense to measure autoguiding with 35 arcsecond/pixel image scale? Of course you want to have an image scale at least 1 arcsec/pixel for doing these types of meanurements.
Ray
Huh? How did I imply that? I am not saying the stars are nonGuassian because of under sampling. Individual shortexposures can still be significantly nonGuassian at my pixel scale of 0.5"/px because of dispersion. examples here
And that's why I said you need to average many stars (or pick only "good" stars). Some stars are going to have better shapes than others. The idea is to get a BASELINE performance of the optics with minimal effects from seeing...
Ray

Ray Gralak
Vendor (PEMPro)
Reged: 04/19/08

Re: PEPrep help
[Re: Ray Gralak]
#5599591  01/01/13 01:10 PM

Edit

Reply

Quote 
Quick Reply



Quote:
Quote:
Quote:
Does it really make sense to measure autoguiding with 35 arcsecond/pixel image scale? Of course you want to have an image scale at least 1 arcsec/pixel for doing these types of meanurements.
Ray
Huh? How did I imply that? I am not saying the stars are nonGuassian because of under sampling. Individual shortexposures can still be significantly nonGuassian at my pixel scale of 0.5"/px because of dispersion.
examples here
And that's why I said you need to average many stars (or pick only "good" stars). Some stars are going to have better shapes than others. The idea is to get a BASELINE performance of the optics with minimal effects from seeing...
Ray
An example might help you understand what I mean... using the same scope one could do two long exposure images. One in focus, and the other out of focus. If the autoguider could be left in focus both images would theoretically have the same autoguiding quality however the one out of focus is going to have a much higher FWHM. So, it is important to have a baseline measurement of short exposure FWHM to measure against long exposure FWHM. That is, the higher FWHM is not because of poorer autoguiding. I've seen a few people try to say their autoguiding is better because their FWHM is tighter but that is an apples to oranges comparison. The measurement of autoguider performance in a long exposure should be related to the amount of degradation that occurs compared to a short exposure.
Hopefully this makes things clearer?
Ray
Edited by Ray Gralak (01/01/13 01:17 PM)

orlyandico
Postmaster
Reged: 08/10/09
Loc: Singapore

Re: PEPrep help
[Re: Ray Gralak]
#5599593  01/01/13 01:11 PM

Edit

Reply

Quote 
Quick Reply



i believe Frank has said as much...
compare FWHM of short (5 to 10 second) exposures with FWHM of long guided exposures, this will tell if the FWHM is atmosphere limited or mount limited..

Alph
Carpal Tunnel
Reged: 11/23/06
Loc: Melmac

Re: PEPrep help
[Re: Ray Gralak]
#5599602  01/01/13 01:17 PM

Edit

Reply

Quote 
Quick Reply



Quote:
Having looked at PECPrep's source code I saw that it only uses FFTs, which are only fair approximations. PEMPRo uses FFTs to ballpark the frequencies but then PEMPRo uses much more advanced techniques to get the correct frequencies, amplitudes, and phases. PECPrep's values are probably higher because it is including noise that is removed via PEMPro's more advanced techniques.
As you know every continuous periodic function can be expanded to Fourier series. FFT is a smart and efficient numerical algorithm that decomposes a sequence of values into components of different frequencies thus in essence it expands a periodic error curve (PEC) to Fourier series. This method is well established and reviewed by mathematicians and practitioners. DTF would be more accurate but more time consuming and there would be no practical difference. PEMPro, from what you said in the past, uses proprietary techniques that I am highly skeptical of. Publish your advanced techniques and let specialists review and approve them. In the meantime I will stick to PECPrep and FFT for PEC analysis.

Ray Gralak
Vendor (PEMPro)
Reged: 04/19/08

Re: PEPrep help
[Re: Alph]
#5599615  01/01/13 01:24 PM

Edit

Reply

Quote 
Quick Reply



Quote:
Quote:
Having looked at PECPrep's source code I saw that it only uses FFTs, which are only fair approximations. PEMPRo uses FFTs to ballpark the frequencies but then PEMPRo uses much more advanced techniques to get the correct frequencies, amplitudes, and phases. PECPrep's values are probably higher because it is including noise that is removed via PEMPro's more advanced techniques.
As you know every continuous periodic function can be expanded to Fourier series. FFT is a smart and efficient numerical algorithm that decomposes a sequence of values into components of different frequencies thus in essence it expands a periodic error curve (PEC) to Fourier series. This method is well established and reviewed by mathematicians and practitioners. DTF would be more accurate but more time consuming and there would be no practical difference. PEMPro, from what you said in the past, uses proprietary techniques that I am highly skeptical of. Publish your advanced techniques and let specialists review and approve them. In the meantime I will stick to PECPrep and FFT for PEC analysis.
FFT's are oldschool. The techniques are easily found if you do searches for them but I consider the specific methods PEMPro uses as trade secrets (and there will be advances in PEMPro V3). I can tell you the techniques used are related to singular value decomposition. Many tool vendors, include the PECPrep author has copied PEMPro's ideas so I'm not about to give people more things to copy.
Ray
Edited by Ray Gralak (01/01/13 01:31 PM)

korborh
scholastic sledgehammer
Reged: 01/29/11
Loc: Arizona

Re: PEPrep help
[Re: Ray Gralak]
#5599617  01/01/13 01:25 PM

Edit

Reply

Quote 
Quick Reply



The stars that have better shapes are still not Gaussian. If the program does not calculate correct star FWHM of short exposures, averaging the result from several such nonGaussian stars will not help.
It seems to me that getting BASELINE performance is not as straightforward. Even with video averaging (stacking/culling) and better centroiding, seeing effects can be unknown and significantly different during long exposure vs short.

Ray Gralak
Vendor (PEMPro)
Reged: 04/19/08

Re: PEPrep help
[Re: orlyandico]
#5599623  01/01/13 01:28 PM

Edit

Reply

Quote 
Quick Reply



Quote:
i believe Frank has said as much...
compare FWHM of short (5 to 10 second) exposures with FWHM of long guided exposures, this will tell if the FWHM is atmosphere limited or mount limited..
Frank's the major culprit to whom I was beating this point. I'm glad he finally accepted that.
Ray

korborh
scholastic sledgehammer
Reged: 01/29/11
Loc: Arizona

Re: PEPrep help
[Re: korborh]
#5599640  01/01/13 01:33 PM

Edit

Reply

Quote 
Quick Reply



Ray  thanks for your replies. I am going to do some more experiments to gain a better understanding from the observations on my equipment.

orlyandico
Postmaster
Reged: 08/10/09
Loc: Singapore

Re: PEPrep help
[Re: Ray Gralak]
#5599647  01/01/13 01:39 PM

Edit

Reply

Quote 
Quick Reply



don't quote me on that...
i vaguely remember Frank saying something of the sort but i'm not about to dig through a bazillion forum posts to find his particular statement.
it does make sense from my perspective which is why i mentioned it.

korborh
scholastic sledgehammer
Reged: 01/29/11
Loc: Arizona

Re: PEPrep help
[Re: Alph]
#5599654  01/01/13 01:45 PM

Edit

Reply

Quote 
Quick Reply



Quote:
Quote:
Having looked at PECPrep's source code I saw that it only uses FFTs, which are only fair approximations. PEMPRo uses FFTs to ballpark the frequencies but then PEMPRo uses much more advanced techniques to get the correct frequencies, amplitudes, and phases. PECPrep's values are probably higher because it is including noise that is removed via PEMPro's more advanced techniques.
As you know every continuous periodic function can be expanded to Fourier series. FFT is a smart and efficient numerical algorithm that decomposes a sequence of values into components of different frequencies thus in essence it expands a periodic error curve (PEC) to Fourier series. This method is well established and reviewed by mathematicians and practitioners. DTF would be more accurate but more time consuming and there would be no practical difference. PEMPro, from what you said in the past, uses proprietary techniques that I am highly skeptical of. Publish your advanced techniques and let specialists review and approve them. In the meantime I will stick to PECPrep and FFT for PEC analysis.
I cannot understand how a well established and robust procedure (FFT) is not enough to model the relatively simple undulations of what we see in a mount periodic curve. FFT is used in complex DSP and spectrum analyzers for many hundreds of MHz of frequencies.

Ray Gralak
Vendor (PEMPro)
Reged: 04/19/08

Re: PEPrep help
[Re: Alph]
#5599665  01/01/13 01:53 PM

Edit

Reply

Quote 
Quick Reply



Quote:
Quote:
Having looked at PECPrep's source code I saw that it only uses FFTs, which are only fair approximations. PEMPRo uses FFTs to ballpark the frequencies but then PEMPRo uses much more advanced techniques to get the correct frequencies, amplitudes, and phases. PECPrep's values are probably higher because it is including noise that is removed via PEMPro's more advanced techniques.
As you know every continuous periodic function can be expanded to Fourier series. FFT is a smart and efficient numerical algorithm that decomposes a sequence of values into components of different frequencies thus in essence it expands a periodic error curve (PEC) to Fourier series. This method is well established and reviewed by mathematicians and practitioners. DTF would be more accurate but more time consuming and there would be no practical difference. PEMPro, from what you said in the past, uses proprietary techniques that I am highly skeptical of. Publish your advanced techniques and let specialists review and approve them. In the meantime I will stick to PECPrep and FFT for PEC analysis.
BTW, how did you determine that PECPrep is even accurate? The critical FFT routines are NOT open source. For all you know there could be a bug. The author seems to have a standing policy of letting the users test his code rather than always doing it himself.
Ray

Alph
Carpal Tunnel
Reged: 11/23/06
Loc: Melmac

Re: PEPrep help
[Re: Ray Gralak]
#5599670  01/01/13 01:56 PM

Edit

Reply

Quote 
Quick Reply



Quote:
I'm not about to give people more things to copy.
You can't stop people from using deobfuscators.

Alph
Carpal Tunnel
Reged: 11/23/06
Loc: Melmac

Re: PEPrep help
[Re: Ray Gralak]
#5599730  01/01/13 02:27 PM

Edit

Reply

Quote 
Quick Reply



Quote:
BTW, how did you determine that PECPrep is even accurate? The critical FFT routines are NOT open source. For all you know there could be a bug. The author seems to have a standing policy of letting the users test his code rather than always doing it himself.
The reason I touched on this subject is that PEMPro results in too many misfits. I have seen it on my own and other data. On the other hand I compared many results from PECPrep to Software Bisque's PEC software and they were in a pretty good agreement. Looks like we are digressing from the OP's question so this will be my last comment.

Ray Gralak
Vendor (PEMPro)
Reged: 04/19/08

Re: PEPrep help
[Re: Alph]
#5599769  01/01/13 02:54 PM

Edit

Reply

Quote 
Quick Reply



Quote:
Quote:
I'm not about to give people more things to copy.
You can't stop people from using deobfuscators.
Good luck with that, Adam!
Ray

Ray Gralak
Vendor (PEMPro)
Reged: 04/19/08

Re: PEPrep help
[Re: Alph]
#5599778  01/01/13 02:58 PM

Edit

Reply

Quote 
Quick Reply



Quote:
Quote:
BTW, how did you determine that PECPrep is even accurate? The critical FFT routines are NOT open source. For all you know there could be a bug. The author seems to have a standing policy of letting the users test his code rather than always doing it himself.
The reason I touched on this subject is that PEMPro results in too many misfits. I have seen it on my own and other data. On the other hand I compared many results from PECPrep to Software Bisque's PEC software and they were in a pretty good agreement. Looks like we are digressing from the OP's question so this will be my last comment.
Show me!!
Ray

Ray Gralak
Vendor (PEMPro)
Reged: 04/19/08

Re: PEPrep help
[Re: korborh]
#5599789  01/01/13 03:06 PM

Edit

Reply

Quote 
Quick Reply



Quote:
Quote:
Quote:
Having looked at PECPrep's source code I saw that it only uses FFTs, which are only fair approximations. PEMPRo uses FFTs to ballpark the frequencies but then PEMPRo uses much more advanced techniques to get the correct frequencies, amplitudes, and phases. PECPrep's values are probably higher because it is including noise that is removed via PEMPro's more advanced techniques.
As you know every continuous periodic function can be expanded to Fourier series. FFT is a smart and efficient numerical algorithm that decomposes a sequence of values into components of different frequencies thus in essence it expands a periodic error curve (PEC) to Fourier series. This method is well established and reviewed by mathematicians and practitioners. DTF would be more accurate but more time consuming and there would be no practical difference. PEMPro, from what you said in the past, uses proprietary techniques that I am highly skeptical of. Publish your advanced techniques and let specialists review and approve them. In the meantime I will stick to PECPrep and FFT for PEC analysis.
I cannot understand how a well established and robust procedure (FFT) is not enough to model the relatively simple undulations of what we see in a mount periodic curve. FFT is used in complex DSP and spectrum analyzers for many hundreds of MHz of frequencies.
There's nothing seriously wrong with FFTs but I think if you did some research you would find that there are definitely disadvantages and limitations to FFTs. There are definitely better (i.e more accurate) methods employed today. PEMPro does use FFT's to approximate and identify frequencies but it uses more advanced methods to actually calculate the magnitudes, phases, and removing noise from the signal.
Ray
Edited by Ray Gralak (01/01/13 03:08 PM)
