I increased PHD2 hysteresis. Wow!

[kirkster501]

Been really struggling with guiding recently after moving over to guidescope in anticipation of a dual rig imager, it has been all over the place.   Not had the best of winters to experiment really.  Anyway, it was crap again tonight.  Increasingly depserate, I increased the hysteresis from the default 70 up to 90 on 5.8 sensor and 600mm guidescope.  Wow. Pancake flat.  Will keep my eyes on this before I get too smug prematurely!

Please let this be the fix I have been looking for......!

 

[vlaiv]

You mean aggression rather than hysteresis?

What mount is that? I see both mesu 200 and eq6 in your signature.

[kirkster501]

Correct I meant aggression and yes, a MESU.  Guided with a 600mm FL Celestron guidescope.

[vlaiv]

Mesu should be "stiff" mount and it should benefit from 90 aggression.

Can I make few suggestions?

Star that you are guiding on is clipped - choose one that is not saturating.

Don't let that indicator turn to red.

By your SNR I suspect that you are using ASI174 with native drives? How about switching to ASCOM driver instead. Choose 16bit format. You'll need to redo your darks for that.

Lower min mo for both RA and DEC to 0.10. Switch DEC guide algorithm to hysteresis as well and use 90/10 for both. Experiment with 3,4 and 5s guide exposure lengths (I have sneaky suspicion that 5s will work the best ).

 

[kens]

The Mesu could well benefit from the Z-filter algorithm with a shorter exposure time - especially if you are getting good enough guide images to saturate stars.

The exposure factor can be set so you don't chase the seeing.

[vlaiv]

3 minutes ago, kens said:

The Mesu could well benefit from the Z-filter algorithm with a shorter exposure time - especially if you are getting good enough guide images to saturate stars.

The exposure factor can be set so you don't chase the seeing.

I'm not sure that is the case - at least if I'm understanding what Z-filter algorithm does. It "learns" fast frequency periodic oscillations and adapts? Mesu does not have harmonics at all (or at least it should not have those), so I'm not sure if algorithm that looks for regularities would benefit it.

[kirkster501]

Thanks vlaiv/folks, will check into your suggestions.  How do I stop the Saturation of the stars?   A lower exposure?

[Laurin Dave]

Turn the gain down ..  it’s in the camera control tab 

[kirkster501]

It is so termpermental; I reduce the minmov to 0.1 and it all goes crazy.  I put it back to 0.2 and it remains crazy.  Look at this start of crazy swinging of the DEC......

 

 

[kirkster501]

I find yo have to STOP the guiding. Let things settle down and THEN start again to get teh graph flat again.

[RolandKol]

It looks like strongly overcorrecting...

Try reducing guide pulse down to 1000...

And maybe enable the time lapse option adding 2 or even 5sec delay

[vlaiv]

I would say that such positive feedback has something to do with DEC aggressiveness being set to 100 and calibration.

If there is even small calibration error - and that would be perfectly normal, you can think of it as calibration SNR - there will always be some error due to measurement, and you set aggressiveness to 100 - you might actually give too much correction. Due to calibration error, PHD2 thinks that it needs to counter 0.2" with pulse correcting it by 0.5" - mount promptly responds and places star to 0.3" on the other side of graph - phd2 sees that, tries to correct with 0.7" pulse - which brings star back to first side of graph and position of 0.4" (0.3 - 0.7 = -0.4) and cycle continues with even wider oscillations.

This is why it is good to have aggressiveness lower than 100% - you will dampen down any sort of positive feedback that might arise from calibration error.

Combating star saturation is easy - just choose fainter star to guide with . There are other things that one can do - use lower gain (requires redo of darks), but also using ASCOM driver instead of native and going for 16 bit helps with this quite a bit, since there is larger dynamic range in guide images and stars won't saturate so easily. This last one also helps with SNR - which is good for star centroid precision.

[kens]

1 hour ago, vlaiv said:

I'm not sure that is the case - at least if I'm understanding what Z-filter algorithm does. It "learns" fast frequency periodic oscillations and adapts? Mesu does not have harmonics at all (or at least it should not have those), so I'm not sure if algorithm that looks for regularities would benefit it.

I know how it works - I wrote it  It is a digital Bessel filter with a tuneable cutoff frequency.  Unlike the other algorithms it has flat response in the pass band and no nodes in the stop band and being a Bessel filter has linear phase response. The filter is fourth order so approximately 24dB per octave suppression above the cutoff.

It is the ability to tune the cutoff frequency as a function of exposure time that makes it useful. By allowing short exposure times you can reduce latency but still filter out the high frequency seeing-related deviations.

[vlaiv]

5 minutes ago, kens said:

I know how it works - I wrote it  It is a digital Bessel filter with a tuneable cutoff frequency.  Unlike the other algorithms it has flat response in the pass band and no nodes in the stop band and being a Bessel filter has linear phase response. The filter is fourth order so approximately 24dB per octave suppression above the cutoff.

It is the ability to tune the cutoff frequency as a function of exposure time that makes it useful. By allowing short exposure times you can reduce latency but still filter out the high frequency seeing-related deviations.

Ok, so it's not what I thought it is, and it indeed sounds interesting. I'll actually give it a go next time I'm out.

This basically means that it's working the best with rather short exposures - like 0.5s-1.0s?

How does it deal with wind - being sharp jump, I'm suspecting that frequency response will be spread out for such signal?

Edited February 21, 2019 by vlaiv

[kens]

It does work best with short exposures and those provide more scope for tuning. At longer exposures it still works well but offers no specific advantage over the default algorithms. It is designed for mounts that respond well to frequent, small corrections. I think the Mesu would qualify there. Large deviations due to wind will cause a relatively slow response but it will be smooth. In some cases it responds better because there is no ringing. Slow response in Dec is most noticeable as I normally recommend a low frequency response (typically a factor 8 of exposure time) This minimises or even eliminates dec reversals but step changes take longer to recover.

[kirkster501]

Thanks guys, definite progress tonight and some great tips.  Many thanks.  Feel I made some progress with this.

[kirkster501]

9 hours ago, vlaiv said:

Combating star saturation is easy - just choose fainter star to guide with . There are other things that one can do - use lower gain (requires redo of darks), but also using ASCOM driver instead of native and going for 16 bit helps with this quite a bit, since there is larger dynamic range in guide images and stars won't saturate so easily. This last one also helps with SNR - which is good for star centroid precision.

Thanks Vlaiv, I have done this and using the ASCOM driver now.  I have also set the DEC to Hysteresis and with 90/10 just like in RA.  It was getting late so did not have time to really test this out but it was looking promising.

I will look into the Z-filter algorithm suggested by kens too.

I'd be interested to know what other MESU users have their guiding parameters set to.

Edited February 22, 2019 by kirkster501

[RolandKol]

9 hours ago, kens said:

I know how it works - I wrote it  It is a digital Bessel filter with a tuneable cutoff frequency.  Unlike the other algorithms it has flat response in the pass band and no nodes in the stop band and being a Bessel filter has linear phase response. The filter is fourth order so approximately 24dB per octave suppression above the cutoff.

It is the ability to tune the cutoff frequency as a function of exposure time that makes it useful. By allowing short exposure times you can reduce latency but still filter out the high frequency seeing-related deviations.

Sounds like adaptive guiding algorithm to me  I have completely forgot PHD has different options to choose

Will test my NEQ6 on your "Z-ombie" algo tonight if weather allows  as my mount went mad and does not want to guide properly on 3sec, - however on 1sec it works quite good, - around 0.55 Total RMS.

I doubt it can perform any better, but.... "curiosity killed the cat"

by the way, - why Z?

[kens]

For continuous waveforms a Fourier transform is used to convert from time domain to frequency domain. For a discrete, that is sampled, waveform the equivalent transform is known as the Z transform and the discretised frequency domain as the Z domain. The Z filter algorithm is designed using the Z transform and operates in the Z domain. It is the digital equivalent of capacitors and inductors used to filter audio signals so I would not characterise it as adaptive. Bessel and Butterworth filters are often used in audio systems e.g for crossovers due to their characteristics. These same characteristics are useful for guiding.

[kirkster501]

2 hours ago, kens said:

For continuous waveforms a Fourier transform is used to convert from time domain to frequency domain. For a discrete, that is sampled, waveform the equivalent transform is known as the Z transform and the discretised frequency domain as the Z domain. The Z filter algorithm is designed using the Z transform and operates in the Z domain. It is the digital equivalent of capacitors and inductors used to filter audio signals so I would not characterise it as adaptive. Bessel and Butterworth filters are often used in audio systems e.g for crossovers due to their characteristics. These same characteristics are useful for guiding.

You took the words straight out of my mouth kens ?

[kirkster501]

Miraculous, another clear night here.....

Set to ASCOM 16 bit driver with maximum dyanmic range.  Aggression  80 on both axis and hysteresis algorithm on both.  It's looking good........  Seeing is rubbish and blowing out Alnitak.   May try the Z algorithm.

[vlaiv]

Just for fun, why don't you try 5-6s exposures? It should really smooth seeing effects out. Maybe do comparison with Z-thingy algorithm, but we won't be able to tell the difference from guide graph - we would need to measure FWHM in corresponding subs to see real advantage of one over another.

[Stub Mandrel]

21 hours ago, kirkster501 said:

How do I stop the Saturation of the stars?   A lower exposure?

Easiest way is to auto-select star or choose one that is less bright.

[Stuart1971]

On 21/02/2019 at 22:59, vlaiv said:

I would say that such positive feedback has something to do with DEC aggressiveness being set to 100 and calibration.

If there is even small calibration error - and that would be perfectly normal, you can think of it as calibration SNR - there will always be some error due to measurement, and you set aggressiveness to 100 - you might actually give too much correction. Due to calibration error, PHD2 thinks that it needs to counter 0.2" with pulse correcting it by 0.5" - mount promptly responds and places star to 0.3" on the other side of graph - phd2 sees that, tries to correct with 0.7" pulse - which brings star back to first side of graph and position of 0.4" (0.3 - 0.7 = -0.4) and cycle continues with even wider oscillations.

This is why it is good to have aggressiveness lower than 100% - you will dampen down any sort of positive feedback that might arise from calibration error.

Combating star saturation is easy - just choose fainter star to guide with . There are other things that one can do - use lower gain (requires redo of darks), but also using ASCOM driver instead of native and going for 16 bit helps with this quite a bit, since there is larger dynamic range in guide images and stars won't saturate so easily. This last one also helps with SNR - which is good for star centroid precision.

I know this is an old thread, but can I ask why the ASCOM driver is better than native with the ASI174M,??  I have just got the mini version of this camera, and found the star that PHD chose was already saturated, or rather had a very flat top on the graph, yet the saturated label stayed green….🤔🤔 I had to bring gain setting in PHD down to 15 to get a peak on the star, and because of this there were barely any visible stars.

ALSO there is an option in the native ASI 174m PHD native driver to choose 16 bit, so is this not enough, is the ASCOM driver better in some way…?

oh, and is it really needed to use different darks when altering gain in PHD, if so why…??

Edited February 25, 2022 by Stuart1971

[vlaiv]

2 hours ago, Stuart1971 said:

know this is an old thread, but can I ask why the ASCOM driver is better than native with the ASI174M,??  I have just got the mini version of this camera, and found the star that PHD chose was already saturated, or rather had a very flat top on the graph, yet the saturated label stayed green….🤔🤔 I had to bring gain setting in PHD down to 15 to get a peak on the star, and because of this there were barely any visible stars.

ALSO there is an option in the native ASI 174m PHD native driver to choose 16 bit, so is this not enough, is the ASCOM driver better in some way…?

For many cameras, native driver is optimized for high FPS and SNR in single exposure is not as good as with ASCOM.

Guiding is similar to imaging - you want the best data you can have and ASCOM is preferred method for long exposure imaging.

Another thing is that 8bit seems to be sufficient for guiding, but I've found that 16bit gives better results and it is much easier to get 16bit data from ASCOM (I can't choose 8/16 bit with native drivers in PHD2 on my camera).

2 hours ago, Stuart1971 said:

oh, and is it really needed to use different darks when altering gain in PHD, if so why…??

Because level of bias / dark signal depends on gain setting. If one new exact e/ADU values for different gain settings - one could scale one dark to match the gain of another - but it is far simpler to do another set of darks and don't bother with all that gain stuff.