New PHD2 Z-filter algorithm

[kens]

This follows on from 

 

In the latest PHD2 dev release 2.6.5dev4 the algorithm has been tweaked to simplify its configuration. The former "Order" has been fixed at 4 as testing showed that to be the optimal and other values made little difference anyway. The "Corner" value concept has been altered to an "Exposure Factor" to make it simpler to understand. Multiplying your guiding exposure by the Exposure factor  gives a "virtual exposure time" where the guiding response is similar to choosing that exposure time with a hysteresis algo. Finer control for the exposure factor has been added and it can be adjusted directly on the graph.

So lets say you normally use 3 second exposures with Hysteresis algo: you could instead use a 1 second exposure with Zfilter using an exposure factor of 3x. This has the advantage of smaller corrections with less lag time. If the guide star brightness drops you can increase the exposure time to, say, 2 seconds and reduce the exposure factor to 1.5x

In most cases, a MinMo of 0.0 is recommended. If you are guiding with OAG at small pixel scale or have severe backlash a small MinMo may be helpful but in general, increasing the exposure factor works better.

Using the on-graph control you can see how the algorithm works to smooth out the guiding as it is adjusted. If you choose the "Corrections to scale" option in the graph menu you can see the size of corrections in relation to the movement of the guide star.

The algorithm is still a beta release so any feedback is welcome

 

[wornish]

PHD2 is great and I use it all the time,  but. ....   I wish I understood a word of this post.  

 

Sorry.

[Andyb90]

I'll definitely be trying this. My guide exposures are short and seeing tends to be average at best.

For 0.5 second exposures should I start with say exposure factor of between 8 to 16?

Regards,

Andy

[kens]

I'd start with a factor of 4 to 8 on RA and 6 to 12 on Dec. The algo is specifically meant to be an improvement on the LowPass algos recommended for the Avalon mounts so it should work well for you.

[kens]

35 minutes ago, wornish said:

PHD2 is great and I use it all the time,  but. ....   I wish I understood a word of this post.  

 

Sorry.

To help explain (maybe).... The current guiding algorithms used in PHD2 rely heavily on setting the guiding exposure time to smooth out the seeing. But that means the corrections to the mount are delayed accordingly and don't correct as well as they should. With this algorithm you can use faster exposures and use the exposure factor parameter to smooth out the seeing. But the corrections are sent more quickly and are smaller so they work better.

The main caution is that when you use longer guide exposures your graph looks smoother. But with this algorithm and shorter exposures your graph looks like a mess because it shows up the seeing. You need to look at the corrections to see how smoothly the algorithm it is working. Or better still, look at the stars in your images

[michael8554]

I've had good results with the original version at 0.75 seconds exposure and 0.0 Min Move, on RA only.

Those fast exposures may have ramifications on your Dec guiding, particularly if you have not-so-good Dec backlash.

Good PA, with a slight drift in one direction, the scope balanced slightly heavy in that direction too, and then the mount should only need guiding in one direction, but you can leave it on Auto if you Dither.

Looking forwards to trying dev 4.

Michael

[asterisme32]

Hello,
I obtained very good results with the original version of Zfilter with an exposure of 0.5 seconds, a min move 0,0 and a factor 4 or 8 (really superior to hyteresis or PEC); but with order 2.
Now with order 4 my results have similar to other algorithms.
Is it possible to give a choice between order 2 and 4?
Bernard.

Mount = Celestron AVX

[kens]

Hi Bernard

It would be useful if you provide some example guide logs. During testing we found that the order did not make much difference so to keep things simple I fixed it at 4. Note that with the new setup that the old corner factor 4 or 8 is now an exposure factor of 1 or 2.

[asterisme32]

8 hours ago, kens said:

Hi Bernard

It would be useful if you provide some example guide logs. During testing we found that the order did not make much difference so to keep things simple I fixed it at 4. Note that with the new setup that the old corner factor 4 or 8 is now an exposure factor of 1 or 2.

Thank you for your reply.
I will do tests with guide log as soon as weather allows.

Which parameters do you put in the log guide, are the graphics saved?
Bernard.

[AKB]

On 07/09/2018 at 22:49, kens said:

The algo is specifically meant to be an improvement on the LowPass algos recommended for the Avalon mounts so it should work well for you.

I’ll definitely give this a go, then.  I’ve often wondered about the delay involved in longer exposures.   This makes a lot of sense in theory.  Let’s see how it turns out in practice!

[kens]

2 hours ago, asterisme32 said:

Thank you for your reply.
I will do tests with guide log as soon as weather allows.

Which parameters do you put in the log guide, are the graphics saved?
Bernard.

Do one run with order 1 or 2 and another with order 4. Keep all other parameters the same. If you can also provide a sub taken from each run that would also be helpful. With good guiding the graph really only shows the seeing. So what really matters is small round stars in your subs.

[kamayok3]

If my guide exposure is 1.5sec, meaning I have to set 0.5sec exposure and 3 for exposure factor? Do I need to set it in DEC as well with the same value as RA? Thanks

[kens]

You can set different values for RA and Dec. Its often beneficial to use a longer value on the Dec axis

[asterisme32]

On 9/18/2018 at 11:21 AM, kens said:

Do one run with order 1 or 2 and another with order 4. Keep all other parameters the same. If you can also provide a sub taken from each run that would also be helpful. With good guiding the graph really only shows the seeing. So what really matters is small round stars in your subs.

Hello Kens,
I tried to do some testing. I attach the logs

PHD2_DebugLog_2018-09-27_201705_expfact1.txt

PHD2_DebugLog_2018-09-27_202043_expfact2.txt

PHD2_DebugLog_2018-09-27_202400_expfact4.txt

PHD2_DebugLog_2018-09-27_204022_hysteresis.txt

PHD2_GuideLog_2018-09-27_201705_expfact1.txt

PHD2_GuideLog_2018-09-27_202043_expfact2.txt

PHD2_GuideLog_2018-09-27_202400_expfact4.txt

PHD2_GuideLog_2018-09-27_204022_hysteresis.txt

[kens]

I'm a little confused as to your objectives. On the one hand the algorithm is working just as I would expect it to.  But if you wanted to test with the now unavailable Order variable you need to revert to the previous dev release.

You'll also need to record longer runs to check that the PE is being guided out. There looks to be quite variable seeing in these logs as illustrated by my attachment. The top trace is with the Z-filter and the bottom one is Hysteresis.

In terms of RMS the bottom one looks much better but you can see that the corrections are closely following the guide star deviations. On the top trace the guide corrections are very much smaller despite the larger deviations. Clearly the deviations are not due to the corrections so the large deviations are due to poor seeing and/or a bad spot in your mount's gearing.

[asterisme32]

Thank you for the comments. I do not doubt that my mount has gears problems, but I do not nose disassembled, I may go see a specialist.
Bernard

[Andyb90]

Just to check should I still download v2.6.5dev4 to try the Z Filter? Or should it be a later dev version? I can see up to dev7 on the website.

Also do I need to un-install my current PHD2 version first or can I install the dev version alongside it?

Andy.

[Dr_Ju_ju]

Get the latest & it will install over what you have....

[jmel]

I have pretty bad seeing most nights, so I typically do 2 second exposures with my ssag. I tried z filter on both Ra and Dec and didn't really have much luck over my standard settings...

I did .5 exposures, 1 sec... .75... And tried 4x factor, 6x, 8x... Still seemed will over the place...

[kens]

Shorter exposures with bad seeing will always give bad looking graphs but you need to look at the end result to see if it is better or not. If you look at the corrections you can see see how Z-filter does not chase the seeing and instead gives a smooth correction.

The normal use case for Z-filter is with a mount that responds well to small, frequent corrections.

[cuivenion]

Is similar to how Metaguide works?

[kens]

I don't think so. Metaguide has a unique guiding algorithm and centroid calculation. It is not open source so I can't comment beyond that. Z-filter uses the normal PHD2 centroid calculation as input. It differs from the default hysteresis algorithm in that it is designed (and I mean designed) to give flat response in the pass band with a sharp cutoff, minimal lobes in the stop band and linear phase. It is a digital version of a Bessel filter. And the cutoff is adjustable so it can be adapted for different exposure times to give a consistent response

[Andyb90]

Just to check if I install the latest Dev release will my profiles and settings be retained?

[kens]

Yes they will.