The effect of Star Centroid Estimation Errors on Guiding

[MarkS]

In an effort to understand some of the limitations of guiding I wrote a simulation last night to see just how accurately the guiding software can estimate the centroid of a (non-saturated) guide star.

As an example, if you are guiding on a star with FWHM (full width half max) of 1.5 pixels and a SNR (signal to noise ratio) of 20 then the software will frequently (25% of the time) make errors of 0.3 pixels or more in the estimation of the centroid position.  This is simply due to the presence of random noise in the pixel values representing the star.  If you are guiding at 10arcsec/pixel (Ivor!) then that translates into frequent positional estimation errors of 3 arcsecs.  The effect will be to cause the guiding software to make unnecessary corrections and of course makes guiding less accurate overall. 

Now I finally understand something that has puzzled me for a long time - why guiding is continually making corrections in the Dec direction.

Also it shows that the careful selection of guide star really does make a big difference.

I intend to produce a table of expected RMS centroid position errors for different values of guide star FWHM and SNR.  In general, the larger the FWHM and SNR the better the guiding will be.

Watch this space!

Mark

[Ivor]

Well that explains the graph we were seeing on Saturday and confirms our initial conclusions.

On my internet travels to find a solution to my guiding issues I read an article arguing the case for guiding SW to move to guiding off multiple stars to further improve SNR which I thought made sense, you would think today's PCs could cope with the extra resource loading this creates.

[MarkS]

I've been arguing the same for quite some time (i.e. multi-star guiding).  As far as I can see (from the PHD source code) the overhead would not be too great and it would dramatically improve guiding when guide star SNR is low or when the seeing is the "slow wave" variety.

Trouble is, adapting the code for multi-star guiding would be just another entry on my long list of things I don't have time for

Mark

[Mac]

A Few spanners questions to throw in.

Is the problem due to the actual calculation of the exact centroid?
If so, would the accuracy be improved if the star was larger? if it was de-focused,
i.e. a larger circle, which would be easier to calculate the centroid accurately ,
would it not also smooth out the errors caused by scintillation,

Just a spanner to think about,

Mac.

[Mike]

You can guide on multiple stars with MaximDL (with a plugin) and also with MetaGuide.

[MarkS]

if it was de-focused,
i.e. a larger circle, which would be easier to calculate the centroid accurately ,
would it not also smooth out the errors caused by scintillation,

Yes defocusing the star (when its undersampled) is a well recognised way of improving guiding.  It helps because the centroid estimation is less affected by the pixel noise. The FHWM of the (defocused) guide star needs to be around 2 pixels I think.

I'm not sure what effect defocusing has on smoothing scintillation.

[MarkS]

You can guide on multiple stars with MaximDL (with a plugin) and also with MetaGuide.

Interesting ...