Oversample or undersample?

[michaelmorris]

I'm starting the process if considering whether to purchase a CCD camera or not. I'd like to use it with a range of scopes from 1400 mm to 410 mm focal length. I realise that using a single camera on this range of scopes is going to lead to some oversampling or under sampling (or possibly a bit of both) at the extremes of this range.
From what I've read so far it seems to me that it's better to oversample than undersample - is this correct?
What are the downsides of oversampling? Is it only a matter of potential guiding issues?
Thanks

[ollypenrice]

In long exposure DS imaging I can't see much point in oversampling with the amateur kit most of us use. OK, you can drizzle provided you dither guide and, maybe, tease out a bit more resolution. However, the price you pay is having less than the ideal amount of light per pixel. The 'real world' problem in amateur AP is getting enough signal so I prefer to be somewhat undersampled and so have plenty of light on each pixel. At the moment I work at 3.5 arcsecs per pixel in the Taks and 1.8 in the TEC140. Sure, 3.5 is far from ideal but it's the price you pay for covering the best part of 3x4 degrees of sky on the big Kodak. And it doesn't look that bad, I don't think! How many APODS has the Tak 106/Kodak 11 meg rung up? It's a lot...

Assuming you go for mono you do have the option of binning but choose your camera carefully. The SXVH36 I use here, for instance, will not bin cleanly despite some adjusting and I've seen this in other cameras too. I have never seen any manufacturer or supplier warn about this issue but, believe me, I've seen it on several cameras.

Olly

[Uranium235]

Im with Olly on the undersampling, but it depends on what you are imaging. For nebulae its fine, since I like to go for the overall impact of the bigger picture, but for galaxies its better to get a bit more resolution (1-2" p/p is good enough). The Star71 + 383 I have gives 3.2", but overall its quicker since Im covering a lot more sky with one frame (less need for mosaics) and allows the capture of multiple targets or proper framing of the bigger stuff.

[ollypenrice]

Im with Olly on the undersampling, but it depends on what you are imaging. For nebulae its fine, since I like to go for the overall impact of the bigger picture, but for galaxies its better to get a bit more resolution (1-2" p/p is good enough). The Star71 + 383 I have gives 3.2", but overall its quicker since Im covering a lot more sky with one frame (less need for mosaics) and allows the capture of multiple targets or proper framing of the bigger stuff.

Agreed. Even in the galaxies, though, there are sometimes faint extensions which can take an age to collect. The full M31 outer regions, the Leo tidal tail, the Antennae interactions, etc.

What I should have included in my original post is the fact that, if you have two focal lengths with the short one undersampled to its disadvantage, you can do a quite quick shoot in luminance only with the long one and this will tickle up key details. I use this technique very often. It's surprising how little long FL data you need to make a difference when you are downsizing it to match the short FL image - and it's rare for the whole of an image to need fine detail.

I use Registar to make the long FL fit perfectly then blend it gently in Ps for a seamless result. (Well, I try to make it seamless, that is!  )

Olly

[opticalpath]

Bear in mind that the theoretical 'optimal' sampling rate isn't a fixed value; it's expressed in terms of FWHM. So in practice the value will change quite a bit depending on seeing conditions.  Under-sampling on a good night could become over-sampling on a bad one with the same setup.

Adrian

[andrew s]

Just a thought from a non imager. You can always bin to increase the effective size of your pixels but you can split them. Maybe this can give you more options.

Regards Andrew 

[Astroboffin]

Very interesting thread

I have a William Optics ZS66 (312mm FL with the WO 0.8x FF/R, and 389mm without) And an Atik 414ex mono camera with 6.45 micron pixels,

This gives me approx 3.4" p/p without the FF/R and 4.2" p/p with it, so is this kit not matched very well, also I am going to use the FF/R most if not all of the time, so will be imaging at 4.2" p/p all the time ??

I was told that this camera and scope combo would be good, now am not so sure

AB

[Astroboffin]

Obviously binning is not an option for me with the above figures ! As I would then assume I would be imaging at 6.8" p/p and 8.4" p/p respectively, have I got that correct ?

AB

[ollypenrice]

Very interesting thread

I have a William Optics ZS66 (312mm FL with the WO 0.8x FF/R, and 389mm without) And an Atik 414ex mono camera with 6.45 micron pixels,

This gives me approx 3.4" p/p without the FF/R and 4.2" p/p with it, so is this kit not matched very well, also I am going to use the FF/R most if not all of the time, so will be imaging at 4.2" p/p all the time ??

I was told that this camera and scope combo would be good, now am not so sure

AB

With this focal length you'll be imaging nebulae, clusters and, perhaps, groups of objects. It will not often be all that important to capture fine resolution on such target so I think you'll be fine. I post lots of images at 3.5""P/P.

It would be nice to have smaller pixels but, in this chip size, it can't yet be done. And you can't tell a widefield junkie to use a smaller chip!     

Olly

Edit. By the way, the best amateur astrophoto I've ever seen was taken at 3.5"PP. The imager in question has not yet realeased the final version but it is utterly devastating.

[Astroboffin]

With this focal length you'll be imaging nebulae, clusters and, perhaps, groups of objects. It will not often be all that important to capture fine resolution on such target so I think you'll be fine. I post lots of images at 3.5""P/P.

It would be nice to have smaller pixels but, in this chip size, it can't yet be done. And you can't tell a widefield junkie to use a smaller chip!     

Olly

Edit. By the way, the best amateur astrophoto I've ever seen was taken at 3.5"PP. The imager in question has not yet realeased the final version but it is utterly devastating.

WOW

If my images come out half as good as any of those then I would be more than happy

So even at 4.2" p/p do you think I would get away with it, so I can use my FF/R and just leave it on the scope ?

Thanks Olly

AB

[ollypenrice]

WOW

If my images come out half as good as any of those then I would be more than happy

So even at 4.2" p/p do you think I would get away with it, so I can use my FF/R and just leave it on the scope ?

Thanks Olly

AB

Although I have a WO ZS66 with dedicated FF/FR, and began imaging with it as my first scope, I can't remember the size of the image circle without the FF/FR so I don't know if you'd be able to get clean stars to the edge without it. I'd suspect not. Be sure to get the correct chip distance (reducer to chip) and then give it a try. Don't panic about the 4.2"PP. Just take some pictures and learn the ropes.

Olly

[Uranium235]

Just a thought from a non imager. You can always bin to increase the effective size of your pixels but you can split them. Maybe this can give you more options.

Regards Andrew 

Already done that mate. When using the 130pds and Atik 383L+ binned it yields ~3.4" p/p, which is still under the 3.5 limit I set for myself. You wont be increasing the amount of photons collected, but its does give a nice boost to the S/N ratio (depending on your chip type). However, its only worth doing if you have a high pixel count to start with, otherwise the image scale will suffer.

Example of a binned image (2x2 mosaic):

[Pompey Monkey]

My limited (by the weather!) imaging has all been at 1.65" p/p and it looks pretty good.

Maybe I'll try some 2X2 binning when I get some clear skies to see how it looks - At least my stars will be round

[ollypenrice]

My limited (by the weather!) imaging has all been at 1.65" p/p and it looks pretty good.

Flaming Star 2.jpg

Maybe I'll try some 2X2 binning when I get some clear skies to see how it looks - At least my stars will be round

If I had to choose one single pixel scale for my site it might well be that for most targets. It certainly sits in the sweet spot.

Olly

[pete_l]

Personally I wouldn't get too hung up about it. People produce highly acceptable images with a wide range of arc-sec per pixel values. Aim for something close to an average-ish resolution for the focal length of your most used scope and you'll still get pretty good images on either side of that focal length.

Remember, over time this will not be your only CCD, only your FIRST.

[Pompey Monkey]

Personally I wouldn't get too hung up about it. People produce highly acceptable images with a wide range of arc-sec per pixel values. Aim for something close to an average-ish resolution for the focal length of your most used scope and you'll still get pretty good images on either side of that focal length.

Remember, over time this will not be your only CCD, only your FIRST.

The way the weather's going at the moment, my SBIG will not only be my first, it will most definitely be my last CCD  

[ollypenrice]

The way the weather's going at the moment, my SBIG will not only be my first, it will most definitely be my last CCD

These long gloomy spells do descend on occasion. Ne panic pas!

Olly

[Uranium235]

Sorry... couldnt resist!

[opticalpath]

If anyone is interested, there is a very good article by Herbert Raab about detecting faint objects, SN ratio etc.

http://www.astrometrica.at/papers/PointSources.pdf

It contains an interesting section on pixel size and sampling.

He concludes that (for point objects), theoretical critical sampling for maximum SN ratio and deepest limiting magnitude is when pixel size equals 1.2 x FWHM.  However at that scale, when most of the light from a star falls on one pixel, some information about the position and shape of the object is lost; for practical purposes a scale should be chosen 'so that the FWHM of stellar sources spans at least 1.5 to 2 pixels'.

I know this is just theory, and probably doesn't apply exactly to extended objects like nebula, but it does emphasise that the 'ideal' sampling ratio is not a fixed  'xx arcsec per pixel', but depends on FWHM - a variable affected by focusing, aberrations, seeing etc.

Adrian

[pete_l]

The way the weather's going at the moment, my SBIG will not only be my first, it will most definitely be my last CCD

On the bright side: you won't wear it out quickly. And if it's permanently cloudy, then the arc-second resolution is largely immaterial.

Remember: a Maserati travels just as fast as a Trabant, in a traffic jam. So it is with astronomy. When it's cloudy, a supermarket "special" is as good as a Planewave.

[ollypenrice]

Are there really any point sources to amateur optics?

Olly

[michael8554]

Here's an article that tries to give real-life results when Seeing, FWHM, and Nyquist sampling are combined:

http://www.astro-imaging.com/Tutorial/MatchingCCD.html

For my setup, with seeing 3 arcsecs, and focal length 1280mm, the calculated pixel size is 5.45um.

My 450D just happens to be 5.2um pixels, and a SPC900 webcam is 5.6um

Michael

[opticalpath]

Are there really any point sources to amateur optics?

Olly

I think the premise is that stars are point sources, but by the time their light has passed through our atmosphere and our optics, a star's light is no longer a point but a smeared out point spread function (PSF).  Raab's article acknowledges that and goes on to consider optimal sampling based on how wide the spread is - i.e FWHM - so it is real-world to that extent.  He then goes on to say that optimal sampling for best SNR is not optimal for reproducing accurate star shape or positions and proposes 1.5 to 2.0 times higher sampling than that for practical purposes.

Rob Kantleburg's article actually argues along similar lines, except that he expresses the smearing of the point source in terms of actual Airy disk diameter, plus seeing effects, and compares that to pixel size in microns, whereas Raab expresses it in a dimensionless way: 'FWHM should cover 1.5 to 2 pixels'.

The answers they come up with are not that different except that Rob applies a higher multiplier to convert 'theoretical best' to 'practical best'.  So he reckons that a pixel size of about 6 microns is optimal for his scope (1500 mm FL) under conditions when seeing is 3 arcsec.  Personally, I find that too high and feel that Raab's figure is closer to my experience: about 1.1 arc sec per pixel under good seeing conditions (FWHM ~ 2") and (binned) 2.2 arcsec per pixel when FWHM is bigger.

Adrian

[pete_l]

So he reckons that a pixel size of about 6 microns is optimal for his scope (1500 mm FL) under conditions when seeing is 3 arcsec.  Personally, I find that too high and feel that Raab's figure is closer to my experience: about 1.1 arc sec per pixel under good seeing conditions (FWHM ~ 2") and (binned) 2.2 arcsec per pixel when FWHM is bigger.

Adrian

Hmmm.

I followed the line of thought in that article. It sounds good on paper, but I'm not sure that it lines up with empirical evidence.

First of all, we need a sensitivity analysis. Although the article gives a theoretical optimum (for the conditions quoted in the piece), we should look at how quickly the "best" results drop away when any of the parameters are changed. To do that, we'd have to take into account the quality of the image - something that the article doesn't go in to as it's far too dependent on a large number of factors: the optics, the CCD noise  and Q.E. and the degree / skill of image processing.

The infamous (and often wrong ) "gut feeling" is that with a notional star size, at the CCD plane in the 15-20µ range then the difference between sensing that on a 5µ pixel and a 9µ won't be that much (5µ is slightly oversampled, 9 - slightly undersampled), especially when you take into account the P.E. of the mount and the post-processing / stacking that is done.

[opticalpath]

Hmmm.

I followed the line of thought in that article. It sounds good on paper, but I'm not sure that it lines up with empirical evidence.

First of all, we need a sensitivity analysis. Although the article gives a theoretical optimum (for the conditions quoted in the piece), we should look at how quickly the "best" results drop away when any of the parameters are changed. To do that, we'd have to take into account the quality of the image - something that the article doesn't go in to as it's far too dependent on a large number of factors: the optics, the CCD noise  and Q.E. and the degree / skill of image processing.

The infamous (and often wrong ) "gut feeling" is that with a notional star size, at the CCD plane in the 15-20µ range then the difference between sensing that on a 5µ pixel and a 9µ won't be that much (5µ is slightly oversampled, 9 - slightly undersampled), especially when you take into account the P.E. of the mount and the post-processing / stacking that is done.

That's why I prefer Herbert Raab's analysis. Because there are so many complicating factors such as seeing and tracking accuracy, the theoretical Airy disk size in microns is not that useful; the real-world size of actual star images on the night, as judged by FWHM already includes the effects of those variables and is a better basis for deciding optimal sampling, imo. 

Rob Kantelberg concludes that pixel size of about 6 microns provides optimal sampling for his scope (1500mm focal length) on nights when seeing is 3 arcsec, borne out by his experience ... which I can't argue with.  But that works out at 0.8 arcsec per pixel.  In my experience, if FWHM is 3 arcsec, I would be binning in order to image at 2.2 arcsec per pixel, not at less than 1 arcsec per pixel. For my scope and conditions, I've found the guideline in Raab's article nearer the mark.

In any case, I think the main point is that the optimal sampling rate is not a fixed number for a particular scope; it is directly related to FWHM and so varies considerably depending on conditions, accuracy of focus, tracking etc.  So we should not get hung up on a single number, but rather think about a pixel size for our scope that would make the best of ideal conditions, and be ready to bin to match larger FWHM in poorer conditions.

Adrian