What is your real-world experience with oversampling?

It really depends on what your goals are. Cameras and software keep evolving so the advice that made sense even 5 or 6 years ago might not be 100% the case now. If you want to go as deep as possible with wide fields and long subs, I don't think oversampling will buy you anything. If you're more interested in going for better resolution then I think there's some wiggle room.

I don't have enough experience yet with my larger scopes to really have a deep understanding but with my 4" refractor the useful sampling was limited to about 0.8*. This could have been seeing limited or more likely scope limited. My point is to remember that it is a whole system with guiding issues, scope issues and the camera. You specifically mentioned the 3.76um pixels and this will avoid some of the challenges I faced with using small pixel cameras to get well below 1* of sampling. OK, I've added my limited experience. Now I get to follow the thread and see what results others have with much bigger scopes.

Thanks for the replies.

@CraigT82  that is a great point about brighter targets I haven't thought of - since the SNR is going to be so much higher on them, you can cull your subs more aggressively and keep the ones with the best FWHM. However, do you believe that even when doing this, if your BEST subs are captured under 1.5" seeing, that one could benefit from over-sampling beyond 0.67" per pixel? The former makes sense, and you might not be suggested the latter - but I wanted to check.

It's a good question Jeff, and indeed I have made that direct comparison, for similar reasons as yourself. Feel free to check out the whole story on my website. In short my finding is that the oversampled images came out better. Noted, this was on two quite different systems. But even when comparing on the exact same system Bin1 (0.3"/px), Bin2 (0.6"/px) and Bin3 (0.9"/px), the 0.3"/px came out best. At my site seeing is rarely under 2", so 0.6"/px is already oversampled. Obviously there are diminishing returns. 

Just a few random thoughts as I have collected them when doing these comparisons.

- The difference shows up when using ML-based deconvolution tools such as BXT. This is a deconvolution tool and designed to 'calculate away' seeing effects. So it's not strange that any seeing based limitation (e.g. this sampling rule of thumb) will behave a bit differently when using these tools.

- The seeing/2 or seeing/3 rule of thumb originates from the Nyquist sampling theorem of digitising analog signals. While in theory there is nothing wrong with this, in reality apparently higher sampling frequencies can have benefits. According to Nyquist, CD quality (44kHz) provides an accurate representation of an analog music signal. Still, oversampled audio of 96kHz and even 192KHz offer an improvement in sound quality and are sold as HiRes Audio.

- The theoretical considerations are often focused on stars (how many pixels do I need to properly represent a star). But it works differently for larger structures (nebulosity, galaxy arms, etc). Especially these structures show up better after BXT in oversampled images.

- The noise is also oversampled. While this may not affect the SNR or the object in any quantitative way, the appearance of this 'finer' noise is often more pleasant. Just a nicer 'look' Also, it is possibly easier to reduce, but I've never tried that comparison.


When selecting a telescope/camera combination, I would start with the objects you'd like to image and what FoV would be required for that. From there get the biggest aperture that budget, mount and handling allows. Whatever the pixel scale is that comes from that's just what it is.

It depends on what you want to do really.

If you’re going for dusty and faint then oversampling for your average seeing won’t help.

But if your project is getting in close on a bright target M31, M42 etc) and trying to get as much fine detail as possible then there will be some worth to oversampling. Be prepared to image again and again over multiple nights and only keep the very sharpest subs and you can start to “beat the seeing” as it were. Well, you can do better than average that way, certainly.

I'm honestly well-acquainted with the math and theory involved in the conclusions about "not" oversampling, but my direct real-world experience has been exactly as Craig says above here. Every time. If I remove that focal reducer, or if I step up my focal length (I've done this 4 times with SCTs- from the 8" focal reduced, to the 8" prime focus, to 11" reduced, to 11" prime focus), my results (mind you, on smaller targets of course, not large nebulae) are always better. And not even a little. A lot, actually. More detail, crisper, sharper, and overall more visually appealing. Even the same scope, with a focal reducer operating up to spec, never meets the level of sharpness I obtain at prime focus upon re-sampling the images to compare apples to apples. And I've run this little experiment more than once on different targets. I also agree with Craig's reasoning; you can throw out as many softer subs as you want from bad seeing.

But there's no free lunch. Be prepared for many hours of integration time to make up for the mathematical fact that you're imaging with a much slower system at that point. Want to image fast (few clear nights available, etc.) with larger targets? Keep the lower focal length. Want more detail on smaller targets and get better sharpness and crispness? Higher focal length is key, at the cost of many hours of integration (especially in light pollution).

As an example here's an experimental L frame shot at 0.33" per pixel.

https://app.astrobin.com/i/joqhd2

It's a 6 inch aperture so 2.2 over-sampling from the optical resolution of the system.

In my place seeing is usually around 2".  

Going from 1"/pixel to 0.67"/pixel (same scope, different camera) was a huge improvement in resolution. It seemed litteraly like I traded my 8" newtonian for a larger scope.  

Next I moved from 0.67"/pixel to 0.45"/ pixel by trading the scope for a larger one. Again a significant improvement, but not as spectacular. 

This makes me think that the ideal sampling is between seeing/3 and seeing/4.