Callibration frames deminishing returns

[Pryce]

So I'm checking here and there about callibration frames and I've found a lot of conflicting information. At what point would you reach deminishing returns for your bias and darks?

Does it change based on exposure time? 

Sensor? ISO? Temperature? 

For now I'm using an EOS 550D unmodded and of course uncooled.

Edited September 8, 2020 by Pryce

[Ouroboros]

I use a Canon 450D. I get so little sky time that it feels a dreadful waste to spend  time taking darks when the sky is clear.  So I gave up shooting darks at all some time ago and instead use dithering to try and smooth out repetitive noise.  I haven’t noticed my images being any worse for that. Flats and bias on the other hand are quick to take. I normally shoot off 32 or 64-ish of those. 

I experimented for a while building up a darks library by putting my camera out at night in different temperatures and taking darks of my typical exposure length. It worked OK but not strikingly better than not using darks at all. 

Edited September 8, 2020 by Ouroboros

[Pryce]

I guess I should just do some testing and see what works for me and my camera? 

[Seelive]

Using several calibration frames is all about reducing the 'random' noise applied to each light during the calibration process. Consider darks, they are taken at the same temperature and exposure as the lights so, ignoring photon noise, each light and each dark will have the same random noise (created by the camera). Since random noise adds statistically then for a noise level N in both the light and dark then just using a single dark will result in a calibrated light with a noise of 1.41N, not what we want!  Combining several darks reduces the random noise statistically, so using D darks, the combined master dark will have a noise reduced by the square root of D which will then add statistically to the light noise.  So putting some numbers together:

1 dark - calibrated light noise 1.41N

2 darks - calibrated light noise 1.22N

4 darks - calibrated light noise 1.12N

8 darks - calibrated light noise 1.06N

16 darks - calibrated light noise 1.03N

.

.

64 darks - calibrated light noise 1.01N

And if you dither and don't use darks like @Ouroboros suggests you won't add any noise. But the other calibration images will have (a lesser) effect.

And apologies for the maths 😊

 

 

Edited September 8, 2020 by Seelive

[vlaiv]

2 hours ago, Seelive said:

And apologies for the maths

I personally don't mind the math, but you need to be aware that above is correct only in one particular case - perfect dither.

If you have opposite case - perfect alignment then same math applies not to single sub but to whole stack - which is important.

If we take only read noise - let it be 3e and one does 100 darks (which includes that read noise as well), single sub calibration will be polluted with 0.3e of read noise from master dark alone - it does not sound much if combined with 3e of read noise of light itself.

But when you have perfectly aligned subs - master dark can be "pulled in front of sum of lights" and it will act on stack of lights instead of each single light (like in case of perfect dither) - in that case, let's have 100 lights as well - so read noise in final stack will be again 0.3e from those lights, but now we are adding another 0.3e from master dark and we end up ~ 0.4243e of noise in total stack, or as if you only stacked 50 subs (with respect to read noise and other noises present in dark - not shot noise or LP noise which are related to lights only).

Moral of the story is - dither and dither often, and people that don't dither will in all likelihood be somewhere in between prefect dither and perfect alignment. In fact - better the guiding - closer to perfect alignment one is and more calibration frames are needed ....