Which calibration frames to use on CMOS and WHY? - ASI533

Yes, modern trend is not to use bias frames and use dark flats instead.

For a CMOS camera, it is important to note that dark frames do not scale in the way they do for CCDs.

Thus, you need dark frames that exactly match in length, sensor temperature, and gain/offset - at this point it's probably clear that it's best to choose two gain/offset combinations (one for broadband, one for narrowband), and then use a limited set of sub exposure lengths to avoid the need to capture too many exposure frames and reuse a library of calibration frames.

Since dark frames do not scale, in the same vein, if you wish to use darks for flats, they also need to match exactly. As such, bias frames aren't really necessary in image reduction (since they are used to de-bias darks to allow them to scale).
If you reduce the data in Pixinsight (and possibly other software), you need to uncheck the optimise darks option to stop it trying to scale the darks to reduce noise - if it does it introduces systematic noise (amp glow that isn't properly subtracted usually).

(For Bias frames, you might also want to note that really short bias exposures use a different mode to longer exposures in a lot of sensors, and so the bias data from v short exposures isn't truly representative of what goes on in a real life long exposure)

In my first attempt I simply defocused and used  the same night sky but in narrowband they behaved pretty much the same as darks, there was no serious signal to speak of.

I also tried the dusk sky because the sun is a rich Ha source but they would burn at 3 seconds.

So I waited for a cloudy night and pointed the telescope towards the clouds. Living in a light polluted area (and using a relatively cheap 12nm narrowband filter) the clouds are bright enough to bring the flats at ~ 50% of maximum value. Not ideal but not too bad.

It's tricky and I appreciate it wouldn't work with all combinations of sky/equipment. Not on broadband for instance. There, I would prefer using the western sky where there's Athens' light bubble, this will allow me to do perhaps 10 second flats.

I emphasize the bit about the long flats because I agree with Graeme that sub-second frames seem to have a very different  profile from frames that span at least a few seconds. The latter are much more similar to the long exposure lights in things like noise profile, banding and glows. In my experience this is quite important in narrowband because signal is so weak compared to all those things so you want to calibrate away as much as possible.

Cheers,
Dimitris

Note that all this applies to DSLR too. They are CMOS sensors.

I generally prefer no calibration at all, but when I have to it is (in order of importance):
1) flats of at least 3 seconds exposure
2) a couple of darks.

I also think BPMs do a very good job compared to darks, if you want to correct hot pixels and not dark signal.
Allow me to share a small discovery:

This type of vertical banding in the Nikon D7500 was something that I just could not calibrate away (except with paranoid levels of dithering):

Even though I have stretched it beyond hope and exaggerated the saturation it is not immediately obvious but there are red (or green, I guess) vertical bands spanning the whole image. One of them goes right through M101.

So, dark frames help a little in the sense that most hot pixels are red so they reduce the red bands too. Bias frames sure aren't helping at all. Normal flat frames of less than 1 second make things worse because they introduce some crazy patterns in the red channel. Actually looking at a dark frame shows

So one night it dawned (heh!) on me that if incoming light plays a part, this is clearly uneven field illumination. In the red channel only, not really the shape we are used to seeing, definitely swamped if you  are imaging a bright part of the sky like Sagittarius or Orion, not terribly apparent in short subs, but very much uneven and very much illumination.

Therefore I took a 60 second flat using a wall (shown here with contrast and saturation exaggerated even more than in the light sub):


So now I believe (but will not be able to fully confirm until my next project) that one needs long flats to calibrate away the red banding in the D7500's IMX321 sensor. It appears to be the single most important kind of calibration sub for that particular sensor, since bias signal is practically negligible and darks only correct some hot pixels which most people can live with. If true, it would probably apply to other large modern Sony sensors too.

Thanks for the additional input, everyone!

@Alberto Ibañez THATS a very elaborate documentation. Thanks!

I found another very helpful source. @Adam Block talks in his amazing video tutorial regarding the new WBPP-Script about CMOS and pre-processing! Can recommend it very much!

WBPP - Playlist

CS
Robert

@Jérémie Ochin you may use a thick artist's paper sheet in between the iPad screen and telescope in order to further dim your iFlat . This method works like a charm with my laptop screen and my f/2 lens. I can achieve 1s exposures without reaching the nonlinear regime of my DSLR's sensor. I usually take 60 flats with different paper angles (with the paper surface always parallel to the screen of course) in order to average out any paper/light source inhomogeneities.

Bias frames or dark flats are essential for flat frame calibration to work. See for example this. Also dithering is essential too. I've read somewhere that one of the professional astronomers' motto is "Dither of die!".