I'm jumping into this late. I think that you just put your finger on the issue with the 1600 is there is no "Standard" exposure for a bias frame. they can and do vary.
Just mt 2 cents.
I agree that this is the issue and for me, I just became aware of it a few weeks ago - previously blissfully thinking a simple "standard" bias was the way to go. It's possible Jon Rista or someone else has posted on this, but I haven't seen an evaluation of the 1600 bias characteristics at very short exposures from essentially 0 to 1 sec and if and where its purported "odd" bias characteristics manifest. I plan to work on this today. If I can make the comparison comparison competently, I'll post the results. If someone has already done this, I would be grateful for a link.
Since you divide the lights by the calibrated flats and not subtract them, it is actually important to subtract bias from the flats, and it does not contribute to the double bias subtraction problem. Instead, this danger is with the darks: if you subtract the bias from the lights but not from the darks before subtracting those, the bias gets double subtracted from the original lights.
That's why you either
- subtract bias from both the lights and darks separately, and then subtract bias-free darks from bias-free lights:
([light + bias + dark] - bias) - ([bias + dark] - bias) = (light + dark) - (dark) = light (before dividing by the calibrated flat master)
- or don't subtract bias from either lights or darks, and then it gets subtracted from the lights as part of darks:
(light + bias + dark) - (bias + dark) = light (again, before dividing by the calibrated flat master)
Note that in both cases above we still need the bias (or flat dark) frames for flat calibration: so that in the end we divide by bias-free flats.
I'm just repeating what spokeshave's written in his post #15 of this thread (https://www.cloudyni...ion/?p=10050922), it's very clear.
I assume this can be answered for every sensor by taking darks at various short exposures and statistically comparing them with bias. Although it seems to me this is where the mentioned Poisson noise may come into play obscuring the results. I need to read on it somewhere.
I agree, spokeshave's write-up is probably the best and most clear I've seen. I've have pretty much always used the approach that he describes as:
(Light-Dark)/(flat-bias) At least that what I've thought I was doing!
My lights and darks are not bias-subtracted, but flats are bias-subtracted, so I've had that right notwithstanding the issue of bias vs. flat dark for the asi1600 sensor. However, I've always added a bias file in PI's ImageCalibration dialogue box process when I calibrate lights. I got this from Light Vortex, or one of the other web sites that were very popular a couple of years ago (maybe still are?). But I have no idea what PI is doing with that bias frame.
With these discussion in mind, I tried re-processing the data I'm currently working on without adding the bias frame: adding only raw light frames and an uncalibrated dark master and a bias-subtracted master flat. There was virtually no difference visually or in measured noise using NoiseEvaluation. Is there another way to assess possible differences in the two approaches? Having previously added a bias master in that dialogue box is why I've begin to wonder if I was double subtracting bias somewhere along the line.
As for the difference between a bias and a ~0.5sec flat-dark on the ASI1600mm, your suggestion is spot on and exactly what I have planned to experiment with today. I've not seen anyone directly address this. Jon Rista has posted a fair amount on the subject, but I haven't seen a direct evaluation of the bias characteristics of this sensor.