CMOS Bias stability

[andrew s]

Several threads on here have commented that in effect that bias frames with CMOS cameras can give problems. I have started a series of tests on my ZWO ASI 1600 at -20c. So far the statistics have remained stable and subtracting pairs of bias frames gives a sensible looking histogram.

I am keen to uses scaled darks as my observation program requires a wide range of exposures and doing darks for them all is possible by messy.

Does anyone have any data on the issue or can point me to any papers which discuss the issue scientifically rather than anecdotally.

Thanks Andrew

[Adam J]

I think the first thing is to understand the issue correctly myth or otherwise.

As I understand it the problem is that every time the chip is powered it performs an internal calibration and so its not that the bias is unstable between images in a single run its that the bias is unstable between imaging sessions. As such you cant use bias from one session to scale darks from another session or vice versa because the bias component of the light frames will not be the same as the bias component of the dark / bias frames.  

Now the second issue is that this calibration is only applied at shorter exposures of less than a few seconds and so the bias component of the dark frames at 2 mins is not the same as the bias component of bias frame of 0.01 seconds. I have seen some people claiming that you can just use a 6 second bias frame as opposed to the normal minimum exposure hence avoiding the issue as the calibration will not be applied to darks or bias but have not tried this. In fact this is why its recommended to aim for longer flat frames with this sensor.  

 The final nail on the coffin is that the amp glow does not grow in a linear fashion with time or at least that is what I am told, as such it will not scale anyway and prevents the use of dark frame optimisation as this relies on a linear model, hence why all the stacking programs tell you to turn off dark frame optimisation for CMOS in their manuals now to prevent residual amp glow. Finally there is something called telegraph noise in very short exposures (Horizontal banding that moves slightly / jumps between exposures) you will see it in a large stretch of a Bias but would not appear in frame stats as these are averaged across the frame.

So although I don't have any raw data to hand I hope that this at least frames the issue correctly for you so that you can perform your own investigation. I note that none of the above conflicts with your result.  

Adam

 

Edited August 28, 2019 by Adam J

[andrew s]

Thanks @Adam J that gives we something to investigate. When you refer to powering the chip I assume that is via USB as I can run it without the 12v needed for cooling.

I don't  think amp glow is not an issue for me as I use max 20s exposure and the science image is on a quite small area near the centre of the chip but will test.

Regards Andrew 

[Adam J]

Just now, andrew s said:

Thanks @Adam J that gives we something to investigate. When you refer to powering the chip I assume that is via USB as I can run it without the 12v needed for cooling.

I don't  think amp glow is not an issue for me as I use max 20s exposure and the science image is on a quite small area near the centre of the chip but will test.

Regards Andrew 

Yes although I have never tried to have the 12 volt cooler powered without the USB attached, I suspect that removing USB power would interrupt the cooling control.

[smr]

With my DSLR (CMOS) I've found that using Bias frames gives a much smoother background. Without them I get all sorts of streaks and artifacts. Unless you mean dedicated astro CMOS cameras?

[andrew s]

7 minutes ago, smr said:

With my DSLR (CMOS) I've found that using Bias frames gives a much smoother background. Without them I get all sorts of streaks and artifacts. Unless you mean dedicated astro CMOS cameras?

Yes thanks, but I did mean dedicated astro cameras specifically ASI 1600 MM

Regards Andrew 

[vlaiv]

I've found two types of bias issues with cmos sensors - both would prevent dark scaling how it's usually done. However, if you want to do dark scaling, I think there is way to do it with ASI1600 and we can devise simple procedure to test this.

First problem is the one @Adam J described - internal calibration and difference in bias between power cycles.

ASI1600 does not suffer from this problem, it has another thing that is making regular dark scaling problematic.

It has to do with the fact that there are sort of two regimes of operation - "internal clock" and "external clock", at least that is what I've read, and might not be related to this issue at all, but I suspect there might be some relation. When I did my measurement of ASI1600 I noticed that bias sub has higher average pixel value than dark sub of relatively short exposure - like 20-30s.

This simply cannot happen if bias is proper. If you take set of bias files and later another set of bias files, you stack each and subtract stacks - you will get what you expect - average 0 and noise (stddev) as expected. Same will happen with darks of same exposure (and same temperature of course).

Where you can see the problem is if you try to match two different set of darks. Take bias subs for master bias, take set of shorter darks and take set of longer darks, for simplicity you can take 30s and 1minute.

Now if you take short darks stack and subtract bias and then multiply by 2 (or which ever ration of exposure lengths you chose) and take long darks stack and remove bias, and then subtract the two, you should get 0 average + expected noise. If this works, then you can scale darks, but I suspect it will not work, at least it did not work when I did my measurements.

Things might have changed in the mean time with drivers or whatever (there was change of how offset is applied - fixed/not fixed/what is default), so it's best to run above test to see if you can effectively scale darks.

There is another way of doing dark scaling that does not involve shooting actual bias subs, but rather extracting common bias from sets of darks. To test this, you would need three sets of darks.

Let's say 1m, 2m and 3m.

2m - 1m = X (1 minute of dark current only since bias cancels out)

1m - X = bias

now you can test if

3m - bias = 3*X

and if it is, you can scale your darks with two different exposure only and doing some math (2m and 1m is enough to extract "unity" dark current and common bias).

[vlaiv]

I forgot to expand on internal vs external clock. I've read somewhere that for exposures of less than 1 seconds, either driver or camera firmware is controlling the exposure length, and above 1s it is up to application using ASCOM driver to start/stop exposure.

Might be that there is different bias applied to these two modes, and using sub 1s bias with over 1s exposures leads to problems described above.

[vlaiv]

Btw, this is something that Christian Buil also noted in his report on feasibility of CMOS sensors for spectrographic purposes, here is a screen shot:

He ascribes this to linearity error on small signals, but I in fact suspect it is due to bias not being proper.

Source: http://www.astrosurf.com/buil/CMOSvsCCD/index.html

[andrew s]

I will investigate all these issues and see what happens then report back.

Regards Andrew 

[andrew s]

Ok, I have done some tests and research. As @Adam J noted the killer is the nature of the amp glow. So I will have to do sets of darks for each exposure.

My results showed there were very small changes in bias +- 2 adu in different sessions including camera and PC booting after power off. Darks were never below Bias with exposures in the range 0, 0.01 0.1,1,2...8,16,64 and 128s and progress linearly. Subtracting pairs of bias or dark frames showed just well behaved noise with no structure across the frame.

Apart from the amp glow I found nothing to worry about.

Regards Andrew