CCD bias instability?

[vlaiv]

Recently I had a chance to do some measurements on bias subs from CCD sensors (two in total) and I've found something that really confused me.

At first it seemed like issue with power source, but second measurement on one of the sensors with what could have been power issue corrected - gave similar if not even worse results.

I have not taken subs myself, but have no reason to believe there were any issues with acquisition (-20C, 0.001s exposure, no light - regular "protocol" for bias subs). My intention was to measure read noise, so idea was to take 16 bias subs - split into two groups, stack, subtract and measure standard deviation of result (corrected for stacking addition). I expected this sort of bias - bias calibration to provide nice uniform noise with mean value of 0. Neither set of data actually produced such result. There were residual mean value quite different than 0 in each case. This should not be the case so I inspected mean values of bias subs in each run and found what I can only describe as instability of the bias. Here are results:

Camera 1, 16 bias measurement, nothing done with data (no e/ADU conversion or anything - values in table are measured ADU values straight from 16bit fits):

I pointed out funny results. I would expect bias mean value measurement to roughly behave as other 13 subs in this list - slight variation around some value - in this case 363.4 or there about. What happened with those three outlined subs?

Camera 2, again same measurement done on 16 bias subs as with above camera (same sensor, but different vendor hence different offset and e/ADU values - resulting values will differ but I expect again same to be true - bias mean value should be very close to some value across measurements):

Here we see much more scatter and larger deviation. Mean ADU levels per sub vary from ~661.5 up to ~655. Not something that I would expect. Standard deviation varies much less than mean ADU value, so read noise remains pretty much the same.

Camera 2, second take with different power supply conditions (and probably changed gain since mean values are larger than in first batch of subs from that camera, but stddev is lower):

This one is even more interesting - very large variation in mean ADU values - and almost sorted in descending order - difference of about 20ADU from first to last bias sub.

Set of associated darks show similar thing (taken at -20C, 300s exposure, again 16 of them used):

Camera 1:

This time only two subs out of 16 had +6 ADU increase (same thing as with bias for this camera), while mean value of other subs is relatively stable (there might be slight trend to rise - probably due to cooling).

Camera 2 (set1 and set2 - same 300s, -20C):

One can say that these variations in mean ADU level of darks are associated with cooling (how fast it reaches set temperature and if there is overshoot), but I don't think they are - look at noise values - they don't follow the same trend. It is bias related.

I don't own CCD camera, nor I have ever worked with one, but think I have fairly good understanding on what should be expected in above tables. As a comparison here is similar measurement done on 16 subs from my ASI1600 at -20C, unity gain, 240s exposure:

Mean values are nice and "uniform".

Anybody have a clue what is going on with above CCD sensors and their bias?

 

[Filroden]

In the three bias frames on camera 1 that show the higher average ADU, can you detect any significant variance within parts of the image or is the higher value fairly constant across the image?

[vlaiv]

24 minutes ago, Filroden said:

In the three bias frames on camera 1 that show the higher average ADU, can you detect any significant variance within parts of the image or is the higher value fairly constant across the image?

It does not look like there is significant difference, here is one sub with "normal" mean ADU level and one sub with higher ADU level side by side. I subtracted uniform level of 6.6 from higher ADU level one and applied same linear histogram stretch to both:

FFT of both frames show pretty much same thing (this is central part of each FFT next to each other):

[Filroden]

Is there an edge glow on the left side of the second image in the montage? It looks noticeably brighter. Might be worth checking the ADU values across a row.

[vlaiv]

Here is what should be the difference for camera 1 - "normal" subs and "higher ADU" subs:

I stacked normal subs to one stack, higher ADU subs to other stack (both average stacking method) and then subtracted resulting stacks. There is obvious gradient in difference of the two.

Here is again comparison of the two groups - side by side, this time using stacks instead of single subs (one stack has 13 subs, other only 3 and is therefore more noisy). Higher value stack was "shifted" to get same average ADU value, and same linear stretch again applied. I added dark background so any edge brightening can be seen more easily:

What could be causing the difference? For a moment I suspected light leak of the sorts, but doubt it would be visible in 0.001s bias subs and not in 300s darks.

[andrew s]

@vlaiv are these differences to be expected? Maybe a students t test would tell you if they are.

Regards Andrew

[vlaiv]

3 minutes ago, andrew s said:

@vlaiv are these differences to be expected? Maybe a students t test would tell you if they are.

Regards Andrew

I thought about that, but can't figure out how to properly do calculations.

My simplistic (although might not be accurate) method would be: Bias noise per pixel is order of 10e. There are about 8.3Mp on this sensor, so average value of all of those pixels should be some value +/- 10 / sqrt(8300000) = 0.0035e

So for 99.7% of time, one would expect average of bias (average value of all pixels in bias) to be some value +/- 0.01e (3 sigma). This indicates that for most part, if it were only for read noise of pixels, average bias value should be fairly constant.

There seems to be some other noise component related to bias in case of ASI1600, as similar calculation performed on my darks would give something like 0.001 sigma, so most frames would be +/- 0.003 yet as we have seen from above table, most values are in range ~63.945 - ~63.971 so actual range is about 4 times as large, or +/-0.013.

Having said all of that, and comparing ASI1600 results with my rough calculation - here bias varies by two orders of magnitude more then expected in case of these two cameras. Other explanation is of course that my approach to estimating variance in mean bias value is completely wrong

 

[vlaiv]

23 minutes ago, andrew s said:

@vlaiv are these differences to be expected? Maybe a students t test would tell you if they are.

Regards Andrew

Well, I still don't understand t test completely (will need a bit more reading), but here is quick calculation of it with online calculator:

I just took set of bias subs from other camera, split into two groups, calculated mean ADU value per each sub and treated those numbers as members of each group1 and group2 for t test and performed calculation.

Result speaks for it self - it looks like these two groups of bias files (although from the same camera taken in succession) don't belong to same distribution.

Not sure if this is the test you had in mind, if not, can you expand on what I should do?

[Adam J]

30 minutes ago, vlaiv said:

Well, I still don't understand t test completely (will need a bit more reading), but here is quick calculation of it with online calculator:

I just took set of bias subs from other camera, split into two groups, calculated mean ADU value per each sub and treated those numbers as members of each group1 and group2 for t test and performed calculation.

Result speaks for it self - it looks like these two groups of bias files (although from the same camera taken in succession) don't belong to same distribution.

Not sure if this is the test you had in mind, if not, can you expand on what I should do?

It's the correct test and your interpretation is also correct. 

[vlaiv]

So no one is having a clue what this might be?

Could someone do the same measurement with their bias subs on CCD camera?

I'm inclined to believe that this might not be fluke and that this stuff can happen to people. Something like this happening on occasion could well be solution to flat calibration mystery that was never solved but happens sometimes to @ollypenrice.

[ollypenrice]

2 minutes ago, vlaiv said:

So no one is having a clue what this might be?

Could someone do the same measurement with their bias subs on CCD camera?

I'm inclined to believe that this might not be fluke and that this stuff can happen to people. Something like this happening on occasion could well be solution to flat calibration mystery that was never solved but happens sometimes to @ollypenrice.

At least I'll be glad to hear that it isn't just me! I shoot more data than most members, doing this for a living and running up to three rigs per night, so if it's a random problem it's likely to hit me more often than most. With some dual rigs operating remotely, though, some other members are now shooting more data than I do. I always follow this issue with interest because there must be an explanation for it.

Olly

[andrew s]

2 hours ago, vlaiv said:

So no one is having a clue what this might be?

I would speculate that it is in the A/D and gain circuits. Providing high stability in a consumer chip is perhaps too much to hope for.

If you stress the circuit with multiple reads with no time to stabilise I would not be suprised  noise were introduced. 

Regards Andrew 

[vlaiv]

8 minutes ago, andrew s said:

I would speculate that it is in the A/D and gain circuits. Providing high stability in a consumer chip is perhaps too much to hope for.

If you stress the circuit with multiple reads with no time to stabilise I would not be suprised  noise were introduced. 

Regards Andrew 

I wonder why it it not "known" thing then? Many people use CCD sensors that are based on "consumer" chips. In fact dark scaling depends on stable bias and many people do it.

In any case - it's worth a try to take subs with a bit of time in between.

[andrew s]

11 hours ago, vlaiv said:

I wonder why it it not "known" thing then? Many people use CCD sensors that are based on "consumer" chips. In fact dark scaling depends on stable bias and many people do it.

In any case - it's worth a try to take subs with a bit of time in between.

Not many people take the trouble to measure their system. Also the differences are small as is the error introduced. It is probably just one of several issues that lead to  the anecdotal comments on some cameras being easier to process than others.

It has been discussed before, I finally remembered I had discussed this with Andy Wilson (who raised it) on the BAA forum here https://britastro.org/node/12121 .

Regards Andrew

Edited November 12, 2019 by andrew s

[Magnum]

Im no expert at all on this, but may as well offer My first thought that it could be Voltage instability, I have done experiments with my Kaf8300 camera and voltage definitely has an effect on the amount of noise, my cam is significantly cleaner at 13V than at 12V. 

Maybe the examples you have been playing with were taken with a non regulated power supply, or supply that is failing?

Anyway just a thought

Lee

Edited November 12, 2019 by Magnum

[tooth_dr]

1 hour ago, Magnum said:

Im no expert at all on this, but may as well offer My first thought that it could be Voltage instability, I have done experiments with my Kaf8300 camera and voltage definitely has an effect on the amount of noise, my cam is significantly cleaner at 13V than at 12V. 

Maybe the examples you have been playing with were taken with a non regulated power supply, or supply that is failing?

Anyway just a thought

Lee

I'll chip in here.  These were taken using a Maplin regulated powersupply, rated 6-8Amps, and I've measured it and it delivers 14v.  It is 2 years old.

[Magnum]

1 hour ago, tooth_dr said:

I'll chip in here.  These were taken using a Maplin regulated powersupply, rated 6-8Amps, and I've measured it and it delivers 14v.  It is 2 years old.

Ok was just a thought, though while we are talking about voltage 14V is a bit on the high side, Atik told me that 13.8V is the absolute maximum I should use with my camera. My Maplin 5amp bench supply puts out exactly 13.8V so I decided to edge on the side of caution and use my other power supply that is 13V.