Sat, 01/09/2016 - 19:30
I am interested in learning the linearity response of the Atik 383L+. It mounts the Kodak Kaf 8300 that is also equipped in other CCDs such as SBIG, etc. I wonder whether there are AAVSO observers that actually do the photometry with this CCD and if anyone has ever tested the linearity of this CCD. Thank you.
Gianluca
Hi Gianluca,
I have a friend who had an Atik camera. There was a white plastic ring as part of the attachment fitting. His split, so the camera fell onto the floor while the telescope was not in use or anyone present in his observatory. The camera fell from about 6 ft (2m) onto a concrete floor, destroying the camera. He did not have a safety line that attached the camera body to the telescope. I'd suggest you (and everyone else with any non-metal parts holding their cameras) put at least one (maybe 3?) safety line on!
I do have a NEW SBIG 8300M camera that uses an identical chip, so I'd appreciate hearing what others experiences are regarding this chip/camera.
Lew
Hi Lew,
I started using an SBIG STT-8300M earlier this year and have been quite happy with the photometry with it. Before I swapped out the old ST-9 I ran a bunch of linearity tests on the 8300 and was pleased with the results. The pixels are quite small for my LX200 10" so I bin 2x2, and from my tests the camera stays linear to at least 64000 ADU. This matches up pretty well with my 1x1 results, where linearity started failing around 25000 ADU. Of course, there's a danger when binning if the images are too sharp where one of the pixels in the 2x2 grouping exceeds the linear range, but with my poor tracking and near-sea level site sharp images are not something I have to worry about!
Shawn
Hello Shawn!
My focal length on the 29 inch prime focus reflector is a bit long, even if it is f/4.36. I have 24 micron pixels (huge) in my primary camera (SBIG STL-1001E). Those give me a scale of 1.52 arc-sec/pixel. I've been under-sampled a couple of times when I had the focus sharp and I was working on a 19th mag star. Most of the time it's about right - 3-6 pix per star FWHM. The only worry I have with the 8300M is the image size of the 1x1 images - 16.5MB!* Even at Amazon Web Service Rates, that is soon going to add up! So a 5 micron pixel size can be binned 2x2, 3x3 or even 4x4 with a net smaller pixel size than the STL 1001E size, provided I save it as a 32 bit file. I will plan to take the images full size and compress them via software. Rob Peter to pay Paul!!
Lew
*The "!" is an exclamation symbol and should not be interpreted as the "!" factorial operator.
- here's an old picture of the 'scope:
Hi -
I'm the person Lew Cook referred to in his post.
I had an Atik 16HR which I had mounted onto my telescope with a plastic ring, which I had machined out of 3/4 inch HDPE. It was that ring that failed, not any part of the Atik camera. The accident wasn't Atik's fault!
I think the ring which stayed on the telescope (while the camera enjoyed air-conditioning) during several hot summers in the observatory, was weakened by the heat. I will clamp my next camera with an aluminum ring!
- Jerry Hudson, HJA
I had thought the plastic part that failed on Jerry's camera was supplied by Atik. It was not Atik's fault, and, once again I am in error (i.e. WRONG!!). My apologies to Atik! I was mistaken, and I am sorry.
Chagrinned and embarassed,
Lew
Hey Lew -
Stop apologizing all over the place!
Probably mea culpa for not explaining to you exactly what happened that night in the observatory. And no doubt, I was a bit stressed. Anyhow, I don't blame Atik for the accident.
- Jerry, HJA
Sorry to hear about the accident of the plastic ring. However coming back to the original topic shall I assume that the kaf 8300 chip stays linear up to about 25KAdu in binning 1x1? That seems very little to me. According to Richard Berry's book CCDs that lose linearity below 1/2 full well deliver poor photometric results and even calibrating with flats may be not optimal. Are there people out there at AAVSO that use CCDs with the Kaf 8300 for their photometry that have previously determined the linearity of the CCD and that are willing to report their results? Lew, what about your experience with the SBIG?
Gianluca - RGN
Hi Gianluca,
If you are just talking about the KAF-8300 sensor itself, we are using a QSI-583 on the AAVSOnet TMO61 telescope in 1x1 binning, and getting very good photometry. While an ABG chip is only linear to about 1/2 full well, most vendors increase the gain so that half full well saturates the ADC, giving a linear response throughout its range. I assume Atik does the same.
Arne
Hi Arne,
I can confirm that probably most vendors increase the gain to extend the linearity of ABG CCDs. I have run a series of linearity tests of the Atik 314L+ that turned out to be linear throughout most of its range. I would like to upgrade from my 314 to a bigger sensor mainly to increase the FOV and feel the Kaf 8300 can be a good choice considering the useful field of my scope. My concern with Atik however is that they probably do not run extensive linearity tests of their CCDs though they sell good cameras for "pretty images" having little noise. I have come across an Atik 490 that showed a very poor linearity response. According to Atik website the 383L+ has "....Great linearity, with a regression correlation coefficient of R = 0.9998 over the range 1,000 to 64,000, making the camera suitable for sensitive photometric measurements." Unfortunately that statement seems also to apply to the Atik 490 EX. All that said I would be glad to know if some AAVSO observers actually use the Atik 383 for photometry and if they can share the linearity tests done with this camera.
Cheers
Gianluca - RGN
Finally, I managed to find an owner of this CCD and then I was able to do the test. I have taken a set of flats with increasing exposure and one corresponding dark. After averaging the calibrated flats I measured the average ADU reading using Maxim DL, area mode. As previously suggested by Arne I have taken here and there a set of 5 secs flats to check light source stability during the test. Here are the results:
Exp. (secs)
Average ADU (whole image analysis)
1
286.005
5
1418.478
10
2829.861
15
4236.022
5
1415.326
20
5637.022
25
7035.323
30
8425.804
5
1417.499
35
9785.699
40
11171.325
45
12560.431
5
1403.797
50
13916.985
55
15301.066
60
16657.252
5
1401.119
65
18025.611
70
19378.645
75
20739.061
85
23455.842
100
27534.879
115
31586.877
130
35620.41
145
39645.043
160
43667.035
180
48966.16
200
54251.656
Exp. (sec)
rate of incr. (ADU per sec)
1
286.005
5
283.6956
10
282.9861
15
282.4014666667
5
283.0652
20
281.8511
25
281.41292
30
280.8601333333
5
283.4998
35
279.5914
40
279.283125
45
279.1206888889
5
280.7594
50
278.3397
55
278.2012
60
277.6208666667
5
280.2238
65
277.3170923077
70
276.8377857143
75
276.5208133333
85
275.9510823529
100
275.34879
115
274.6684956522
130
274.0031538462
145
273.4140896552
160
272.91896875
180
272.0342222222
200
271.25828
Based on this test I understand that this CCD is linear (1% deviation) in the follwing range in ADU:
1400-9000; 9000-15300; 15300-27500; 27500-48000
so care must be exercised in selecting the comparisons stars for photometry.
I would be happy to get some feedback on the above-mentioned test and conclusions.
Gianluca
Hello Gianluca
The results that you posted for linearity testing. Did you subtract the bias and dark for each one?
Gary
I have taken 2 flats and 1 dark for each individual exposure time following the recommended procedure of Aip4win. I can however redo the test with different recommendations, if any.
Gianluca
Looks pretty linear to me. From the fit I would expect a mean bias level offset of 203 ADU - is this in accordance with a bias histogram?
Frank
Frank I got exactly the same trend line when I applied an affine function fit (Y=mX+b). However, since Gianluca subtraced dark frames of exactly the same EXPTIME as the flats there should not be a bias frame offset. If you do a true linear fit using the LINEST function with the constant term forced to zero (Const = FALSE, Stats = TRUE) you will see that the r^2 value is slightly closer to 1.0 and the F statistic is larger than with a non-zero constant, meaning the regression line constrained to go through zero is a slightly better fit even though the SumSquare residuals is larger. I think the offset comes about mostly because the light source is not constant over time. There could also be some contribution from imperfect dark current cancellation due to differences in dark currents in the single dark frame vs the dark currents in the flat frame. However, I would expect the contribution from random pixel dark current fluctuations to be much less than 203 counts since you are taking the mean of a large number of pixels UNLESS the temp of the camera changed slightly. This could occur, for example, if the temp of the camera were settling back to the steady state value from a slightly lower temp due to cooling "overshoot" that often happens when a camera first cools down. I suspect, however, Gianluca gave the camera plenty of time to settle down before starting the test sequence. Dark currents are subject to Poisson noise just like a light source. That is why in my subseuent e-mail, I suggested that for the purposes of a linearity test it is probably better to subtract a master dark composed of the same number of darks at the same temp and EXPTIME as the flat just as you would do for a data frame when you are not scaling darks. That is as close as you are going to get to perfect dark current cancelation, but as stated above, I think the main source of the offset in the early frames is due to source intensity change, not dark current noise and this conclusion is supported by a plot of the ADUs of the 5 second exposures vs their rank order in the sequence of images.
Also, the very small deviations from linearity that we are trying to detect are difficult to see in a plot of ADU vs. EXPTIME. Athe Handbook of Astronomical Image Processing (AIP) suggests plotting ADUs per second vs EXPTIME since that amplifies differences, but I think it is even better to plot % residuals 100*(ActualADU - ModelADU) / ModelADU vs. Actual ADU since that shows % deviation from linearity at different image depths directly on the graph and no ancillary calculations are required.
Brad Walter, WBY
Hi Gianluca:
What I see in your ADU versus time graph is a linear section up to about 75 sec and then a divergence. This is about 23000 ADU. I have seen other reports of non-linearity of ABG chips above 25000 ADU. Also, I noted that the last two 5 sec images gave lower ADU/sec than the first three. Problem? Using a 1 sec image may start to be impacted by shutter effects? Ignore?
I'm not sure why you broke it up into several smaller 1% linear sections. Are you planning on running all your photometric analyses by hand. I don't think most photometry software is set up to do such an analysis scheme?
Ken
Gianluca, this camera seems to have significant non-linearity compared to, say, a camera using a non ABG KAF 6303. However, I think there is a problem with both our techniques. I did not use the AIP4WIN 2-flat method so I have some light source variation mixed in which is evident from comparing the plot of EXPTIME vs Mid UT time of the image to the plot of ADU vs. exposure time. It is clear that my EL screen dimmed over the duration of the test. But it is also clear that most of that would disappear by averaging one of the images taken during the sequence with increasing EXPTIMES with an image taken in the sequence with decreasing EXPTIMES. unfortunately I did not continue the decreasing EXPTIME sequence to the shortest exposures and I should have.
I had some difficulty seeing the extent of non-linearity from your graphs so I made some of my own using your data showing the ADU vs. exposure time and the % residual of the difference between the actual ADU vs the linear regression model ADU. I also made a graph comparing the % residuals of the 5 second exposure ADU values relative to the average of the 5 second exposure ADUs. See attached graphs. The percent residuals from a true linear regression line (force to go through 0,0) ranges from about +4.5% for a 1 second exposure to about -0.9% for your longest exposure. It appears that you took your exposures in a progression from shortest to longest. Your 5 second exposures showed a definite1.4 % dimming compared to the average for the 5 samples you took. I am assuming each of the 5 second image values you posted are averages of the values taken during the increasing and decreasing EXPTIME sequence of exposures. If my assumption is correct then your light source (and I think mine as well) dims more quickly when it is first turned on so the intensity for your first short exposures in the increasing EXPTIME sequence is not fully counterbalanced by the dimming of the corresponding short exposure in the decreasing EXPTIME flat sequence. I can't tell how much of the non-linearity that appears is due to the source and how much is due to the camera. I think you can separate the effects by plotting the ADUs of the individual 5 second exposures vs. the elapsed time from the first flat in the increasing EXPTIME sequence (but of course this assumes the light source was on uninterruptedly during the whole process).
Compare the shape of the % residual curve for the ATIK 383L+ camera to the STXL 6303 Camera. the 6303 % residual stays relatively constant at a value between +/- 0.5% for exposures longer than 5 seconds and I think much of the variation for the initial short exposures is due to higher lamp intensity when it is first turned on. At this point that is surmise but it is indicated by comparing the % residual graph for the 6303 vs the UT mid-image time vs exposure duration plot. Much of the scatter between the 6303 images taken at the same EXPTIME is clearly due to the light source intensity reduction over time but the shapes of the residual curves are very different.
Rather than using a single dark frame at the same EXPTIME as the flat, I used a master dark composed of 20 Darks each at the same EXPTIME as the corresponding flat frame. That is a lot of dark frames, but I wanted to keep the noise low and avoid having to use bias frames which increase noise.
So what can we do to eliminate the effect of changing lamp intensity? My idea is to take a 5 sec check flat exposure before the first exposure of the increasing EXPTIME sequence and at two or three intervening points on the way up. Take one in the middle of the longest exposure and then take them at the same points in the decreasing EXPTIME sequence as on the way up and at the conclusion of the decreasing sequence. Then it is possible to construct an intensity function with respect to the elapsed time form the mid-frame time of the first image to correct for intensity variation. I also think we should take two frames at each EXPTIME in the increasing and decreasing sequences so that we can compare the scatter between the two adjacent frames compared to the scatter between corresponding EXPTIME frames in the increasing and decreasing EXPTIME sequences.
This is a bit complicated, but I hope I am communicating my idea for a sound procedure well enough that it is understandable.
Also, I think it is really hard to set your exposures so that different stars have ADU values that fall within different relatively narrow ADU ranges that are within 1% or less of the linear regression line. It is probably easier to mathematically correct for the non linearity as long as it is consistent for different pixels. "Ay," could that be our "rub" in creating a mathematical correction or can we avoid the "slings and arrows" of unwanted complication and simply assume all pixels are similar enough? I asked Yorick, but, alas, he's not talking.
Brad Walter, WBY
Hi Brad,
thank you for your reply. I think you have highlighted the main point in question, that is light source stability over the test. Since the CCD is not mine I went to my frined's place without bringing my flat box that delivers a proven stable light output. Instead we have used a monitor screen of an Ipad that has most probably slightly dimmed over the test increasing the effect of non-linearity. By the way I have not taken a set of increasing and decreasing exposure times but followed the order I have posted (progressive incresing exposures). The only way to understand whether part of the non-linearity of this CCD is actually due to the dimming of the light source is by redoing the test using my flat box and taking here and there some 5 sec exposures to check for light source stability. Is there any good reason I should order the sequence of exposures by taking increasing and decreasing exposure times given that I use 5 sec exposures here and there (say every 3 exposure till the end of the test this time) as a check of light stability?
Gianluca
Yes, I think there are good reasons for taking a descending as well as ascending EXPTIME sequence.
Using increasing and decreasing sequences allows you to take your analysis a bit further. First, it highlights any intensity change effects. Then, after you have eliminated systematic source intensity variation using a 'smooth curve" function you derive from the 5 second "check" flats, it lets you see if your linearity is affected by any time dependent or order dependent effects other than variations in source intensity. It gives you a way to separate any remaining effects that are related to elapsed time or order from things that are related to integration time but not related to saturation. Also if you look at figure 6.7 in AIP by averaging increasing and decreasing sequence images with corresponding sequence times you eliminate most of the shorter term source intensity effects that can't be eliminated by a "smooth curve" correction derived from the 5 second "check" flats. So averaging them also provides an alternate (or checking) means of correcting for source intensity variations.
So I think having both sequences provides substantial benefit but you want the entire pair of sequences to be automatic so that they are as symmetrical as possible in elapsed time around the longest EXPTIME. That is easy to do in Maxim DL and probably most other popular camera control programs.
If you look at section 6.3.3.3 in AIP they recommend using an increasing and decreasing set of flats with two flats and a dark for each EXPTIME in each sequence. They want two flats at a time in both sequences to determine the CCD transfer function. You don't need two at a time for the linearity test, However, by taking multiple images in the increasing and decreasing sequences you can separate short term (mostly random) variations that may be more than just the normal Poisson noise (for example, short term supply voltage changes, or the i-phone screen illumination being affected by other processes going on in the device, which affect a single image).
Also for linearity testing, It seems to me you are better using master darks at each EXPTIME comprised of the same quantity of individual darks that you would use for normal photometry to minimize dark frame Poisson noise in the calibrated flats. This is probably a small thing, but if we do it in calibrating our data frames we should do it for the linearity test as well.
Finally, I prefer looking at % residuals vs. mean ADU rather than ADUs per second vs. EXPTIME as is done in AIP. Looking at % residuals gives you the % deviation from linearity at various ADU levels directly, rather than requiring ancillary calculations to get at the value you really want to see.
Brad Walter, WBY
PS.: It was because of your reference to AIP4WIN and the contents of section 6.3.3.3 in AIP that I thought you had taken increasing and decreasing EXPTIME sequences.
Gianluca,
Attached is a pro-forma file showing the sequence of flats I tried to describe. Of course the range of flat EXPTIMES is appropriate for my EL screen. You would have a different range.
Also, when I did my test I did not allow very much time for the light source to stabilize before I started the series of exposures. When I do it again, I will turn it on 15 or 20 minutes before I start the test. I was surprised by the amount of change in the EL screen output as it warmed up. I expect large changes with conventional lamps but did not expect that much variation with an EL Screen because it has only a small fraction of the temperature change of a conventional lamp. Photometry is a never-ending learning experience.
Brad Walter, WBY
Gianluca,
One more PDF from the STXL Spreadsheet showing ADU/Sec vs. Elapsed time from the UT mid-exposure time of the 1 sec image at the beginning of the increasing EXPTIME sequence. This shows there is possibly some source intensity variation affecting the results. However, because
1. I stopped the descending EXPTIME sequence at 10 seconds instead of going all the way down to 1 sec;
2. I did not have constant EXPTIME check images at the beginning, end, midpoint and other points during the test; and
3. I did not let the light source stabilize sufficiently before starting the test;
I can't tell how much of the variation in ADU/Sec, particularly in the early images of less than 10 sec duration is due to source intensity fluctuation or, mechanical effects like shutter latency, or incomplete bias elimination by the darks (20-darkframe masters at same time and temp as the flats). However, this last plot does allow me to separate non-linearity in images 10 seconds and longer from source intensity change over time, but not as well as if I had include item 2 above in my procedure. The pairs of images at each EXPTIME in the decreasing sequence provide some idea as to how much random variation occurs as well. I think it would be better if I had taken pairs during the increasing EXPTIME as well.
Brad Walter, WBY