Affiliation
None
Mon, 01/21/2013 - 19:48
Hi all,
Since I've taken my ST9e ccd camera off of my scope for the first time in a long while and the weather is generally bad, I've decided to re-do the linearity test for my camera. I just want to ask a few questions:
1.) I presume that all of the light frames must be bias and dark subtracted, correct?
2.) Is a standard fluorescent light indirectly illuminating the camera stable enough to do the test?
3.) Should I average several frames at each exposure time to get better statistics?
4.) What else am I forgetting?
Thanks for sharing your accumulated wisdom.
...Tim (HTY)
Hi Tim,
You do not need to do any of that for the linearity testing. Ideally just used a fixed continuous spectrum slight source, such as from an incandescent light. Adjust the brightness so the first exposure produces ADU counts of around 1000 (assuming a 16-bit camera). Then increase the exposure times and note the ADU counts. You should be able to plot a straight line. The point where the line turns (knee) is where the linearity breaks.
Note: You do not subtract flat frames. You must create the proper flat and then use it to divide into the image, but again this is not necessary for linearity testing.
Jeff 187283 (HPO)
Hi Tim,
A fluorescent light is not a very stable light source. You've noticed that most lamps take time before they reach nominal brightness; they fluctuate after that as well. They also are not "on" all of the time, but pulsate with the changing AC current. However, I've seen decent results even with them. The trick is to find a lamp brightness that gives you about 100ADU/sec, so that you reach the saturation of your sensor in about 60 seconds. Then ramp the exposure up from short (say, 1 second) to saturation, but every so often, take a 5 second exposure. This set of 5-second exposures will tell you how stable your light was during the entire process.
Rather than taking several frames at each exposure time, I'd suggest doing the entire set several times and averaging the results. That randomizes any stability change in your lamp, and minimizes the time available for the lamp to vary. Since what you are looking for is the response of the detector to light, you should dark subtract each frame. That does change the final numbers you get for saturation, so for greatest accuracy, you would add the subtracted dark to any determined value - but in general, you can just ignore it.
Unless you are using an ultra-stable light source, such as a pulsed LED, these tests will tell you in the 1% range how linear things are, and they do a reasonable job of showing saturation. The flats can also be used to determine read noise and gain via the Janesick method.
Arne
I use the same Light box that I use to make flats. It has very low power (7 watts, If I recall correctly) halogen lamps. I set the intensity of the lamps to a level that gives about 500 counts in 5 seconds. I know this seems a very low light level but I don't take exposures less than 5 seconds doing photometry even with a SBIG camera that has a "photometric" rotating gate shutter. It may not make any difference at all, but I try to duplicate the conditions under which I do photometry to eliminate unknown sources of variation, and I try to cover the whole range of image depths I may encounter, even ones with SNR under 100. Then I let the lamps run for about 10 minutes before I start the test. The lamp intensity tends to increase until the lamps reach thermal equilibrium. Then I take my images by increasing the exposures in 1 second steps until I pass the saturation point which for my camera is about 42,000 ADU even though the camera has 16 bit resolution (SBIG ST7E). Then I take the sequence in reverse order. I average the values of the two images at each exposure duration which tends to even out any systematic change in lamp intensity over time.
In the past I haven't averaged the results of multiple runs , but that is also a good idea.
Brad Walter, WBY
Arne,
Could you elaborate on this process more and clear up any misunderstandings? what I understand from your post is: (my ccd is KAI-11002 - abg, .9e/adu gain, and full well of ~65k adu, 16bit)
1) Take series of light frames with uniform illuminated light source at brightness resulting 100adu/sec until saturation (wouldnt this 600 sec not 60 sec?)
2) Capture multiple series using same exposure time
3) create a dark frame for each data point (exposure time)
4) subtract each dark from the corresponding light
5) average each exposure time and plot results?
yes, you are right - for 100 ADU/sec, you would need around 655 seconds to reach saturation, if the sensor remained linear throughout the range. The count rate chosen was for systems with a mechanical shutter. If you are using a CMOS sensor, or an interline transfer sensor like the KAI-11002, where electronic shuttering is used, you can use subsecond exposures and increase the count rate so that saturation is reached in a minute or so.
I'm out of town right now, and can give a more detailed description of the linearity test when I get back. Berry & Burnell give a nice description of the process in their AIP book.
Arne
Ill look…
Thanks Arne!
Ill look forward to hearing from you. My biggest misunderstanding is regarding darks. Seems like bias would need to be subtracted to get a proper floor. But do darks scale with exposure linearly? If not, seems darks with the same exposure as the associated light would need to be subtracted as well? Am I over thinking it....Anyways, Ill patiently wait until your return. Safe travels!
Arne, see my results below of linearity test. KAI-11002 has reported gain of .9e/adu, full well of ~65K, but is abg. Looking at the results below, do you think linearity is good enough for photometry? R squared = .9999 per Excel
awww guess aavso didnt allow the snippet of my graph, oh well, sorry
While I agree fluorscent lights are a poor choice, my procedure is different than Arne's.
No need for darks or flats. A precise linearity break determination is not needed. Just the approximate area of the break. I adjust the light for ADU counts of around 1000 for 0.01 second exposure and then increment up from there until the linearity breaks. Typically it breaks with just a few seconds exposure.
Jeff 187283
HPO
Jeff's use of .01 seconds exposure is fine if you have an electronic shutter, ie a frame transfer device or a photometric shutter. However, if you have an iris shutter, it won't respond that fast. Check the speed of your shutter. I try to never go faster than 2 seconds with mine.
Gary
WGR
Gary is right of course. If the camera cannot take images faster than 2 seconds then my method will not work. But I think most CCD cameras can take very short exposures, some down to a millisecond.
One thing not mentioned is the spectral response of the camera. What may be 40,000 counts at the center of the V band may be very different in the B, R and I bands. So it might be good to do this with a filter in place and do it for each band.
Jeff 187283
HPO
To: Tim,
Thanks for bringing this up. I've been meaning to do this on my anti-blooming camera for awhile. I imagined that it was less than 32K ADU but it turns out to be more like 20.5K!
CCD Autopilot 5 has a routine for testing linearity, as long as you can get a dark subtracted 5-sec image with an ADU of 5K. I used my cousins B filter and a flat field box on low power to get there; took over 20 minutes to finesse this. I enclose the results in the CCDAP provided text file; note I manually wrote in my CCD type and other non-generated parameters over the ******line.
James