For the Aluma 4040, I understand that exposure times, temperature, and imaging mode of dark frames should match the corresponding light frames. My question: once such matching darks are taken, for how long can they continue to be applied? For a week? A month? You see, I take 4-6 different exposure times for different targets in a night, at 300-900 seconds, and if I have to take 9 dark frames to make masters, at each exposure time, every night--well you can see the problem. I don't really expect CMOS darks to last 3-6 months as for high-end CCDs...but if the answer is that CMOS matching darks really must be taken every night, then I can't use a CMOS camera, ever, I guess. This is for photometry, monochrome, high-quality flats taken, chip chilled to ca. -10C, mode very probably to be High Gain StackPro. Any and all guidance welcome, esp. from any photometrists using a CMOS like the 4040. Thx.
Really? No one? (here or at AAVSO, either, so far) How can anyone do CMOS photometry without knowing this? Well, ok--guess I'd best keep my CCD camera working, and cancel my scope and camera plans.
I was travelling when you posted the question, so I missed it. I would have hoped maybe a customer might have provided an opinion. My advice would be identical to that for a CCD camera. The main causes of degradation (cosmic rays) and the detector (silicon) are identical. It's just the details of reading out the sensor that are different. I typically reuse dark frames for several months. Usually by the end of that time I don't see any difference, but just as a matter of course I reshoot them anyway. An obvious thing to do to monitor the quality of your master darks: take a dark frame at temperature and subtract the master. Look to see if there are any outlier pixels or increase in noise. If so, then you will want to reshoot. For flats the most likely cause of a change is a dust speck getting on the filter. Dealing with that is no different from a CCD.
Thank you very much, Doug. Years ago, a couple of early CMOS adopters (not SBIG cameras) had insisted that darks be taken the same night as the lights, for photometry. That struck me as something that would improve over time and engineering effort, as many things did with early CCDs. To my knowledge these early adopters never updated their advice, so I was asking photometrists for any update. So your advice that darks are more stable than that is very welcome. Always great to hear from someone with close technical knowledge of these things. Onward to CMOS. Thanks again.
It likely has to do with bias stability. That is a problem for many CMOS cameras even today. You will often see bias level shifts from image to image, and especially over ambient temperature changes. We've gone to some lengths to optimize the bias stability in our cameras, which is one of our key advantages compared to low-cost suppliers.
Related to the original question but I've just received the camera that will go out to SRO in a few months so I'm trying to get familiar with it at home before hand and use it on my 12.5" RC. Im very comfortable and familar with the CCD line of cameras as I've used them over the past 25 or so years. CMOS is a whole nother story. And this CMOS, the AC4040 BSI is a total stranger to me. As weather isn't very good this week I thought maybe I'd take the time to take dark frames and get a head start. For bias I need to be at focus which won't happen yet so darks I'll do. So is there a quick start setup that can get the ball rolling before I take a deep dive into the manual? It's far easier, I thhink, to get familiar when you can go out and make adjustments versus being 3-4K miles away and remoting in. Thanks for any basic starter pointers.
They are a different beast! A few pointers: For calibration, do not use bias frames. Use darks, flats, and flat-darks matching the flats. I suspect the StackPro subexposure length should be made longer on the BSI sensors. Try to run the sensor at the lowest practical temperature. They have more dark current than the FSI version When doing flats, use a shorter, relatively low ADU exposure and stack more frames. When imaging, use a very large dither amplitude, like 50-100 pixels if you can. This helps mitigate any fixed pattern noise that comes through in the background. When doing short exposure for centering, focusing, etc. use an Auto Dark. There is more pixel-to-pixel bias variability in these sensors, so you can't see faint targets very well until you subtract a dark. I recently put a 4040BSI in my back yard observatory to give it a good try. Naturally the weather has been absolute crap ever since. But there are some things I want to experiment with. One of them is to lower the gain from the default in High Gain mode. Gpixel optimizes the gain for HDR mode, and I'm not sure that's the right choice for High Gain StackPro.
This nice video shows some of why CMOS sensor tech shouldn’t be treated the same as CCD. I can’t recall to whom I’ve already passed this along. There is certainly a difference with respect to bias.
All the CMOS sensors that we've worked with behave very differently on short exposures versus long exposures. Not so with CCD sensors; they always work the same way. And that has a big effect on bias. A lot of this has to do with "amp glow". CCD sensors have a handful of transistors - sometimes literally just a couple - that operate relatively slowly (1 to 20 MHz typically) and at relatively high voltages (often 15V). The high voltages generate LED effect in the transistors, producing light in the corner of the chip. On the other hand they generate negligible amounts of heat. The LED effect is easy to eliminate by reducing the amplifier bias voltage during a long integration, and turning it back up a few seconds before the end of the exposure. This has no impact on the bias behavior. That's why bias frames work. CMOS sensors contain millions of transistors, operating very fast (often over 100 MHz) but at low voltages. They don't generate significant LED light, because they always operate at very low voltage. Instead they generate heat - lots of it, because they're operating so fast. The "amp glow" is caused when all that generated heat increases the dark current. There are transistors at every pixel, plus at the edges of the array there are massive numbers of transistors for the A/D converters, driver logic, interface logic, etc. These things are all clocking very quickly and making tons of heat, and therefore glow. A sensor I'm working with right now increases the camera's power consumption by half when it reads out - that's a lot! On top of all that, most CMOS sensors operate on a frame cycle. When you need an exposure longer than a frame cycle, you have to temporarily disable the cycles. This also allows you an opportunity to lower the heat generation by powering down large parts of the chip, including reducing drive to the pixel amplifiers and completely shutting down the converters and other logic. That's how you mitigate the "amp glow" on a CMOS sensor... but it also radically alters the distribution of heat across the sensor. The massive circuits top and bottom cool down, altering the pattern. On top of that, CMOS sensors often have circuits that do things you'd never see on a CCD sensor, like lower the bias level to compensate for higher dark current. That deliberately makes the bias level variable. Given all that, the bias level on a long exposure will look absolutely nothing like the bias level on a short exposure. So bias frames don't work at all.
(Sorry if this goes on, but believe me--this is the short version.) We've been hearing for nigh a decade now that "CMOS shouldn't be treated the same as CCD" and "CMOS sensors...do things you'd never see on a CCD", yet we still don't have a quantitative calibration protocol for quantitative CMOS photometry. I'm very glad that engineers are paying attention to the details as Doug describes. But as a user I definitely do not want to know--at all--what goes on inside the CMOS. It can be Legos powered by mutant Martian slime mold for all I care. We need CMOS cameras to work predictably, as a black box, so we can get our own work done. Specifically, we need to know exactly how to calibrate CMOS quantitatively. For heaven's sake, at least a starting point. One way to do it. Generalized across exposure times. Without user tinkering, experimentation, crossed fingers, or eye of newt. And not what calibration doesn't work. CMOS is not better for astronomy if its images can't be calibrated straightforwardly and with confidence. I get that no one individual, company, or observing organization is responsible for generating and publishing calibration protocols. I get that CCD calibration best practices didn't appear in a day, either. Certainly SBIG is doing much better than most to help. But even so: overall, for quantitative low-light photometry, the current rumor-driven, contradictory, and confused CMOS state of affairs is appalling, just appalling.
My best advice so far: Acquire all frames at the exact same sensor temperature and exposure mode. Never, ever use bias frames. Acquire dark frames matching your light frames. Acquire flat-field frames. (Tip: GPixel sensors seem to prefer lower ADU levels in their flats - it helps with fixed pattern.) Acquire dark frames matching your flat-field frames. Perform calibration as usual. Unfortunately there are some things that CMOS sensors do that may work against you. Many SONY sensor models have the ability to adjust the bias level automatically on every exposure. They do that based on the optical black pixels. The thing is, when you do this zero is no longer zero - it depends on the dark current. With longer exposures and higher temperatures the average background level is higher, and the sensor automatically pulls that back to the baseline. That has a serious potential to make the flats not work properly. That's why I recommend using the same sensor temperature for the flats as for your lights - it minimizes the impact of any baseline shift. For one sensor I'm working with right now, SONY actually defaults that feature OFF, but strongly recommends turning it ON to optimize sensor dynamic range. I'm not doing that!!! It might be the right thing for a DSLR, maybe, but is definitely the wrong thing to do for an astronomy camera! You want a stable bias level that is identical in all frames, even when there's non-zero dark current. I have no idea what choice other manufacturers are making. Also, exercise some care about temperature. Ideally you want the sensor to be regulated within 0.1C. Whenever a CMOS sensor reads out its temperature goes up. That is unavoidable but the key thing is consistency. Don't run a series of 100 back-to-back 0.1 second exposures to focus and then immediately start running long exposures (dark or light). Let the temperature of the sensor re-stabilize first. Also bear in mind that some camera manufacturers may not understand the subtleties of temperature regulation with these devices. I'm going to some lengths in my designs to try and ensure as stable a temperature as possible, but I don't know what other manufacturers are doing. I'll also recommend that you let the camera stabilize thermally for a while after it reaches its operating temperature, before doing science images. Also be aware that in some cases there can be a risk of ambient temperature coupling affecting dark calibration, because the amount of heat being pumped is a variable based on ambient plus the sensor itself is generating quite a bit of heat. It really depends on where in the system you are measuring the temperature. We've had to go through some serious hoops to handle this on one of our camera models, to ensure that the calibrations would be accurate. It may be a good idea to compare otherwise identical dark frames taken at different ambient temperatures to see whether your particular camera exhibits this sort of drift. Some do, some don't. One other thing... many CMOS sensors tend to have some line-to-line noise. That's because they're sensitive to very subtle, very low frequency power supply noise. That stuff can be nearly impossible to completely filter out, unless you have a capacitor the size of the camera, so many cameras will show a (hopefully) small amount of this line noise. Some are better than others of course. I'm planning to add an algorithm to MaxIm DL for removing this by looking at median background values line-to-line. It will have to wait until the team is done with the x64 port though.
I just started using my AC4040bsi and am grappling with the 4 quadrant pattern. Reading the above I will lower the ADU on my flats. I am using a flat panel and therefore fix my exposure beforehand so that I can match my flat darks exactly. I am wondering if I should introduce a delay between flats to allow the camera’s temperature to stabilize? I noticed the temperature fluctuations you talk about as well as how hard the cooler is working (my ears are still ringing). Lower left quadrant is brightest, followed by lower right, and 2 top quadrants are similar. My logic is to do my best with the acquisition before acrobatics in processing for which I will have to seek help from perhaps Adam Block. Next is the number of darks. With my STX I used 20-30. I wonder if a higher count would work better? Lastly, I am wondering about the levels of gain. I took 2 pics to date. One high gain 480s Stack Pro on faint planetary Abell 35 in narrow band, full moon, which by luck ended up in one quadrant due to my choice of guide star. The second was M13 in RGB which saturated (CDK14) at 2 seconds at high gain, so I tried low gain, which at 60 seconds was about 1/2 saturated which really surprised me. The quadrant pattern was still very much present however. On a star cluster, black clipping, though I don’t like it, works. However, on faint nebulae this will not work. I am wondering if I should lower the high gain, even though it did not work, as high gain seems crazy high, to tame the beast as much as possible. But lower it by how much? All your warnings about it being a different beast were warranted, as I would have thought I had a dud. The learning curve is there for sure, and I pity the guy who waits for their 16803 to die before getting on board the CMOS train (wreck.
I'm still experimenting with the AC4040BSI in my back yard observatory... unfortunately we've had horrible conditions here for months - literally a few hours of good conditions in all that time. Right now we've got massive smoke from Quebec forest fires. We've gone to some effort to fully stabilize the temperature of these cameras, but in general for CMOS it's not a terrible idea to have a pause between short exposures. The chips do generate more heat when reading out. For the darks, try taking two sets of 20 exposures each and stack them separately. Then subtract the two stacks from each other using Pixel Math (add a constant of 1000 to prevent hitting zero). Then use the Information window to measure the standard deviation (noise) in the result. Then repeat with bigger sets, say 40 each. See if there's a significant improvement in the noise. If there is, use more... if not, don't. I don't really recommend using the Low Gain mode for astronomy unless you're doing HDR. One thing I've been trying is using High Gain but adjusting the gain away from the factory default setting (Camera Settings). The choice of High Gain setting by Gpixel was based on optimizing HDR mode. I don't think it's optimal for High Gain mode (or High Gain StackPro). I think you could reduce the gain from the default, get more dynamic range, and still have more than ample sampling. I've been experimenting with that in my observatory, but haven't had enough clear weather to validate the idea. Also we're doing some testing to see if there's a better way to mitigate the cross pattern. Crude early experiments were promising.
I should mention that one of our customers experimented over time and discovered that being able to adjust the mean background ADU for flats on a per filter basis was a way to mitigate the problem. I don’t exactly get it, but I think Doug does? Anyway I have an updated ACP AutoFlat script that allows you to provide background ADU for each filter (one-time setup). Ask (on our forum) and ye shall receive, and will be in the next release.
Thank you both for this information. On my next visit to my observatory, will add the following changes to my work flow: 1) Reduce the ADU count for flats from 50% to 30% saturation 2) Carefully manage 1) through the filter changes 3) Pause for 10s or more between flats to allow the camera time to stabilize it’s temperature 4) Experiment with batches of darks to see where they “plateau” to determine optimal number of dark subs 5) Reduce high gain 15% I will see where all this takes me and will report back (in a month or so). Clear skies.
You might also want to try 10% level flats. You'll have to take more of them to average out the noise.
For my CCD camera, I needed 25 flats at 50% ADU level. Four minutes and done. But to equal that total ADU count with the AC4040M at 10% level, that would require 25 * [(50% * 64K) / (10% * 4K)] = 2,000 flats, per filter. For four filters, flats taken weekly, this is a Terabyte of flat images per month. This does not seem reasonable.
I reduced the High Gain to 1400, reduced the ADU to 10% on the flats, took 20 each of darks and flats, and all the craziness is gone! Albeit on very short exposures on a globular cluster. If the weather changes for the better I am anxious to try narrow band. It’s almost as good as my beloved STX. Thanks for your help.
I've won somewhat more accurate AC4040M panel flats by inserting a 5-second delay between each flat exposure. This did require slightly modifying ACP's AutoFlat script, but it's been worth it. 50 flats to stack per filter seems to work well, which still amounts to only 5 minutes per filter--quite acceptable time investment. My flat exposures have to be very short (0.1 sec, except 0.25 sec for Sloan i'). The between-exposure delay seems to help the chip's temperature etc. stabilize between exposures. Doing it this way, I get pretty good panel flats at about 40% of max ADU (i.e., about 1600 ADU level), which helps both (1) to get the total ADU count up faster (than at 10% ADU) and (2) to minimize any effect of flat darks. Thought I should mention this.
