Ed Wiley suggested that this new topic be created, because there was a lot of general discussion about CMOS cameras in his post from last year "Call for Action: CMOS Photometry", in which Ed suggested that it would be useful to have a manual or guide for photometry with CMOS cameras.
The last post in "Call for Action: CMOS Photometry" prior to Ed's suggestion to create this new topic, concerned gain settings and binning in CMOS cameras.
I copied that post into a document which I've attached. No-one (as of a few minutes ago) has yet posted a resonse to it.
Roy
Thanks Ken,
I'll just need to wait for clear nights. I leave in Brisbane, in the central part of the east coast of Australia. Lots of cloud this time of year.
Roy
Just getting back to CrossoverManiac's post #38, and his concern that binning in CMOS cameras that average the counts from the binned pixels may mask saturated pixels:
One way to avoid that worry would be to take a trial unbinned image and look for saturation prior to taking the binned science images.
Roy
I've updated the document on testing of the ZWO ASI1600 monochrrome CMOS camera posted 6 days ago.
The updated sections are BV transformation coefficients and photometric accuracy at the end.
The bottom line is that, particularly for transformed V magnitudes calculated using standard stars with a range of colour indices, the accuracy is very good. Of 17calculated V magnitudes, 14 differed by less than 0.01 mag from catalog. The other three differed by 0.01 to 0.015 mag. These were individual measures, not averages of multiple observations.
Roy
Hi Roy,
from your doc
Note concerning counts of more than 60,000. This is a 12 bit camera. Therefore the native full well depth at unity gain is about 4096 counts (ADU) per pixel. The camera software however has apparently been set, in RAW16 image mode, to multiply the native counts by 16, hence the high counts seen in image statistics.
Think of it like a violin concert: A concert recorded with an analog digital converter (ADC) which had only 12 Bit, has the quality of 12 Bit. Now if one pushes this into a 16 Bit wav file, the file format has 16 Bit, but the information cannnot be made better with this.
I think with photometry it is the same. One would not achive more to put 12 Bit into 16 Bit Files. The starlight was already reduced to 12 bits!
This RAW16 mode file format is questionable?
Maybe you can make comparisons with other 12 Bit cameras?
regards Bernhard
Hi Bernhard,
I thought it was important to describe what actually happens and the terminology used by ASI. I was not suggesting that the camera has 16 bit characteristics - clearly it does not. 'RAW16' is the name used in the camera software interface for one of the image formats. I am very tempted to regard this as a marketing ploy, because it could perhaps be interpreted (wrongly) to imply 16 bit characteristics.
I have no experience with the use of any other 12 bit cameras, so cannot make comparisons. However, my experience over several years with DSLR photometry was very useful when transitioning to the ASI camera.
Roy
The update to the document is in the section on gain settings between 1e/ADU and 5e/ADU, increasing the dynamic range and allowing more precise photometry with bright stars (because longer exposures can be used without saturating the sensor).
Roy
Hi Roy,
You attached a light curve of FR Cet, not a final Word document. Thanks!
Arne
I am wondering what kind of ADU values to expect from my newly acquired ASI183mm Pro camera. Anyone measured ADU values for these dark and bias frames for this camera? Higher than CCD darks? Lower?
Ed
Hi Ed,
The bias and dark levels vary from sensor to sensor, and from vendor to vendor. So they don't mean anything much. For the ASI183mm-pro on BSM_NH2, at -30C and binned 2x2, the bias averages 656, and a 300sec dark has about 2.7 counts of signal above bias. Much of that dark is due to the small amount of amplifier glow with this sensor. Your values may be higher or lower.
Arne
Arne,
Many thanks.I was getting what I thought were rather high counts (ca 700ADU) on both my darks and bias frames.
Ed
I took some images of M67 with my ASI183MM-Pro. If you or someone else wants to take a look, I can send them out. I think I might also have issues with the transforms, at least the ones concerning the Ic filter.
Thanks Arne. Full document now attached.
Roy
Thanks Arne. Full document now attached.
Roy
Thanks Arne,
I was getting a value of 750ADU on my 10 seconds darks but only 450ADU on my CD (NABG, 1603ME sensor).
Ed
This is a very interesting discussion. Thanks to Roy for posting the ASI1600 paper - it is fun reading!
Here at Hawkins Pond Observatory (HPO), I'm testing the ZWO ASI 183, 294, and 1600 cameras, along with the QHY600 and a little bit of the FLI KL400 (these two latter cameras in collaboration with Gary Walker). The 183 has been used at BSM_NH2 for about 17 months with good success. CMOS can definitely be used for precision photometry. That said, CMOS is more complex, with more parameters to set and different artifacts than CCD. I'm doing a lot of testing in my lab, with results to follow in the next few weeks. The current camera of interest is the QHY600, also being used by Josch. There are lots of things to consider with it to get the optimal photometric performance, and I'm not there yet.
Arne
Arne,
How is the CMOS testing coming? I'm most interested in your impressions of the ZWO ASI 294. My STT-8300 has developed oddball electronic problems yet again, and rather than sink another grand or so into it I'm looking at CMOS cameras. I'm leaning towards the QHY294M since my 1.25" filters will fit in their filter wheel (ZWO can't handle filters with a body > 7mm and my Astrodons are 7.6).
Shawn
CMOS cameras can have very small pixels, which could result in over-sampling. Page 18 of The AAVSO Guide to CCD Photometry states:
"If you are averaging a FWHM of less than 2 pixels, you are probably under-sampling. If the FWHM
of your seeing disk is more than 3 pixels in diameter, you may be over-sampling. Either situation
could pose problems for the accuracy of your photometry, though under-sampling is much worse
than over-sampling."
I'd be grateful is someone could explain why over-sampling could create problems for the accuracy of photometry.
Roy
I took the Exoplanet Observing course, and the manual said FWHM should be 3-5 pixels across. I don't why that is the case with observing exoplanets. I know why the DSLR photometry manual says to spread the FWHM across 8 to 30 pixels - because of the Bayer matrix. Maybe the wider spread is for the smaller drop in magnitude (< 10 mmag for some target stars) is there to gather enough light so that the drop in magnitude is less than the noise. I'm just guessing why there is a difference in the width of the FWHM.
The reason I can think of is read noise and background-noise:
The read noise from the specification of your camera is per-pixel, and if you spread the light over more pixels than necessary/optimal, you are accumulating more read-noise than necessary/optimal.
I guess the same goes for the *noise* that is introduced by the sky background. What counts is the signal to noise ratio. After overcoming losses from undersampling, the "signal" part is fixed (the total starlight) and by spreading it over excessively many pixel, you can only increase the noise-part of the equation.
I guess if you take the idea of oversampling to extremes and do the thought experiment of spreading the light of a star evenly over the entire sensor, it's easy to see that this is not optimal, spreading it over "too many" pixels is bad for pretty much the same reasons.
Cheers
HB
Bikeman wrote: "The read noise from the specification of your camera is per-pixel, and if you spread the light over more pixels than necessary/optimal, you are accumulating more read-noise than necessary/optimal."
And: "After overcoming losses from undersampling, the "signal" part is fixed (the total starlight)"
But spreading the starlight over more pixels allows longer exposure and therefore more signal as well as more noise. Since S/N ratio increases as the number of photons increases, this is beneficial. In my situation (near a city with high background sky signal) shot noise predominates, hence read noise from more starlight pixels will not be as important. And I use a CMOS camera, which has lower read noise than a CCD.
One final quote from Bikeman and a comment: "... if you spread the light over more pixels than necessary/optimal ..."
I'm not sure about the word "optimal", presumably referring to the application of the Nyquist criterion. What Nyquist determined (if my interpretation is correct) is that a periodic signal is faithfully represented by a sample which is at least twice the frequency of the signal. My emphasis here is on the words "at least".
I realize that I asked a question, and am now arguing with the answers given. There were two reasons for asking. First, I genuinely want to know why, for a CCD camera, and particularly a 16 bit instrument, oversampling may adversely affect photometric accuracy. Second, for a 12 bit CMOS camera, my belief (and persisting cloud where I live prevents my getting to the experiment to test the belief) is that, for the right target and equipment, oversampling can improve photometric precision.
Incidentally, for my camera (ZWO ASI1600MM) and 120mm refracting telescope, I have about 3 pixels per FWHM unbinned, which is pretty much spot on for the proper application of the Nyquist criterion for a circular target.
Roy
Roy,
I have the same camera as you - just arrived this week and am playing with it using the ASICap software. I am planning on having my students do a project to measure read nicse, dark current, gain, and linearity.
What I am getting confused on is the scaling of the ADU values. As has been stated before, the 12-bit A/D (values 0-4095) are getting multiplied by 16 so that the full 16 bit range (presumably 0 to 65520) is saved to the fits files.
The question is what scale is used in the specs for the camera and the graphs on page 8 of the manual for ADUs? Is it the native ADUs or the *16 ones?
--- Dave
ps. I wish they would just NOT scale the raw values.
I'm not sure what you're asking but if you're asking about the gain setting, the decibel value is 0.1dB and it is the native ADU count with the actual gain (e/ADU) being 4.95 e/ADU at Gain 0. Unity Gain at 139 is
gain (e/ADU) = 4.95÷10(Gain÷200)
gain (e/ADU) = 4.95÷10(139÷200)
gain (e/ADU) = 4.95÷4.95
gain (e/ADU) = 1 (unity)
I may have answered my own question.
If the nominal full well is 20,000 e- and a gain setting of 0 cooresponds to a gain of about 5 e-/ADU, dividing those two gives about 4000 or DUH, the range of a 12 bit A/D (0-4095). That means that the specs of the camera are based on a range of 0-4095 of ADU (even through their software and I assume the ASCOM driver multiply ADU values by 16 in the program histogram displays and the fits files saved).
That explains why the read noise is so high at low gains - it is mostly quantization error from the A/D conversion.
Dave
Hi Dave,
The top panel (FW(e-)) of the graphs of the camera's performance from the manual has the y axis in native ADUs, as you have noted. Thus, projecting the unity gain value up and across to the left intersects the y axis at 4k in the top panel.
I'll have to leave it to others to comment on any relationship between read noise and quantization error at low gain settings.
Roy
Dave,
Just thinking futher about this. Readout noise can be calculated for a sensor from the gain and two bias frames. Thus, readout noise = gain x (bias1 - bias2) / square root of 2. The units are electrons r.m.s.
There are no signals from incident photons in this calculation, and therefore no quantization error, even at low gain settings. Therefore, I don't think the higher readout noise for the camera at low gain settings is due to quantization error.
Roy
Look at it this way. If the real read noise in electrons is much lower than the gain (which it is in the case of a gain settting of 0), then the effective read noise is much higher because you cannot measure it because the A/D bit-step is too large (quantized).
In the case of our (you and me!) camera, the gain is about 5 e-/ADU at gain of 0, so it cannot measure quantiities smaller than that without large bit flips showing up as noise. I don't think it is a huge suprise that the stated read noise is close to the A/D resolution.
I have my students measuring doing a project in the next few weeks to charactarize the camera at a bunch of settings (measuring gain, read noise, dark current, linearity, etc.). I'll post the results here when I have them.
Dave LDJ
During a previous discussion in this Forum on procedures for imaging for photometry, Ken suggested (in post #42, 16 February 2020) that settings for a CMOS camera should be done in the following order:
I promised to experiment based on those suggestions. I imaged a field in Puppis near the zenith on a recent very clear night. It contains stars up to 5th magnitude. My image scale is 0.86 arc secs/pixel. The seeing was 2.6 arc secs.
Most importantly, I set the gain initially to 5 e-/ADU, the lowest setting. There was no need to bin or to defocus to achieve the Nyquist number, as the ZWO camera is a good match for my 120mm f/7.5 refractor, with 3 pixels per FWHM in a well focussed, unbinned image when the tests were done on 1 March 2020.
Exposures of 15 seconds and 30 seconds were taken.
Then I re-set the gain to unity (1 e-/ADU) and took another set of 15 second exposures.
The attached file, Peak ADUs.png shows that stars brighter than magnitude 8.9 either saturate the sensor or have peak ADUs above the linear range of the camera, even at a short exposure of 15 seconds. At a longer exposure of 30 seconds and higher gain (1 e-/ADU) the population of stars in the linear range of the camera moves to fainter magnitudes.
To perform photometry on the brighter stars, images would need to be defocussed.
Differential v (instrumental) mag photometry was performed on pairs of stars in the well-focussed images. The results are:
Four pairs of 8th and 9th mag stars, gain 5 e-/ADU, exposures 15 seconds
n=6, SDs 0.022, 0.015, 0.008, 0.012
Three pairs of stars, mag range 9.8 to 11.1, gain 5 e-/ADU, exposures 30 seconds
n=10, SDs 0.007, 0.006, 0.014
One pair of stars, magnitudes 10.5 and 11.1, gain 1 e-/ADU, exposure 30 seconds
n=10, SD 0.023
Previous results, on defocussed images, gain 1 e-/ADU, various exposures
Six pairs of 11th mag stars, exposure 180 seconds
n=10, SDs 0.008, 0.007, 0.007, 0.009, 0.006, 0.007
Six pairs of 10th mag stars, exposure 180 seconds
n=10, SDs 0.006, 0.006, 0.006, 0.003, 0.006, 0.006
One pair of 8th mag stars, exposure 150 seconds, on two consecutive nights
n=10, SD 0.003, 0.002
Conclusions
In my hands defocussing and using longer exposures at a gain of 1 e-/ADU give better precision than using well-focussed images and setting the gain to 5 e-/ADU.
In well-focussed images, even at the lowest gain (5 e-/ADU) and with short exposures (15 seconds) bright stars are either saturated or peak ADUs are above the linear range of the camera.
In view of the above, I do not believe it is appropriate, with 12 bit CMOS cameras, to preferentially set the gain to the highest dynamic range. In my camera, that setting means >1 e-/ADU. For fainter stars, you need all the electrons you can get. That is best achieved at unity gain (1 e-/ADU).
Setting the gain to >1 e-/ADU in 12 bit CMOS cameras is appropriate and and may be necessary for bright stars, with defocussing of the image to spread the light across more pixels to avoid saturation. In this situation tuning the gain, degree of defocus and exposure achieves the best precision (lower gain and more defocus allow longer exposures).
Roy
Hi Roy:
Thanks very much for your experimentation! I've been wrestling with your results before commenting. It is interesting, but perhaps reasonable, that the better precision exists for the 1 e-/adu setting. What snr values were you getting for 1e-/adu vs the 5e-/adu when you tried both? More signal would overwhelm noise and lead to smaller SD?
Somewhat embarrassing to say but I continue to get confused by the different terminology used for "gain". Gain may be described in terms of e-/adu or some different numerical value like zero, one, or 100-300. What the heck is gain zero? Generally, through my Google search for "cmos gain settings", it appears that gain zero corresponds to the point of maximum dynamic range that I have proposed is best for photometry (despite the fact that my calculation indicated that occurred at 3.6e-/adu for the ASI183 camera)?
In any case, I continue as a practical matter to dislike the idea of defocusing as the preferred way to deal with saturation. You seem to be a little reluctant to use "short" exposures of a few seconds or even tenths of seconds? They do pose a problem due to scintillation but that can be resolved by taking multiple images and stacking. I just think of defocusing as a normal technique, as more an "art" than "science"?
Most of the discussions I have read so far seem to lean toward "gain zero / max dynamic range" to simplify normal photometric procedures for a wide range of magnitudes, as the way to go, although unit gain is not that far away? Although a CMOS Guide could propose a few different setting options for different observing conditions, I would rather be able to limit that as much as possible to avoid too much ambiguity and subsequent difference in precision and accuracy of reported magnitudes?
Still need more thought about this and perhaps more experimental comparisons of results at a finer gain resolution? Any interest in trying this?
Ken
Hi Ken,
S/N at gain of 5 e-/ADU for 4 stars varied from 506 to 765. S/N at gain of 1 e-/ADU for 2 stars was 533 and 849. The gain was calculated as flux(star-sky) / Square Root of flux background, based on the fact that the target signal was much greater than the background signal. I don't usually do S/N ratios and am therefore unfamiliar with the nuances (although my reading suggests that a 'proper' calculation of S/N for an image is complex), so please correct me if this method is not correct.
I agree that gain in decibel units versus e-/ADU is something to get your head around. Basically the ZWO ASI1600MM camera requires the user to select the gain in units of 0.1dB, from 0 (high dynamic range, 5 e-/ADU) through 139 (1 e-/ADU) to 300 (close to, but just above 0 e-/ADU).
The reason that I do not use multiple short exposures and stack the images for photometry is that almost all of my data is collected as time series, through the night. With multiple short exposures I would have much larger numbers of images, chewing up a lot more computer memory. I don't have software to stack 'on the fly', and stacking manually would be an impossibly long task. The alternative would be to perform photometry on all of the multiple short-exposure images then average groups of them. That can be done, but I don't have a spreadsheet routine to automate that. Rolling averages are easier in spreadsheets, but I don't need an enourmous number of data points for the light curves I draw. In any case, I would still end up with a large number of images using up memory on my hard drive.
Sorry to be long winded, but the short answer is that there are very pertinent, practical issues I'd have to face if I took many short exposures of well-focussed images.
However, my next test (when the clouds eventually clear again) will be just what you propose - lots of short exposures and stack groups of them. I'm happy to try that as a test, but could not possibly do it routinely. Ideally, I'd like to do that at unity gain (1 e-/ADU) so that every electron counts. At 5 e-/ADU by definition you need 5 electrons to end up with one count. In my view, the only reason to do that would be if you had a very bright star that would saturate the sensor at 1 e-/ADU.
As I understand it, high dynamic range (greater range of stops) simply means allowing longer exposures without saturating or exceeding the linear range of the sensor. Please correct me if that is not correct.
Yes, I'm happy to try more tests. But in the end, I can't use multiple short exposures of well-focussed images for the reasons detailed above. Also, I'm getting high precision and high accuracy with this 12 bit camera by defocussing, using 1 e-/ADU gain (5 e-/ADU for very bright stars) and setting the exposure by trial and error. I have no reason to change that, but other observers with different objectives or different software than I have may be able to use the camera differently with results just as good or better.
Roy
Ken,
The S/N figures I gave you are correct, but I described the calculation incorrectly. The calculation I used was: signal / square root of signal, where the signal is the star minus sky flux.
Roy
Just one more comment on defocussing for photometry. It has been used to good effect in exoplanet research. The authors set out to do exactly what I have been describing - defocussing the image to spread the light of the star over more pixels, thus allowing longer exposures and thereby increasing the number of collected photons, increasing S/N ratio and improving precision. A paper is attached.
Roy
Hi Roy:
I accept the fact that under certain circumstances, defocusing may be useful. However, if I may use an analogy below, I think simplicity in photometry technique (e.g., normal autofocusing, binning to match nyquist criterion, good dynamic range) is important for 'most' of our observers? It is certainly useful for remote observing (distant or backyard) where human oversight is limited by necessity or desire.
Most of us are 'cooks' who use a simple recipe/cookbook to achieve edible meals. A few of us are 'chefs' who use experimentation (touch and feel) to achieve that perfect souffle or artisan bread!
IMHO, most of our observers use/want a simple method/guide to achieve 'acceptable' photometry. A few of us are 'photometrists' who again use experimentation rather than rules of thumb to achieve that 'perfect' (accurate/precise) magnitude. While the latter may be our goal, it is not easily achieved in practice, and would be difficult to require? (This may lead to arguments!) ;-)
i propose that the future CMOS Photometry Guide err on the side of basic settings/procedures (e.g., gain for max dynamic range, binning to match seeing (FWHM)). However, since CMOS cameras are a slightly different animal from CCD cameras, perhaps (?) the Guide should also include some modifications that allow one to 'tune' settings / alternative procedures to address certain conditions. Of course, this is the tough part that leads to ambiguity and disagreement/arguments!
I am hopeful that this forum thread can lead to an understanding of both the basic settings and the more involved? BTW, I think of you as a chef!![wink](//cdn.ckeditor.com/4.5.4/full-all/plugins/smiley/images/wink_smile.png)
Ken
Ken,
I urge you to think further about 'always' setting the gain to maximum dynamic range if that means >1 e-/ADU. if you image faint stars so that even at unity gain you are within your linear range, setting to gain >1 e-/ADU means that you are throwing away signal.
Roy
For those of you thinking about purchasing the QHY600, probably the most advanced CMOS camera on the market today, I have one that I'm testing with Gary Walker here in NH. It really is a good camera, and checks most of the boxes that you might have for a digital camera.
That said, it is also a complex camera to set up. There are a lot of adjustments that you can make, such as readout mode, pixel gain, pixel offset, whether to include overscan and dark row pixels, binning, etc. Here is an interesting section of the QHY site regarding many of these parameters:
https://www.qhyccd.com/index.php?m=content&c=index&a=show&catid=94&id=55&cut=1
I've personally verified most of those plots. They highlight a point: vendors tell the truth, but select for the best values when describing this camera. You can't set up a camera that meets all of those specs simultaneously. For example, the full well of ~50Ke- for normal readout comes with the lowest gain setting, where you have the highest readnoise. If you want the lowest readnoise, then the resultant full well is only 5-10Ke-. So you have to look at the plots and find the combination that best fits your desired needs. That may be high gain because you are doing spectroscopy where low readnoise is the most important quanity, or low gain because you want to do photometry of both bright and faint stars in an image and need the highest dynamic range.
For photometry, I'm leaning towards the "high gain" readout mode with gain=0 and offset=10. This yields about 3 electrons read noise and a full well of 50Ke-.
However, another important factor with the QHY600 is that the current driver only supports 1x1 or 2x2 binning. Additionally, the 2x2 binning is done by summing, not averaging. This means for the resultant 16-bit image transferred to, say, MaximDL, 2x2 binning only uses 16K counts per native pixel, with whatever number of electrons based on the amplifier gain. In other words, say that you can accumulate 50K electrons in a single native pixel. With 2x2 binning, you still can only accumulate 50K electrons per binned pixel, but now with sqrt(4)=2x more readnoise. So you lose with binning compared with some other vendors/CMOS cameras like the ZWO ASI183mm-pro.
The best way to avoid this binning issue is to acquire images in 1x1 mode, and do your binning after the image is read out, with your own software (or through built-in or plug-in options in MaximDL). If you store the full 62Mpix 1x1 binned image, you will eat up disk space fast (you can only store eight 16-bit images in one GByte, for example, without compression). So the processing done after image acquisition becomes important, and I haven't had time to fully study the best way forward with regards to software.
The QHY600 has MANY attractive features, and is probably the best value in the market today. I and others will find good operational methods in the near future.
Gary and I will be at NEAIC and SAS in the coming months, if anyone wants to discuss CMOS in greater detail.
Arne
Thank you! I appreciate your feedback.
I'm a bit confused about the binning. On its website, QHY states that in standard mode, the full well is 51k elecgtrons and in 2x2 binning, it goes to 204k electrons per binned pixel.
If I am interpreting your comment correctly, the 2x2 binned pixel stays at 51k rather than going to 204k electrons? Does the binned gain vs. standard gain affect this? Best regards.
Mike
Hi Mike,
That is correct. With the current QHY600 driver, 2x2 binning doesn't buy you much. To get the ~204K number, you have to read the camera in its full 61MPix mode, and then bin in software after the image is acquired.
Arne
Thanks! It sounds like there is no point in binning at this point with the QHY600. Best regards.
Mike
Hi Mike,
If your goal in binning is to get greater "full well" or dynamic range, then the 2x2 binning option in the driver is not for you. If your goal is to get bigger physical pixels because of your image scale or fwhm, or to decrease the size of the stored image, then binning is still a good option. Binning with average rather than sum, and 3x3, 4x4, etc. binning, are just software addtions to the driver and not changes in the camera, so I would bet that those options will be available sometime in the future.
The small pixels of most CMOS cameras are not for everyone. They work really well with short focal length systems, such as the Takahashi E-180 (180mm aperture, f/2.8) astrographs we are now using in the Bright Star Monitor network. For a C-14 (which has ~3000mm focal length), you have to bin 4x4 or even 6x6, and then the readnoise advantage of the CMOS sensors goes away in comparison with a standard CCD.
Arne
This follows on from post #82, in which I suggested that we should not 'always' set the gain to maximum dynamic range when that means a gain of >1 e-/ADU. For the ZWO ASI1600MM Pro 12 bit camera the maximum dynamic range is at a gain setting of 5 e-/AdU. The following shows the loss of precision which can occur if maximum dynamic range is selected preferentially.
The two illustrations showing differential v (instrumental) magnitudes are probably self explanatory. The same field was imaged on a very clear night near the zenith. Three stars (mags 11.65, 11.74 and 12.24 approximately) were the targets for differential photometry, performed on the three possible pairs (11.65 and 11.74, 11.65 and 12.24, and 11.74 and 12.24).
Photometry was performed on well focussed images (no defocussing here). Two different settings were used: 30 second exposures at unity gain (1 e-/ADU), and 60 second exposures at maximum dynamic range (5 e-/ADU). 10 frames were taken at each setting. Three frames had imperfect tracking in the 30 second exposures and were not used.
The illustrations show clearly that there is better precision with 30 second exposures at 1 e-/ADU gain. Of course, precision could be improved at the 5 e-/ADU gain setting by further increasing the exposure, but why do that when better precision is obtained with shorter exposures at unity gain?
Roy
Following on from post #88, I processed the last results from the series of tests on this issue. The attached compares the standard deviations (SD) of differential v (instrumental) magnitudes of 11th and 12th magnitude stars at maximum dynamic range (5 e-/ADU) and exposures of 60 and 120 seconds with the results at unity gain (1 e-/ADU) and 30 second exposures. The graph for the latter looks good but that is deceiving; concentrate on the SDs adjacent to each plot.
The bottom line from the results is that precision at maximum dynamic range (5 e-/ADU) and exposure of 120 seconds is essentially the same as the precision at unity gain (1 e-/ADU) and an exposure of 30 seconds. These were well-focussed images. Note however that, even at maximum dynamic range the brightest star in the linear range of the sensor at an exposure of 120 seconds was mag 10.5. For photometry on brighter stars stacking multiple shorter exposures (and that would be a lot of exposures) (or averaging the photometry from them) would be needed, or the images could be defocussed. My personal bias is that it would be much easier to defocus and lengthen each exposure, but of course for all but the brightest stars I would do that at unity gain.
The above applies to my 12 bit camera. Perhaps photometry could be performed on brighter stars with a CMOS camera having a greater bit depth without resorting to stacking etc, but I've no experience of this.
I am sympathetic to Ken's concern that for observers starting out in CMOS photometry simple procedures (such as starting off with maximum dynamic range) and not varying the gain would be easier to understand. The downside to this, for well-focussed images taken with a 12 bit camera, will be a limitation in the range of stellar magnitudes that fall within the linear range of the sensor for single exposures. Multiple shorter exposures (and that could be many) should deal with that problem (although I've not tested the precision), but as I noted earlier in this Forum that raises issues for observers using time series photometry. Software modifications in the future may remove this problem, but we're not at that point yet as far as I know.
Thus, two ways to go, both I suspect of equal validity, but both needing care to get the best out of the cameras we use.
Roy
If one looks seriously at the contents of the AID, I believe it's clear that the main problem with AAVSO photometry is not inferior equipment, but poor technique. CMOS cameras will do nothing to change that. If anything, the greater complexity of these sensors makes more room for errors.
If half of the enthusiasm for embracing CMOS technology was directed, instead, to helping observers "up" their game, AAVSO would be much better off.
Tom
Tom,
Perhaps you should think about a new thread adressing your concerns. Might be a valuable forum for discussing such issues. You could define the "main problem" and we could all discuss the subject. For example, is the main problem uncertainty or accuracy or both? What steps could be taken to improve the photometry? Etc.
Ed
Hi Tom,
I can't let your comment, which I think is a little unfair, pass without a reply. The first point is that there have been views expressed in this Forum which do not support a rush into CMOS cameras. The second is that most if not all of the posts I have made to this Forum (ad nauseam, I would have thought) concerning my experiments with a CMOS camera have addressed what I believe is one set of 'good practice' suggestions for getting the best precision and accuracy out of the sort of camera I am using. Similar techniques (apart perhaps from those involved with different gain settings?) could be used by CCD observers. The sort of tests I have used and described could, I submit, also be used (in modified form if necessary) by PEP observers.
My posts, and those of others I believe, were aimed at encouraging or seeking good observing practice, which is precisely what you have just asked for.
Roy
Hello, Roy! Thank you for your comments and work with the CMOS camera. I appreciate it.
From a personal note, the reason that I've been asking about CMOS cameras is that it has become apparent that CCD cameras will not be available in the near future. Just in the past month, a number of cameras models appear to have been discontinued as ON production ceases.
I've been mulling over how I might want to increase my ability to perform photometry. Currently, I use an 8" LX200 with SBIG ST-402. That combination could keep me going for years, if I so desire. However, if I consider expanding my equipment, it is likley that CMOS cameras might be my only option.
So, I consider at CMOS cameras, not because they are better than CCD cameras, but because they might be my only option. If so, the question for me becomes how I might want to use them and when. Are they appropriate for long focal length systems, like a C14? Or should I consider short focal length, wide field systems, like the AAVSO uses with the BSM network and scopes like the Takahashi epsilon 180?
In any case, I appreciate comments by observers with more experience than I who provide guidance. Best regards.
Mike
I finally managed to get around to this test with the ZWO camera. The night was not ideal, as I was dodging clouds. I eventually found a large area of clear sky, but the field unfortunately contained only two bright (mag 6) stars, when I was hoping for more. Results attached.
Roy
Roy wrote;
"Conclusions
Photometry performed on stacked frames at high dynamic range from short exposure images (in this test, 5 seconds) of bright stars produces good results in differential photometry. Averaging instead of stacking yields essentially identical results. In my opinion, it should be possible with bright stars such as these (magnitude 6) to obtain even better precision by varying the technique (same gain, at 5 e-/ADU, defocus the images, take longer exposures).
It should be noted that 5 seconds is the longest exposure possible with the ZWO ASI1600MM Pro camera with well-focussed images of bright stars (magnitude 6) under the conditions of this test, as the ADUs are near the top of the linear range."
Thanks for your efforts. These results look really good to me. Do you feel that you need to improve much in precision? You are a perfectionist in that regard! It is hard to object to that, but it still seems atypical of regular photometric practice, for better or worse?
If you used 0.5 sec, you could go 10x brighter (~2.5 mags) but only with a lot more images! :-(
Note that the new version of MaximDL 6.21 has a method to stack images in real-time! It is a bit tedious to use for remote observing but could be a game changer in terms of hard drive space.
Ken
Ken,
Does the Maxim real-time stacking work with time series (that is, take 10 images, stack them and save, take another 10 and stack, and so on)? If so, that's good, and I could therefore go that way, if I bought Maxim. If not it would not help me because most of what I do involves time series through the night. Maxim is not cheap. I use AIJ which is free and very fast.
It should be easily possible with bright stars of 6th magnitude to achieve very high precision with low bit depth cameras using ordinary good methods (including defocussing, which is always used by DSLR photometrists anyway). I just think that's worth aiming at.
I think one of the concerns you expressed early on about defocussing was that it was not a good idea for routine use on public access robotic systems. I agree with that. But for a setup in a private observatory it's no problem.
And as we said earlier we would not be having this discussion about defocussing if the camera had 16 bit ADC.
Roy
Roy:
Do not accept this as the gospel or run out and buy MaximDL. I do not have a cmos camera and thus have not tried this new option! Actually, I could try it with my ccd on a bright star that I wanted to stack to average out the scintillation effect. Maybe I will, BUT it would be nice for a current cmos user and Maxim DL license owner to try it and report how it works? In fact, there may already be a known bug that Diffraction Limited is working on?
BUT, based on what little I have seen, it is most efficient for time series imaging. Effectively, you select an exposure that works for your target and put that value into a "hidden fixed" stack setting box. Then you select some longer exposure that will be your total exposure time for each "stacked image" and put that longer time into the normal exposure box. Then, the software automatically takes images at the shorter fixed setting time AND stacks these images when you get to the end of the "normal" exposure time. Does this make sense? I'm not explaining it very well? One of my concern is how much RAM (or other temporary drive space?) you need to hold the intermediate images before final stacking into one stored stacked image?
BTW, for experienced users, any of the procedures you are using work fine and yield good photometry. It works very well for imaging one target over a long time but may be less effective for taking a few grab images of many targets and might be a PITA during such a long night?.
Ken
Ken,
Don't worry, I'm not going to buy Maxim - am very happy with AstroimageJ. However, your description of stacking in Maxim is very easy to understand and sounds extremely useful. Maybe it holds the images for a stack in 'virtual memory' which is what AIJ does (basically on disk I think, then pulls each image out when needed), and quite fast.
I agree that having to defocus for best precision is fine for time series, but for multiple different targets during one night would be a pain, because of checking the degree of defocus to avoid saturation or imaging in the non-linear range of the camera on one hand, and getting good precision on the other. I presume well focussed images, and known exposure times for stars of various magnitude ranges, would be much easier.
Roy
I am just getting interested in variable star photometry (based upon an inspiring talk at the ATMoB April 9 2020 Virtual Monthly Meeting) and have a hopefully simple question.
I have read a lot of comments about the ZWO ASI 183 MM on this thread. Getting into photometry very late in the game, I purchased a ZWO ASI 183 MC (color) (cooled) for a completely different purpose. It seems to me -- from my limited experience -- that the DLSR instructions apply almost completely to this camera. It has the standard DSLR configuration, including a Bayer mask. Is this correct?
Thanks for your expertise in advance,
- Alva