34 replies to this topic

[GaryShaw]

HI All:

 

In the context of photometry-related imaging, I'm finding myself in situations where Binning at Bin 2 and Bin 3 offers some advantages. The aspect of this which I have not been able to learn about is how do you determine the Full Well depth of 'binned pixels' so you can avoid saturating key stars in the image?

 

I feel that I should have learned this somewhere along the road but the subtleties of 'on chip', hardware binning versus 'software' binning has me baffled. I hope that some of you might help explain this or steer me to resources where I can learn how to determine Full Well for binned images with both CMOS and CCD sensors. The two cameras listed above are of particular interest (both CMOS) but I sometimes use CCD imagers so I'm interested to understand the issue with CCD sensors as well.

 

Thanks very much for your help...!

Gary

[pbealo]

Gary,

 

I just picked up a used ASI294MM Pro, but have not yet begun using it for photometry. It will be used with a new set of Chroma filters.

 

A starting point for Bin 1 and Bin 2 saturation comes from ZWO:

 

"Unlocked Bin1: 12bit ADC, 2.3um pixel size, 47 megapixels, 8288*5644 resolution, 14k full well capacity.
Bin2: 14bit ADC, 4.6um pixel size, 11.7 megapixels, 4144*2822 resolution, 66k full well capacity."

 

So I would think that in Bin 2 you'd be good to 55K or so ADUs. But this needs confirmation.

 

Page 39 of the AAVSO CCD and CMOS manual describes the procedure to test for linearity and saturation. You can do it indoors even.  https://www.aavso.or...ometryGuide.pdf

 

Peter

[robin_astro]

From discussions in the spectroscopy field at least, the consensus is CMOS sensors should not be binned in camera as it offers no advantages and post filtering of oversample images can be used to advantage to reduce telegraph noise inherent in CMOS. The situation is different in CCD cameras where in camera binning reduces read noise where that is a limiting factor.  My understanding is you can't increase full well depth by binning though as there is always a risk that one of the say 4 pixels is saturated. (Not seen in evenly illuminated linearity tests)

 

Cheers

Robin

[Ed Wiley]

What Robin said.

Ed

[pvdv]

Have a look here

https://www.photomet...-basics/binning

CCD 4 times the signal, 1 time the read noise
CMOS 4 times the signal, 2 times the read noise.

In practice, this is sensor dependent because the hardware can average or sum the pixels (will get back to this later).  

and here

https://andor.oxinst...y-ccd-detectors

 

The internals of the fits file of a 14-bits camera provides some insight - the acquired data is simply shifted by 2 bits (multiplied by 4) in the camera in order to get 16 bits integer values in the fits files (300 s dark)

Filename: dark300.fit
No.    Name      Ver    Type      Cards   Dimensions   Format
  0  PRIMARY       1 PrimaryHDU      34   (3008, 3008)   int16 (rescales to uint16)   
(3008, 3008)
INSTRUME: ZWO ASI533MC Pro
[[2800 2796 2796 ... 2796 2788 2804]
 [2796 2800 2796 ... 2796 2796 2800]
 [2800 2796 2792 ... 2800 2804 2792]
 ...
 [2776 2792 2800 ... 2796 2804 2804]
 [2796 2804 2808 ... 2792 2800 2796]
 [2796 2800 2800 ... 2792 2804 2804]]
Contains numbers not divisible by 4: False

Since summing and averaging 4 numbers that are divisible by 4 obviously gives a number that is divisible by 4 you can see where this is going.

Now looking at a native 16-bits camera, we get on the same 300 sec dark

No.    Name      Ver    Type      Cards   Dimensions   Format
  0  PRIMARY       1 PrimaryHDU      41   (6248, 4176)   int16 (rescales to uint16)   
(4176, 6248)
INSTRUME: ZWO ASI2600MC Pro
[[499 494 500 ... 499 505 498]
 [498 514 499 ... 501 497 489]
 [499 494 502 ... 507 511 493]
 ...
 [505 503 523 ... 499 505 506]
 [508 500 509 ... 503 510 504]
 [499 505 499 ... 498 505 506]]
Contains numbers not divisible by 4: True

See the difference?

 

Futhermore, here's what a calibration frame might end up looking like when it is actually used for calibration (it has been rescaled to 32-bit floats)
 

No.    Name      Ver    Type      Cards   Dimensions   Format
  0  PRIMARY       1 PrimaryHDU     126   (9576, 6388)   float32   
(6388, 9576)
INSTRUME: ZWO ASI6200MM Pro
[[0.00763523 0.00765812 0.00764191 ... 0.00769341 0.00760185 0.00765145]
 [0.00763523 0.00765812 0.00766575 ... 0.00762856 0.00763809 0.00764954]
 [0.00764286 0.00760662 0.00763714 ... 0.00764668 0.00760948 0.00766384]
 ...
 [0.00769436 0.00759422 0.00768292 ... 0.00762474 0.00767147 0.00765717]
 [0.00766384 0.00763714 0.00764668 ... 0.00764096 0.00764858 0.0076257 ]
 [0.0076524  0.00766003 0.0076524  ... 0.00763142 0.00764858 0.0076257 ]]

But without knowing exactly the internals of the on-chip binning and the exact algorithms used by the calibration software you can't really quantify an eventual gain in practice.

Going back to the hardware summing or averaging binning strategy (depends on camera electronics, configuration options available, etc...) there are a couple of things you need to take into account

a 14-bit camera summing could be OK, but keep in mind that since you get a sum of 4 numbers that are divisible by 4 you are still getting a maximum 16384 levels (at the very maximum and not linearly) and you do run the risk of overflowing.
a 14-bit camera averaging will not overflow, but it will still give you a maximum number of 16384 levels.

One could possibly argue that by oversampling the number of photons hitting the area, you are increasing the resolution of your photon counter but you will still end up with 16384 electrons maximum (at unity gain)

Bottom line: don't bother with the false hope of increasing actual FWC by 4 per advertisement material.


 

[pbealo]

FWIW,

 

I checked with Arne Henden, and ALL the AAVSONet CMOS imagers are binned in the camera, and I have personally seen that they are linear until almost 60K ADU, not the 15K ADU saturation for a single pixel unbinned.

 

Peter

[GaryShaw]

FWIW,

 

I checked with Arne Henden, and ALL the AAVSONet CMOS imagers are binned in the camera, and I have personally seen that they are linear until almost 60K ADU, not the 15K ADU saturation for a single pixel unbinned.

 

Peter

Great Info Peter.

 

1. Does that indicate that the binning on those cameras is 2x2 so the full well appears to be additive when binning?

2. What cameras are used on AAVSONet?

 

thank you!

Gary
 

[pbealo]

A lot of ZWI ASI183MM Pro cameras and a mix of others.

 

The binning is done on camera. Looks additive to me!

 

Peter

[Octans]

From discussions in the spectroscopy field at least, the consensus is CMOS sensors should not be binned in camera as it offers no advantages and post filtering of oversample images can be used to advantage to reduce telegraph noise inherent in CMOS. The situation is different in CCD cameras where in camera binning reduces read noise where that is a limiting factor.  My understanding is you can't increase full well depth by binning though as there is always a risk that one of the say 4 pixels is saturated. (Not seen in evenly illuminated linearity tests)

 

Cheers

Robin

The IMX492 cameras (like the ASI294MM) are a bit unusual in that the 2x2 binning mode does reduce actually seem to read noise over software binning. It's not entirely clear why, but that's at least what they show in the published measurements, so there's presumably something happening at the hardware level even if it isn't CCD-like binning.

Edited by Octans, 17 March 2024 - 02:53 AM.

[GaryShaw]

Hi

FWIW, I started binning to reduce my final image file size as recommended by one of the AAVSO VPHOT Gurus. I had been encountering multiple problems uploading 50 MB images from the ASI294MM to VPHOT so it was suggested to reduce file size by binning so long as this didn't under-sample the images. 

Gary
 

[robin_astro]

 

FWIW, I started binning to reduce my final image file size as recommended by one of the AAVSO VPHOT Gurus. I had been encountering multiple problems uploading 50 MB images from the ASI294MM to VPHOT so it was suggested to reduce file size by binning so long as this didn't under-sample the images. 

 

If you are just binning to reduce image size then  binning the original images using software is fine with no risk of saturated pixels provided the resulting image stays within the dynamic range of the format used

Edited by robin_astro, 17 March 2024 - 09:17 AM.

[robin_astro]

FWIW,

 

I checked with Arne Henden, and ALL the AAVSONet CMOS imagers are binned in the camera, and I have personally seen that they are linear until almost 60K ADU, not the 15K ADU saturation for a single pixel unbinned.

 

Peter

Peter,

 

How was linearity checked? For example if you check it using even illumination you  will indeed get to 4x the number of e- before saturation. A star however may saturate a particular pixel but not the surrounding pixels which would give lower than saturation count in a 4 pixel bin so go unnoticed. 

[robin_astro]

Here is a simple hypothetical example. You have a sensor with 100e- FWD and in camera bin it 3x3.  You do a linearity test using even illumination and it saturates at 900.  You then take an image of a star centred on the binned pixel

 

 

The central pixel is saturated so you get a total count of 600 e- so all looks well based on the linearity test. In fact however the true count is 650.

 

How much of a problem this is in real life depends on the point spread function of the star image and the exact definition of critical sampling but I would say the risk is always there.

 

There are other reasons not to bin, for example cosmics and hot pixels are harder to spot and healing them causes more damage.

 

Cheers

Robin

Edited by robin_astro, 17 March 2024 - 09:53 AM.

[Ed Wiley]

Robin is correct about binning, linearity and saturated pixels. So what to do? I suggest to my students that they be very cautious about setting their saturation level when binning. Looks like you are linear to 60K? Back off to 50K. Don't push to the limits with your integration times and beware of the possible problem. If its critical (bright target), then have a look at your image in "pixel view" where you can search for saturated pixels or image at bin 1x1 and go to bin 2x2 later after insuring that there are no saturated pixels in the target and comps.

 

The question is: how much work are you really doing at such high ADU levels? Even with exoplanets? No challenging, just asking as I am curious.

 

Ed

[pvdv]

FWIW, in camera binning can either be set to summing or averaging. Moravian is one of the manufacturers that exposes that feature clearly.

One may have the impression that a 14-bit cameras is "linear" to 60k ADU, but that isn't real dynamic range/snr etc. See how the 533 above _acquires_ data. If it wasn't arbitrarily shifting it by 2 bits to the left, we'd get 16384 ADUs at best. and the result would be equivalent, except for the fact that our software is typically designed to handle 16-bit values and, especially for visual inspection, we would be served with even darker images if only the righmost 14 bits of the data were used. 

And, to some extent, 16-bit cameras don't do much better as can be seen by the fact that a 533MC maximum DR is a bit below 14 bits or stops while the maximum DR for a 2600/6200 is a just a bit above 14 bits or stops (basically the application of the formula for FWC and read noise). Even GSENSE 4040 BSI sensors in dual reading modes don't reach an actual 16 bit of DR.

People may run whatever stats or analysis they want in their favourite applications once the data has been loaded and massaged, that doesn't change the reality of the acquisition.

And yes, it is possible to get a bit more DR on CMOS by binning, there isn't anything surprising about that (see the teledyne link above) but it is fraught with potential issues at almost every level if one doesn't know the internals of the particular CMOS sensor and configuration on top of the other issues that were raised in this thread.

[GaryShaw]

Ed & Robin

I’m checking test images as I capture them by zooming in and checking each pixel in the bright, central region. I’m keeping them generally below 50k ( on the ASI294mm)when setting exposures at Bin3. 
 

Forgive the naive questions but 1) would it make sense to run linearity testing at Bin 2 and Bin 3? and 2) when I zoom in and measure each pixel during capture, am I measuring the sub pixel or the super pixel? It seems like I must be viewing and measuring the super pixel and have no way to view, measure or determine whether any sub pixels are saturating. 

 

The 294’s default is Bin 2 and each 4.63u pixel saturates at 66k e. When I zoom in and measure one of these pixels, can I be any more confident ( than with Bin3) that none of the 2.3u hardware pixels are saturating?

 

thanks for your input!

Gary

[Ed Wiley]

Sure, Gary, run a linearity test with binning. But also think about the general situation. How much of your photometry is conducted at the limits of some reasonable ADU value, like 50K? If such activities are a normal part of your observing program, then saturation can be a real problem. If not, then not so much. Bright target? Back off on the exposure and stack a short series to guard against scintillation. 

 

Ed

[GaryShaw]

Sure, Gary, run a linearity test with binning. But also think about the general situation. How much of your photometry is conducted at the limits of some reasonable ADU value, like 50K? If such activities are a normal part of your observing program, then saturation can be a real problem. If not, then not so much. Bright target? Back off on the exposure and stack a short series to guard against scintillation.

Ed, did you mean to say the underlined text or were you intending to say that working at or below 50K wouldn't lead to much of a saturation concern? If what's written is intended, I am confused. I would have thought that keeping at or below 50k when FW is 65k, would be a comfortable margin from saturation.

 

Perhaps I should have mentioned why I'm asking this question: I'm working on Alert 842 which involves imaging AB Auriga with a HA filter. This star has a possible planet forming in the star's residual disk of gas and dust. In order to refute or confirm this, the Researchers have asked for HA observations of AB Aur to supplement observations by HST. They have also asked that we use star identifier 104 as the comp star and 120 as the check star. Star 120 has a V magnitude of 11.97 whereas AB Aur's V magnitude runs from 6.9 down to 8.4. It's also requested that the SNR of check star 120 be >100. This creates a situation where we need to avoid saturating AB Aur yet we need to achieve an SNR of 100 for a check star that's 4 magnitudes fainter. The approach to meeting these requirements seemed to suggest  that we image at an exposure value that doesn't saturate the Target star - yet is as long as possible in order to enable achieving the SNR of 100 by stacking multiple images.

 

With that background in mind, is there an answer to my Question 2 in my previous post? How would you approach ensuring/confirming that none of the stars in AB Aur are saturated?

Thank you

Gary

Edited by GaryShaw, 18 March 2024 - 04:13 PM.

[Ed Wiley]

Gary,

 

So you have a target that is brighter than the comps by several magnitudes. That is a hard one. How to insure you have a 12th magnitude stars with SNR > 100 while not saturating a target star at 6.9? That sounds more like a problem to pose to Arne Henden or Ken Menzies directly. I have never attempted this kind of imaging, so I am no help except to suggest that one uses the absolutely largest dynamic range possible for the camera and use 1x1 binning.

 

How would I check for saturation in general? I would examine the seeing disks of all stars I wished to measure at the pixel level and see if there were any saturated pixels.  But you already know this.

 

I think all I was trying to say was that when you work near the limits of your linearity, be careful.

 

Ed

Edited by Ed Wiley, 19 March 2024 - 10:00 AM.

[GaryShaw]

Hi Ed et al:

 

I've been doing fine with the approach I outlined - I think. No obvious saturation has shown up when I check and I'm getting the check star at SNR's in the 80-120 range over 58 observations so far. The variations come from the number of images I have to stack (10-20 usually) and probably also on transient local conditions at my home Obsy and the itelescope UT site.  Guess the Target's Airmass makes a difference as well.

 

My post was just hoping that I could learn more about how saturation works in a 'Binned' situation. I take a test image using TSX and it opens up on my screen, I zoom in and 'mouse-over' the bright central pixels and they appear to be around or below 60k. The FLI camera full well is 100k but itelescope doesn't say outright if that's at Bin 1 - I assume so. Anyway, I was hoping for some general clarity on how FW works in a binned situation for CCD (itelescope T21 and T18) versus my humble ASI294MM CMOS camera. I've been surprised at the lack of clarity on this question around the internet so I decided to bare my ignorance for all to see and ask the question here. So far my homework from your collective responses mainly centers around digesting what has been posted above, looking into the links and numbers from pvdv and running linearity tests at Bin 2 and Bin 3 for the ASI294MM and for the ASI2600MM when it arrives.

 

Thanks to you, PVDV and Robin  for pitching in with your thoughts and advice. I need to pursue this issue until I can wrestle it into some degree of clarity going forward for the benefit of future observations. Maybe you can wrap this topic in the next  AAVSO 'CCD3 Course'...!

 

Regards to all, wishing you better skies than we're having here..

Gary

Edited by GaryShaw, 19 March 2024 - 05:22 PM.

[GaryShaw]

Hi All

 

I've attached an image from The Sky's FITS viewer. I moused over the pixel thats circled and the ADU displays on the lower left to the right of the numbers within the parentheses.

 

This ADU value of 59,823 adu is measured for the the circled pixel. This is a bin 2x2 image so, do we believe the pixels shown are the sub-pixels or the primary pixels? I've been reading them as the 2.3u sub-pixels of the ASI 294 MM. If that's correct, none of these pixels appears to be saturated... Per the attached linearity test, at 59,823 adu, we seem still within the linearity of this sensor.

 

Thoughts or comments?

 

Gary

 

ps: This is an image of AB Aur taken at 75 sec exposure, Bin 2 with a PW CDK 17" at itelescope UT

 

 

 

Edited by GaryShaw, 20 March 2024 - 02:20 PM.

[GaryShaw]

Ed

 

In case it's of interest, here's the seeing profile of AB Aur taken from the image displayed above.  You mentioned checking the 'Seeing Disks' for saturation. Perhaps the image below taken together with the pixel zoom image shown above, allows confirmation of the saturation state. Perhaps just the image above suffices?

 

Gary

 

[pvdv]

Not 100% sure of what you are asking but, if you bin 2x2 in camera, load the resulting file in a visualization software and zoom to the pixel level, each pixel is now either the sum (which may require saving to 32-bit FITS files at imagining time in order to avoid truncating/clipping data) or the average (which will fit in a standard 16-bit FITS file) of a 2x2 square of pixels - 4.6µm in that case. Plate solving and getting the pixel scale should confirm this.

There is no way for the software to walk back, so to speak, to the 4 native pixels values from either a sum or an average. 

The "start your transit at 2/3 rd of the maximum ADU count of your linear range" ( say 40000 for 60000 max is this case ) approximation is a good starting point. I would even start lower than that if the elevation of the target is expected to increase dramatically during the transit. Starting closer to your linearity limit may make sense if the transit starts at maximum elevation. While on paper it may seem to make sense to maximize the exploitation of your linear dynamic range, a potential minimal SNR gain can very quickly turn into unusable data if the target rises or the transparency improves. I suspect those considerations might be a bit less important when one has consistently good transparency using remote scopes, but I have been burned quite a few times with improving transparency during the transit.


 

[GaryShaw]

Thank you. That all makes good sense. For recent captures, AB Aur has been on its way down so I went with the 60k adu and got by without saturation. When binning with cmos, is there a way to  know whether it’s averaged or summed?

Gary

[pvdv]

Not reliably I am afraid. Apparently, in the case of the 6200mm at least, there is another "surprise" per the manual

The ASI6200 camera has a built-in 16-bit ADC. 12bit ADC mode for output will be used when we

do hardware Bin, This camera also supports a custom ROI partial readout mode, with a faster frame
rate at small ROI resolutions.

 

That shoots down the hope of exploiting any fancy advertised "FWC" for those cameras when binned.

That's not as bad as it seems, even the sciency expensive FLI cameras with their huge pixels/FWC struggle to go above 14 bits of dynamic range. The most recent Sony sensors are just above 14 bits thanks to their very low read noise.