Jump to content

  •  

CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.

Photo

Photometry - CCD vs CMOS

  • Please log in to reply
14 replies to this topic

#1 jlandy

jlandy

    Apollo

  • *****
  • topic starter
  • Posts: 1464
  • Joined: 21 May 2014
  • Loc: Groton, CT

Posted 22 December 2016 - 08:44 AM

I recognize this will likely turn into a hot debate when looking at just the title, but I didn't know where else to turn.

 

The professor in my astronomy club is looking to replace their old CCD cameras for her classes where students are doing experiments with photometry. They have a fairly limited budget for both the camera and narrowband filters/filter wheel. My only experience in monochrome imaging is with the ASI1600mm-cool, which for astrophotography has been great. The new monochrome CMOS cameras on the market are certainly within their price point.

 

My (granted uneducated) thought is as long as the quantum efficiency is comparable, or at least a known value that could be accounted for, there shouldn't be much of a difference in the data acquisition. The low read noise of the CMOS might be an added benefit.

 

So the question is - does anyone have any experience with photometry, in particular comparing the results of a CCD vs a CMOS? If not, are there any scholarly articles comparing the two types for this application? My google searches have led just to broad comparisons of the sensors... haven't had much luck finding anything specific to photometry.

 

thanks in advance! hope this thread doesn't turn into a boxing match...



#2 jgraham

jgraham

    ISS

  • *****
  • Posts: 20380
  • Joined: 02 Dec 2004
  • Loc: Miami Valley Astronomical Society

Posted 22 December 2016 - 08:59 AM

First, you might want to check the AAVSO web site. Photometry with CCDs and DSLRs (with I believe use CMOS sensors) is not quite common.

 

Personally, I use both a monochrome CCD and a DSLR for photometry. I use a Johnson V filter on the CCD and the Tri-G (green) channel from the DSLR. I get essentially the same results from both. Regardless of the camera, it is a good idea to check the linearity and to stay within the camera's linear range (usually 80-90% of the camera's dynamic range). You can even use a camera with non-linear response if you calibrate to so that you can linearize the data.

 

Soooo, it should work, but I would check the linearity.


  • Lightning and jlandy like this

#3 jlandy

jlandy

    Apollo

  • *****
  • topic starter
  • Posts: 1464
  • Joined: 21 May 2014
  • Loc: Groton, CT

Posted 22 December 2016 - 10:34 AM

First, you might want to check the AAVSO web site. Photometry with CCDs and DSLRs (with I believe use CMOS sensors) is not quite common.

 

Personally, I use both a monochrome CCD and a DSLR for photometry. I use a Johnson V filter on the CCD and the Tri-G (green) channel from the DSLR. I get essentially the same results from both. Regardless of the camera, it is a good idea to check the linearity and to stay within the camera's linear range (usually 80-90% of the camera's dynamic range). You can even use a camera with non-linear response if you calibrate to so that you can linearize the data.

 

Soooo, it should work, but I would check the linearity.

Thanks for the quick reply - that site looks like a great place to start!

 

After reading their forums for a bit, I'm proposing the college do an experiment with their CCD, my 1600, and also my asi120mm guidecam - I've read a bunch of accounts of people using the 120 successfully for photometry... which would certainly make the budget workable if the results are good



#4 lambermo

lambermo

    Viking 1

  • -----
  • Posts: 905
  • Joined: 16 Jul 2007
  • Loc: .nl

Posted 22 December 2016 - 02:25 PM

It's quite linear. See post-205864-0-57176500-1477635295_thumb.

from this thread http://www.cloudynig...e-zwo-settings/



#5 yawg

yawg

    Ranger 4

  • *****
  • Posts: 354
  • Joined: 18 Jun 2013
  • Loc: Phoenix, Az

Posted 22 December 2016 - 03:09 PM

 it is a good idea to check the linearity and to stay within the camera's linear range (usually 80-90% of the camera's dynamic range). You can even use a camera with non-linear response if you calibrate to so that you can linearize the data.

 

Soooo, it should work, but I would check the linearity.

This is key, knowing your instruments linear response range.  As lambermo has pointed out, it appears linear.  although, a linear regression of a data set is NOT how linearity is determined.  To do it properly, a residual plot of a linear regression needs to be used.  Simply getting R= 1.0 means nothing.



#6 groz

groz

    Vanguard

  • *****
  • Posts: 2148
  • Joined: 14 Mar 2007
  • Loc: Campbell River, BC

Posted 22 December 2016 - 04:48 PM

Linear is part of the battle for photometry, but, adc resolution is also a big part.  With 12 bit cmos resolution, that means you get 4096 points between 'empty' and 'full well' as the best resolution it can do.  With a 16 bit adc, you get 65536 points over the same range, from empty to full well.

 

For photometry, that means using 12 bit will only give very coarse measurements, and will be unsuitable for doing any kind of precision photometry.  Just as an example, if you are trying to measure a 0.02 mag dip for an exoplanet transit, something easily done with a modest ccd, the 12 bit resolution of the cmos will mean  that amount of dip doesn't even register in the data you produce.


  • jlandy likes this

#7 jgraham

jgraham

    ISS

  • *****
  • Posts: 20380
  • Joined: 02 Dec 2004
  • Loc: Miami Valley Astronomical Society

Posted 22 December 2016 - 11:27 PM

Eh, 12 bits isn't too bad for photometry depending on how much dynsmic range you neec to bridge the gap between your variable and reference stars. I don't think that I have ever used a 12 bit camera for photometry, though I have used a 12 bit linear diode array for spectroscopy. My DSLR is 14 bits, which works fine, and my CCD is 16 bits, which works great!



#8 freestar8n

freestar8n

    Vendor - MetaGuide

  • *****
  • Vendors
  • Posts: 8959
  • Joined: 12 Oct 2007

Posted 23 December 2016 - 06:08 AM

Let's say you want to detect 0.02 mag. That is a fractional change of about 10^(0.4*0.02)-1 = 0.0186.

The shot noise fraction goes as 1/sqrt(N) - so for the shot noise alone to be less than 0.0186, we would need (1/0.0186)^2 = 2900 electrons.

If you had all the star photons in a single pixel at that intensity, the electron count would be 2900 +/- 54 due to shot (Poisson) noise alone - and it would still have 0.02 mag standard deviation. If you have gain of 1 e/adu - you would only need adu count of 2900 - and even then the shot noise would be huge (54 adu) compared to the digitization noise. 2900 is much less than 4096 of a 12-bit sensor.

Even if you had 10 e/adu, the adu count would be 290 and the noise in adu would be 5.4 (in adu, the Poisson noise is not the sqrt of the adu count). So even then, the digitization noise of ~1 adu would be much less than the inherent shot noise of 5.4 adu. In fact, the digitization noise isn't even 1 adu but 1/sqrt(12) adu - or 0.3 adu, which is even more negligible. So it would be ok even with 8 or 9 bits.

Finally - an actual star spot would be spread over many pixels - so the total intensity could be much more than the 12-bit limit of one pixel. 4 pixels would be 14-bit and 16 pixels would be 16-bit.

There may be other subtle reasons why cmos isn't ideal for photometry, such as reproducible behavior and good response to applying flats and bias - but I don't think 12-bit is big problem either for stacked images or photometry.

Unless I made a mistake in the math.

Frank
  • Shiraz, Jon Rista and TFT Ben like this

#9 555aaa

555aaa

    Vendor (Xerxes Scientific)

  • *****
  • Vendors
  • Posts: 1484
  • Joined: 09 Aug 2016
  • Loc: Lynnwood, WA, USA

Posted 24 December 2016 - 08:53 PM

Interesting question. If the use is light curves in only one band, then noise and linearity are important but if accurate color indexes are important then spectral response is important.

#10 groz

groz

    Vanguard

  • *****
  • Posts: 2148
  • Joined: 14 Mar 2007
  • Loc: Campbell River, BC

Posted 27 December 2016 - 01:29 PM

Unless I made a mistake in the math.
 

On the surface the math looks ok, for simple absolutes, and would work wondefully doing all-sky photometry under photometric conditions.  But that's not what amateurs typically do.  What we end up doing is differential photometry under less than perfect skies.  So expect more noise in the calcs overall, some from the cameras themselves, some from the sky conditions (seeing, transparency, etc), and the measurement is one noisy source against another (target vs reference).

 

I have not tried things with current generation cmos sensors, but I do remember a few years ago after I did an exoplanet transit measurement for the first time using an SXV-H9 camera in a C8.  At that point, my wife wanted to try the same kind of stuff, she was using a dslr, and had just read an article about folks doing a transit measurement with a dslr.  That particular one was a very bright candidate, ie one of the easier measurements to do, so we decided to set up a test run on a fast variable with a significant change over a few hours, just to see the difference.

 

The end result was as shown in the graphs below.  Top one, the ccd in the C8 with reducer, bottom one is a dslr in a Williams 110.  They were set up on our back deck about 10 feet apart pointed at the same variable and both running exposure lengths to bring the target star up to about 50% saturation, typical of how we do differential photometry.  Suffice it to say, after doing this side by side experiment, the next day I went shopping for another ccd.  the plots below use the same reference stars against the same target star, just different camera setup, admittedly a poor camera for this kind of work.  Camera in question is 12 bits with a bayer color matrix measured against a low end monochrome ccd with photometric filters, in this case if memory serves correctly, we had the V filter in for this run.

 

PhotometryTwo.jpg

 

 

If one now assumes that the modern cmos sensors do give good linear results, then look more detailed at the differential photometry problem.  For this exercise, there is no reason to set the gain at anything other than unity.  Exposure time would be defined by whatever time it takes to get your desired count on the target star, so using the above numbers, expose till the target is at 2900 or so to achieve a 0.02 differential measurement.  Any bright stars in the field will be saturated, and you need suitable reference stars in field that come up to roughtly 2500 as a minimum, and 3500 as a maximum.  If there is any vignetting in the field, you need to choose reference stars that are not in the vignetted portion of the field.  Choices of reference stars will be limited on a small sensor with any amount of vignetting in the field.

 

But in general, it really depends on your goals and objectives.  If the goal is to measure variables with significant changes over time, the 12 bits wont be a huge hinderance.  If your goal is to get serious on exoplanet transit measurements, many of them are somewhat less than 0.02 magnitude dips.  I remember when the first keppler data set was released, a bunch of us jumped on what looked like the easiest of them to snag, and we had a bit of a competition going here on CN to see who would get one first.  After the first night with runs on one of them (dont remember which) the conclusion was, gosh these are hard.  We had to go deep enough that binning was required to get measurable numbers on 10 minute exposures, and none of us were able to produce a light curve that showed any recognizeable dip.

 

If one already has the camera setup, it's easy to see what it's capable of.  The AAVSO variable star database has lots of suitable candidates for generating short term light curves, just search the database for something with a period of less than 0.1 days (2.4 hours) and variability of half a mag or so, that's a great way to wet your feet and see what kind of light curve you can produce, the change is large enough to be clearly visible thru any system noise, and fast enough you can get a couple cycles in one session.  This is how I found the star used for the curves above.

 

On the flip side, if you are thinking of buying a camera specifically for the purpose, you have to start by defining the purpose, there are many different applications of photometry, and they have different requirements.  In our case, I want to be able to take measurements of virtually all the stars of a magnitude range in a given field, so we set up a system specifically targetting that application.  It's by no means a starter system, 16803 sensor in a 12 inch RC capable of illuminating the entire field with no vignetting.  What got me hooked, was the first time I did a run on a transitting exoplanet, this is the curve we got from WASP-10b when set up at a star party many years ago.  When we saw this result, it completely changed our approach to the astronomy hobby, and when we started choosing equipment for our 'retirement project' observatory, being able to do this kind of stuff at higher precision was the primary consideration.  This curve was generated from data taken with an SXV-H9 in a C8 with 0.63 reducer, ie a relatively modest system.  We bought all this equipment from the used marketplace, so it was not a huge outlay to set it up.

 

Wasp-10b-full.jpg

 

It's now the better part of a decade later, we have moved, now live under a dark sky and the observatory build is on hold due to snow, but that should clear up soon.  I have a significant list of projects for this setup over time, it'll measure exoplanet transits, do some asteroid light curves, and likely start some longer term projects studying specific objects which have not yet been defined.

 

The point of this long ramble, photometry means different things to different people, and choosing the camera to use really depends on what it is you are trying to measure.  There will be projects well within the reach of the cmos cameras, but, there will be many projects that are outside the reach of those sensors.  It really depends on what you are trying to measure.  The analogy is, a measuring tape marked out in 1/16" increments will do a fine job measuring a lot of things, but, it'll fall flat trying to measure something where you are trying to get accurate measurements of a half millimeter, for that you probably want a micrometer.

 

So to go back to a proper answer for the question in the original post, is a cmos camera suitable for doing photomtery for the professor in the astronomy club.  To answer that question, one has to refine the question.  What is it that you want to measure ?


Edited by groz, 27 December 2016 - 01:31 PM.


#11 freestar8n

freestar8n

    Vendor - MetaGuide

  • *****
  • Vendors
  • Posts: 8959
  • Joined: 12 Oct 2007

Posted 27 December 2016 - 05:30 PM

I was mainly responding to your claim that 0.02 mag would not register if you had only 12-bits.  0.02 mag is 1.9%, which is actually quite large and would register on a scale with a max count of about 50, rather than 4096.  Exoplanet stuff often refers to "millimag" work - and 0.02 is 20 millimag.  If instead you wanted 1 millimag accuracy, that would be a change of about 1 in 1000.  And for shot noise to be smaller than that - you would need a total electron count of about 1 million.  And that is just so that you aren't limited by shot noise itself - ignoring all other noise sources.  If the well depth is about 10,000 and the stars are spread over 4 main pixels - you would need about 25 exposures in order to beat shot noise without saturating the pixels.  If instead you had an Andor camera with 500k well depth - you could easily do that with a single exposure.  So that does show the benefit of a very deep well depth when doing photometry - but 0.02 mag is not particularly small - so both well depth and bit depth aren't a problem and you should be able to do much better than that before those aspects of the camera become a problem.

 

I don't know the OP's specific interest for photometry - but I think that well depth would be more of a factor for general work than resolution - simply because at that well depth the shot noise is huge compared to the bit resolution.  For that reason I would use a high value of gain in e/adu - so the well depth is near the max of 20k.  If the stars are spread over about 4 pixels that should allow a wide range of magnitudes to be captured.  Only at the low end would the stars be faint enough that the digital resolution would be greater than the shot noise.  For 20k well depth in 4096 steps that is about 5 e/adu and the digitization noise is 5/sqrt(12) = 1.4e - which is tiny.

 

It's unfortunate that comparisons of ccd with cmos are often comparisons of astro ccd with dslr - because a dslr has many other problems that have nothing to do with it being cmos.  The ASI-1600mm is monochrome and cooled - and you have direct access to the raw data - as opposed to a dslr with its bayer matrix, no cooling, and semi-raw format.

 

Here is some work I did a few years ago regarding absolute magnitudes with sloan filters - using APASS stars as references.  That's very different from exoplanet differential work - but I showed a fundamental error of about 0.05 mag just due to inaccuracies in the reference star mags.  That was with ccd but I think the same would apply to cmos.

 

http://www.cloudynig...iz-differences/

 

Here is a link from AAVSO regarding the ASI-120m for photometry - and they don't have a problem with it.  The ASI-1600 is much larger and would allow more reference stars.

 

https://www.aavso.or...ry-zwo-asi120mm

 

Frank



#12 gregj888

gregj888

    Vanguard

  • -----
  • Posts: 2332
  • Joined: 26 Mar 2006
  • Loc: Oregon

Posted 27 December 2016 - 07:53 PM

After testing cameras for speckle, I'll add another fly to the ointment.  CMOS sensors have an amp and ADC per row (or column).  These are calibrated "on the fly" with when depending on the camera.  In some cases, this cal appears to be done per frame, in others at startup.  Since there's no way to freeze the calibration, there's no way to do precise measurements, even differential on the same frame as the stars are being read by different analog channels with unrelated calibrations.

 

So the question is how far off are the measurements and how much variation can you take?  No question CMOS will wor, the question is how precise they are?

 

IMHO, a camera that re-cals each frame is prefered with an exposure spread over a number of frames (100s or thousands).  If you can find software that stacks, that is aligns and adds the pixels you would average out the calibration artifacts and build up 16 or more bits of dynamic range. 5lii)

 

Easy to test this BTW.

1) capture 2 sequences of 500 frames

2) stop and start the camera, change the exposure and change it back (resets my QHY5Lii)  or make other changes...

3) with the setting set back to those used in 1, take another 500 frame set

4) Average each of the sets, and subtract one from another looking at residuals/patterns.  My CCD frames are very flat, the QHY5Lii is flat 1-2, but shows a distinct pattern 1-3...

 

J, If you do test the cameras, please post the results.

 

Greg


Edited by gregj888, 27 December 2016 - 07:54 PM.


#13 WadeH237

WadeH237

    Aurora

  • *****
  • Posts: 4932
  • Joined: 24 Feb 2007
  • Loc: Snohomish, WA

Posted 27 December 2016 - 08:52 PM

I realize that this doesn't answer the (very interesting) question about the difference between CCD and CMOS, but if you want to answer the question of what camera to get on a budget:

 

Why not find a used ST-10?  You should be able to get one at a pretty good price.  They are CCD with pretty high quantum efficiency, large well depth, reasonably sized sensor, and NABG with a nicely linear response.  The drivers are rock solid stable, and it should work easily with just about any software that you choose.

 

I believe that they were considered one of the best cameras for amateur photometry back in their day.  Even today, I've used mine for over 10 years, and have held off on buying anything newer because even with the newer sensors, it's hard to find something that is a clear improvement in all areas that is worth the cost to me (ie. I can get a much larger sensor, but with less QE; I can get much lower read noise, but with a smaller sensor and much lower well depth; etc.)

 

-Wade



#14 Astrojedi

Astrojedi

    Skylab

  • *****
  • Posts: 4120
  • Joined: 27 May 2015
  • Loc: SoCal

Posted 28 December 2016 - 12:59 PM

Also one other wrinkle with CCDs which folks ignore is read noise. Comparing output bits is absolutely useless. What you need to compare is the dynamic range of the sensor. There is no such thing as a 16 bit dynamic range.

 

For example the SVX-H9 has 7e-12e read noise as per the camera's manual. Which means anywhere from bottom 3-4 bits (2^3 = 8) in the 16-bits is just noise assuming unity gain. Only gets worse if you increase the gain. So the actual dynamic range for a single exposure is only 12-13 bits.

 

If we take a 14bit ADC CMOS like the ASI178 you are comparing a 14 bit camera with <2e read noise to a camera with 12-13 bit real dynamic range.

 

All the CMOS sensors have the same dynamic range as a CCD (typically in the 70-72 db range). Also you can very effectively use stacking for photometry which I do quite regularly with excellent results. This whole debate is very misleading.


  • gregj888, Miguelo and Jon Rista like this

#15 gregj888

gregj888

    Vanguard

  • -----
  • Posts: 2332
  • Joined: 26 Mar 2006
  • Loc: Oregon

Posted 30 December 2016 - 12:52 PM

To add to Jedi's statement...

 

The new CMOS sensors are 80%+ QE and <1e TDN (read noise), so you are effectively photon counting...  Stacking and the use of short exposures is free...

 

jlandy, if your Prof hasn't looked at doing double stars (w/ autocorelation) as an option to photometry might bring that up.  The 290 based cameras are a very good option here and CMOS works very well. 

 

BTW, when testing the CMOS cameras for photometry, take 500 or so short exposures instead on one long one and dither.  If the camera does "recal" it should average out some of the effects.  Try AstroLive USB if you have an ASI camera (free).  Might just do the trick and adds to the data reduction sampling discussion.  Should also improve FWHM and help improve measurements.  Properly done, CMOS may do better than CCDs, but it is still unclear.


  • Miguelo, Jon Rista, jlandy and 1 other like this


CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.


Recent Topics






Cloudy Nights LLC
Cloudy Nights Sponsor: Astronomics