Affiliation
American Association of Variable Star Observers (AAVSO)
Wed, 04/10/2019 - 00:14
I have been using 2x2 binning for all of my photometry images, mainly because of reduced file sizes, and the speed with which I can work with them. I was under the impression that binning was improving my S/N as well. This doesn't appear to be the case.
The other day, I took sets of 3 images a star field near the zenith using 2x2 and 1x1 binning, and did photometry for a roughly mag. 14 star. The measuring aperture was doubled for the 1x1 binned images. For the three 2x2 images, the error was 0.025 mag., while the 1x1 images produced a 0.017 mag error.
I am curious if others have had this same experience. Is there a simple explanation for the improvement that goes along with not binning? Thanks.
Tom
I think you need to provide additional information if you'd like us to help you to understand the issue. What is the FWHM of these images (both binned and un-binned)? Are the stars far above the background? What is the readout noise of the chip?
Sorry for the slow reply. Here's some specific information for my setup. As you'd expect, 2x2 binning doubles the Max. Pixel and Sky counts. The unexpected result is that 2x2 S/N is significantly worse.
Tom
Hi Tom,
I think you should reduce your aperture in case of 2x2 binning as your star will not cover the same pixels as in 1x1 bin.
I routinely run my setup in 3x3 binning (due to file size) . I actually have not checked your findngs in my case.
Josch
polakis: Actually, wouldn't you expect ~ 4 X (not 2 X) the sky and max counts for 2x2 (4 X the pixels) vs no binning, all else being equal?
Is 16000 ~ your saturation ADU? If your on-camera binning is additive rather than averaging, that can cause earlier saturation than does no binning. And if the binned image is saturated at star center, we can't trust the statistics.
I see that the aperture is in arcseconds, not in pixels.
Thanks for the replies. Josch, as Eric noticed, aperture is in arcseconds. In other words, 9 pixel measuring aperture for 2x2 and 17 pixels (Canopus software needs an odd number here) for 1x1.
Eric, yes, I'd expect 4x, but in several tests now, the result has been ~2x. In my linearity tests, the non-linear count for the ST-6303 at 2x2 begins around 45,000 ADU. Perhaps I should see if I get a similar result using a slightly brighter star in that frame.
For reasons other than photometry, I am beginning to look into a new computer. If I can be confident that 1x1 is really reducing the errors, I'll go all in for a much faster CPU and greatly increased RAM, and image unbinned. As it stands now, processing and measuring unbinned images really brings my computer to its knees.
Tom
Tom,
How do you decide on the size of your measuring aperture?
Does Canopus have a SNR curve of growth tool, i.e. a plot showing SNR on the Y axis and measureing aperture pixels on the X axis?
Phil
Phil,
I'm using the general guideline of 3 to 4 times the FWHM diameter. My seeing is pretty consistently 3.5 to 4.0 arcseconds, so I use a measuring aperture of 16 arcsec.
Oh. *that* Phil! I hope our camera exchange at last year's Symposium worked out for you as well as it did for me!
Tom
Yes Tom, I'm very happy with our camera trade. If you are going to the SAS meeting next month I'll bring you a picture of what your ST402 is doing now.
Back to your binning question.
I would be interested to see how much the choice of measuring aperture affects the results you got with binning. If you could adjust the measuring aperture to give the highest SNR for the same star in the binned and unbinned images, that would clear up that part of the question for me.
Also, how does Canopus determine the uncertainty? I think that Canopus can do ensemble photometry. Am I right about this? Is that how the uncertainties in your experiment were calculated?
Phil
Phil's suggestion of varying the measuring aperture prompted me to make a couple plots. They are families of curves of S/N vs. magnitude for three measuring apertures, 2x2 binned and non-binned. Smaller aperture is better until you start cutting off the tails of the PSF, so I'm going with 7 pixels (12.3 arcsec) for 2x2 and 13 pixels (11.4 arcsec) for 1x1. The two plots have the same vertical scale, and again show the large improvement for my system with imaging non-binned.
The last chart uses the simplistic assumption that error for each star is simply the reciprocal of S/N, not considering comp star errors in the ensemble photometry that I use for asteroids. This plot better illustrates the improvement with non-binned images. For a magnitude 14.5 object, error is reduced from 0.025 mag to 0.015 mag. I do asteroid photometry out to about 15.5, so this would doubtless improve as these curves diverge further.
Tom
Some interesting plots, Tom - thanks! However, I'm concerned that you may be comparing apples and oranges. How is the signal/noise ratio being calculated? Remember that most camera vendors change their gain when binning, and so the electrons per ADU value will differ. Are you taking new darks and new flats for the binned images? What software are you using?
In general, there should be negligible difference in binning for bright stars, as you should be collecting the same number of electrons. 1x1 binning is usually slightly better because, with more pixels within the measuring aperture, it can be "circularized" better, and partial pixels on the aperture boundary have smaller effect. For faint stars, 2x2 binning usually wins because readnoise may be a bigger noise contributor.
Arne
Arne,
Signal to noise ratio is calculated by the sotware. Since my main area of interest is asteroids, I use MPO Canopus. The binned images were calibrated using a unique set of flats, darks, and bias frames. Yes, I would have expected the 2x2 binned images to have lower errors, not higher, for the same reason of read noise that you mentioned.
Many asteroids of interest are magnitude 15 and fainter. From my urban observing site, the errors at this level are large, and when the amplitude s small, the definition of my lightcurves get swallowed up by errors.
Tom
Thanks, Tom. My comments from before still stand: SNR needs to know the system gain in order to be correct. I don't know where Brian is getting that information, and so can't tell whether it is being done correctly for the binned mode. You should calculate it by hand, using the total counts within the aperture, properly sky subtracted, along with the system gain.
Whether 15th magnitude is faint, and whether readnoise is a major contributor to the noise, depends on many factors. I'm betting that you are sky noise dominated, in which case readnoise is not going to be a major player.
The best way to tell the true SNR is to take 10-20 exposures of a field in which you have a comparison star and a constant star. Do that for both the 1x1 and 2x2 modes on a clear night near the meridian. Compare the standard deviation of the (constant-comparison) differential magnitudes for the two modes. Much better to do this experimentally than to trust software.
Arne