May 29, 2022: Tests of gain for ATIK and ASI cameras

Michael Richmond
June 1, 2022

Executive summary:

• ATIK 11000 camera: gain = 0.66 e-/ADU
• ZWO ASI6200MM camera: gain = 0.264 e-/ADU

Inspired by our acquisition of our new ZWO ASI6200MM CMOS camera, I decided to measure the gain of both the new camera, and our ATIK 11000 CCD camera. The method involves taking a number of flatfield images with different average light levels, and performing statistical tests on pairs of images with the same average values. Various descriptions of this procedure can be found in a number of places:

• Tech Note: Pixel Response Effects on CCD Camera Gain Calibration by Michael Newberry
• Canon EOS 20D vs Canon EOS 10D by Christian Buil (skip down to the section "Complement : How to compute the photon transfer curve and the electronic gain ?")
• CCD Gain by Michael Richmond (skip to the final section)

The basic technique involves a number of steps, ending with a graph of variance in signal vs. mean value of signal.

1. acquire a set of flatfield images covering a wide range of average light levels. At each level, take several images.
2. subtract a bias or dark frame from each image, as appropriate
3. choose pairs of images (A and B) with similar light levels. For each pair,
   1. compute mean value in each image
   2. multiply B by a constant to match the mean level of A
   3. create a difference image D = A - B
   4. create a summed image S = A + B
   5. in a number of small boxes, each around 25-by-25 pixels in size
      • compute mean intensity of S
      • compute variance in D, via variance = (stdev*stdev)/2
      • print out mean and variance

One can then make a graph showing the variance as a function of the mean values. On this graph, the slope is related to the gain of the camera.

no binning, or binning via sum

In the simplest case, if there is no binning involved, or if the binning involves replacing a block of pixels by a single super-pixel with the SUM of the individual pixels, then

                             1
            slope   =     -------
                           gain
     
     

Here, the "gain" is defined as the ratio of electrons to ADU in the images produced by the camera. Thus,

     
                   gain  =    electrons / ADU
     
     

binning via average

If the data have been binned with an algorithm which replaces a block of pixels by a single super-pixel with the AVERAGE of the individual pixels (as done by MaximDL, for example), then

                                                2
                             1        (   1    )
            slope   =     -------  *  ( ------ )
                           gain       (  bin   )
     
     

where the bin value in the equation above is the factor by which the image has been binned. For example, if the images are binned 2x2 using an average, then the slope will be equal to (1/4) * (1/gain).

after scaling the pixel values

Because the software I use ( XVista ) can't handle pixel values greater than 32767, I start all my data analysis by dividing images by 2. This scaling modifies the slope of the variance-vs-mean graph in the following manner:

                                      
                             1           
            slope   =     -------  *    scale_factor
                           gain         
     
     

where the scale_factor value in the equation above is the factor by which the image has been scaled. In my case, scale_factor = 0.5, and so the slope in my diagrams is always half of the expected value.

This adjustment stacks multiplicatively with any binning factor. Thus, if the data has been binned 4x4 using an average, and has also been divided by 2, then the slope will be (1/32) * (1/gain).

Measurements of ZWO ASI6200 camera

These tests were run during the day, with the temperature of the sensor set to 0 Celsius. I closed the door to the dome to prevent outside light from coming in, and turned on the dome lights. The telescope pointed straight up at the top of the dome.

Exposure times ranged from 0.1 seconds to 30 seconds. I used a single master dark frame based on 1-second dark images, since I've verified that the dark current is negligible over time periods shorter than one minute.

The gain measured here -- 0.264 e-/ADU -- agrees with the values shown in the graph below, reproduced from ZWO's description of the ASI6200MM. Our camera is set at the factory default "gain" value of 100 units.

I then set the camera binning to 2x2 and repeated the darks and series of flatfield images. Note how the variance is much smaller for a given mean value in these images, compared to those taken with 1x1 binning. This is a result of MaximDL binning data from this camera with an algorithm based on the average of values in each set of 2x2 pixels.

Measurements of ATIK 11000 camera

These tests were run during the day, with the temperature of the sensor set to 1 Celsius. I closed the door to the dome to prevent outside light from coming in, and turned on the dome lights. The telescope pointed straight up at the top of the dome.

Exposure times ranged from 0.1 seconds to 30 seconds. I acquired three dark frames at each exposure time in order to subtract the appropriate amount of dark current, as this camera has a dark value which changes rapidly with exposure time.

The gain for this camera is about 0.66 e-/ADU. That's quite different from the value of 0.92 e-/ADU provided at

• Atik Cameras Specification Tables

but agrees roughly with the measurement of 0.695 e-/ADU made by Christian Buil at

• Noise and electronic gain evaluation of a sample of astronomical CCD cameras

I repeated these tests with the ATIK camera set to 2x2 binning. The results, shown below, are close to the gain measured in unbinned (1x1) mode --- as long as one postulates that the ATIK camera was performing a sum-based binning. I believe that the ATIK camera is set up to perform the 2x2 binning on the chip itself, rather than allowing MaximDL to perform the calculations after the image has been transferred from the camera. I can't prove that at the moment, since it would require running MaximDL while attached to the camera, but it would explain the results.

Once again, I'll show the histogram of the dark frames first.

Last modified 6/1/2022 by MWR.