QHY600 tests

error

sorry i misspoke/typo - Mark is at gain 56 - where the step is in the charts ...

but that was effectively a typo that does not impact the discussion/questions..!

Yes, I'd like to see more discussion; I'm not convinced that I've thought this through properly.

I used a four-step process to come up with my photometry process for the QHY268M:

1. Characterize my equipment and skies
2. Clearly define what I'm trying to measure using scenarios (an idea borrowed from the world of software development)
3. Predict performance for various alternatives and choices
4. Select the alternatives that make sense given what I'm trying to measure

In particular, I identified 4 scenarios that describe the most challenging photometry that I care about:

1. Blue photometry of a faint, mag 17(B), LPV using a mag 12 comparison star. My sky is bright with light pollution (Bortle 4-5), and this challenge is to collect sufficient photons from the star to overcome high noise levels from skyglow and internal camera noise.
2. Ic photometry of a bright, mag 5(Ic), LPV using a mag 12 comparison star. This challenge is about avoiding nonlinearity on the bright variable. Skyglow is a non-issue.
3. Exoplanet time-series photometry of a mag 12 target (with a mag 12 comparison star). I arbitrarily selected an exposure time of 60 seconds (no stacking allowed).
4. Photometry of a Landolt field, a collection of approximately mag 12 stars. Nothing is unusually bright or unusually dim; the objective is to maximize SNR.

I am comfortable with single exposures of up to 2 minutes. I find that stacking multiple exposures generally gives me better photometry than averaging multiple measurements from each individual exposure, so with the exception of scenario (c), stacking is permitted in my scenarios.

I define photometry performance in terms of SNR. So, for example, in scenario (d) I know that I will collect a total of 686,676 electrons within my photometry aperture for a 2-minute exposure (signal). At a gain of 56 (readout mode 1), my noise budget looks like (each number is a total across the entire photometry aperture):

• Target shot noise: 829 e-
• Read noise: 78 e-
• Dark noise: 52 e-
• Skyglow noise: 258 e-
• Total noise: 873 e-

And that gives an SNR of 686,676 / 873 = 787. Further, selection of readout mode and camera gain has almost no affect on the SNR, which is dominated by target shot noise.

Scenario (c) gives almost the same result (SNR = 547), pretty much independent of readout mode and camera gain.

Scenario (b) becomes more interesting. If you stick with readout mode 1, gain 56, then you have to shorten the exposure time to 4 second to avoid non-linearity within the LPV's aperture. But when you do that, the comparison star SNR drops to about 53, driven by read noise. The noise budget (total within the photometry aperture) for the comparison star looks like:

• Target shot noise: 78 e-
• Read noise: 78 e-
• Dark noise: 9 e-
• Skyglow noise: 29 e-
• Total noise: 114 e-

If you choose a different mode/gain (for example, mode 1, gain 0), things change. You can now go to 8 second exposure time and stay linear. SNR rises to about 59 on the comparison star, even though read noise has doubled.

And, finally, scenario (a) requires 120 second exposure times with a readout mode of 1 and gain of 56. This will give an SNR of 25 on the variable and 700+ on the comparison star. The noise budget is dominated by skyglow. Because of that, the readout mode/gain really doesn't matter. For example, mode 1, gain 0 gives an SNR of 20 on the variable and 650 on the comparison star, not tremendously different.

My conclusion:

1. SNR is all about counting electrons, not ADU. For this reason, the effects of altering gain tend to be secondary, and relate to gain's influence on readnoise and where nonlinearity starts.
2. Photometric accuracy (SNR) is relatively insensitive to selection of readout mode and gain in environments that are dominated by skyglow, as long as you avoid the modes with relatively high readnoise (e.g., mode 0 gain 0 for this camera).
3. It is important to understand where nonlinearity sets in for whatever mode/gain combinations you use, and adjust exposure times to avoid nonlinearity.
4. Only one of my scenarios is particularly stressful to the camera's dynamic range (the LPV bright in Ic).

- Mark M

(And, by the way, choosing an offset value is an entirely different thought process that I'll save for another post.)
 

I've been using an offset of 5 with my QHY268M camera. Overall photometric performance does not seem particularly sensitive to the selected offset, as long as you stay away from extremes.

Understanding the purpose of the offset is best done by starting with a mental image of the histogram of pixel values in your images. Imagine a number line running from zero on the left to 65,535 on the right (the largest 16-bit value that the camera's A/D converter will generate). Photometry images are mostly background sky; the stars themselves cover a relatively small number of pixels. So, the histogram will have a distinct peak around the sky background's median value. Around that peak you'll see a gaussian drop-off on both sides. The camera will only generate non-negative values from the A/D converter, so some of the gaussian drop-off on the low side of the peak will be truncated (clipped) at a value of 0.

Of course, if you're making bias or dark images, the peak will be at a lower ADU value than the peak for sky background, because neither bias nor dark will include skyglow (including light pollution) counts. The histogram's peak will be lowest for bias exposures.

Changing the value of the camera's offset will shift the entire histogram left or right. Smaller values of offset will shift the histogram to the left; larger values will shift to the right.

Many image processing algorithms make an implicit assumption that the statistics of the image's background regions are statistically "well-behaved," with a roughly symmetrical histogram shape on either side of the peak. If the offset value is set too small, a significant part of the histogram's shape to the left of the peak can be shifted below the zero ADU level, chopping off some part of the histogram.

On the other hand, if the offset value is set unnecessarily high, the histogram will be nicely preserved, but some pixel values at the high end of the histogram (i.e., in the core of the brightest stars in the image) will be shifted toward the magic number of 65,535. If the shift is sufficient to push a valid star pixel value beyond 65,535, then that pixel has effectively saturated, and you've lost linearity on that bright star. For that reason, you don't want to crank up the offset too much. (Put another way, raising the offset will slightly reduce the dynamic range.)

The offset that I chose, 5, puts the peak of my bias exposure histogram into the range of about 80-90 ADU when I run at mode 1, gain 56. The standard deviation of the background distribution is around 6.0, giving at least 10 standard deviations on each side of the peak before zero-clipping happens. This is way more than enough to preserve clean statistical behavior, even though I often see something on the order of 17,000 pixels that "want to go negative" but have been clamped to zero.

To add another wrinkle, the offset is temperature-sensitive. The offset is actually determined by the value of the reference voltage fed to the A/D converter. The precision voltage source that is used for the A/D reference is located on the CMOS chip itself (this isn't true for CCD devices), so the temperature of the reference voltage circuit changes with the temperature of the sensor. I see a dependence of about 1 ADU per degree C variation in sensor temperature (at this gain setting). This temperature sensitivity is particularly relevant when making a long series of bias exposures. The circuits on the sensor with the highest thermal dissipation are the circuits used for the readout cycle; QHY keeps these circuits idle except when a readout is needed. When making a series of really short bias exposures, the circuits have a high ON time duty cycle, and I frequently see chip temperature fluctuations during bias sequences. Sadly, this means that the act of measuring the location of the zero-exposure histogram peak will itself shift the peak's location due to transient heating of the sensor.

In summary:

• The offset value of 5 preserves (to at least 10-sigma) the bias histogram's shape (at readout mode 1, gain 56) for my camera. (Your mileage may vary.)
• Choose a single offset value and use it for all of your exposures. In particular, bias, dark, flat, and science images should all use the same offset value.
• Poor temperature management can easily lead to bias or dark frames that are brighter than the background of your science images. Get to know your camera and the amount of reserve cooler power you need to handle a series of short-exposure images that raise the CMOS sensor's thermal load.
• Don't spend a lot of time agonizing over the precise offset setting. Its effect on dynamic range is quite small, and it's pretty easy to keep your pixels away from zero-clipping.

- Mark M