CMOS dark frames

Articles in this series:

Characterising RTS

RTS Pixel Behaviour

Dark Frames (this page)



On-sky Photometry


In this post I will look at the dark frames from the QHY183M.

I’ll look at amp-glow, hot pixels, and how well they subtract out.  I had heard that CMOS darks are not the same as CCD darks, in that they have to be taken as close as possible to the time of the science frames, so I will check that.



This is 40 dark frames of 10s each combined with C-Munipack to make a master 10s dark frame.

10s master dark

Even with just 10s exposure the amp-glow on the right of the image is already visible.  Other than that the dark frame is actually very clean.

Amp-glow becomes much more pronounced at 120s.

Single 120s dark

Aside from the amp-glow there is some horizontal banding, and fainter vertical banding.  The amp-glow looks horrendous, but taking a cross section through a master made from 40 frames shows that it is ~40 16bit-ADU high, or only ~3 actual (12-bit) ADUs.

Master 120s dark, with x-section


Dark subtraction

This is a single light frame of 120s exposure, of about 190 ADU, taken several days after the master dark.

Single 120s light frame

The 3 ADU amp-glow is not so obvious, but it’s there.  Then the master dark is subtracted from this frame:

Single 120s frame dark subtracted

It’s had to tell from this how effective the dark subtraction has been, so I also plot the histograms.

Histogram of single 120s light frame

There is a long tail of hot pixels stretching to the right.  Not many – above 600 ADU there are only 1 or 2 at each level.  Now the histogram post dark subtraction:

Histogram of dark subtracted single 120s light frame

Most of the hot pixels have been subtracted out.  There is one solitary uncorrected pixel at >700 ADU, and a few others that are lower than you would expect.  So despite the master dark being a few days old, dark subtraction has worked well.


Some other quirks

Initially I got weird results when I plotted the histograms.  A population of pixels clustered on the left of the chart which I’ve now tracked down.  Turns out that the first 4 columns on the left of the sensor are unresponsive – they look normal on dark frames, but their ADUs don’t increase with light as much as the rest of the sensor.  I looked on the QHY forum and there was a question on this.  And the response was ‘This is normal’!  Incidentally, the histograms above omit the first 4 columns.

There are also a few (5 to be exact) pixels which, although they do respond to changing light levels and have normal scatter, show lower ADUs than expected.  I am not concerned about these.  As long as they respond linearly, flat-fielding should correct this.  Also they are very few.



Looks like dark subtraction works as usual on this sensor, as long as you use the same settings and exposure time.  Amp glow does not seem to be an issue.


Next, I’ll look at linearity.