Darks: high amplitude resolution?

[andriy_melnykov]

Hi,
is there a way to use high amplitude resolution darks?
For example, 32-bit or float FIT files for darks with 16-bit subs?

Reason: particularly at low gains, the real value of dark pixel can lie between ADC units. If we use darks with same amplitude resolution, this value is rounded, and the difference can not be compensated. This can be seen as remainder fixed pattern noise. Can be compensated with dithering, but it is only a partly solution.

Thanks!

[admin]

Hi,

currently no. Dark and flat frames are created in 16 bit, so any detail that might be present at finer scales is not kept.

What camera are you using and what gain value/exposure? It might be interesting to work out how many dark frames you would need to bring the estimated noise level of the dark below 1 16-bit ADU.

cheers,

Robin

[andriy_melnykov]

I am using ASI385MC, uncooled but temperature stabilized. Gain from 135 (must be unit gain) to 300 (+16,5 dB to unit gain, if I understand it correctly).
Read noise is 0.7-1 e-, with dark current at typical 1-8-15 seconds - around 1-2 e-. So, it must be beneficial to have high amplitude resolution dark dark frames, for EAA typical sum exposures 10-15 min. But I am still at testing.
I am trying to produce my own precise dark frames, including, for example, information about very-high-noise pixels, based on some scientific articles.
I am using 1000-5000 subs for dark frames.

May I ask you, what value type is used internal for pixel values for stacking?
Would it be possible in SharpCap, at least theoretically, to use 32-bit or float dark frames?

[admin]

Hi,

the IMX385 sensor has a 12 bit ADC, so pixel values coming from the sensor will be in the range 0..4095. SharpCap automatically stretches 10/12/14 bit data to a full 16 bit range to avoid issues with saved images looking very dark, so the pixel data as saved will be in the range 0..65535, but only have values that are a multiple of 16 (note : if you have used colour balance in the camera controls then that may multiply pixel values in each channel by some constant, so then they will not be multiples of 16, but the quantization still exists).

So, picking 135 gain as an example, at that point you have 1e/ADU (12 bit) or 1e = 16 ADU (16 bit). The read noise is ~1.2e, or 19.2 ADU (16 bit). That means that to bring the noise level of a dark frame down to about 1 16-bit ADU, you would need to average about 370 (19.2 squared) individual dark frames. Actually it would take more than that, since we have neglected to include the shot noise of any electrons registered by the sensor in the individual dark frames as an additional amount of noise.

This sort of calculation indicates a more general idea that you have to be careful when thinking of using higher bit depths, as in some case all you do is end up sampling and recording the noise in more detail. This is also true for capturing at high bit depths when the camera gain is high.

As for the internal data type - for live stacking in SharpCap, pixel data is either a 32 integer bit value (default stacking algorithm) or a 32 bit floating point value (sigma clipped stacking).

cheers,

Robin

[andriy_melnykov]

Thanks Robin for calculation and explanation!

Looks like 500 darks would be a sweet spot for me needs.