Fullwell Capacity – Does it really matter?

Christian,

The dynamic range is what matters. As you've noted, you can have the same DR with different FWDs if you can tune the read noise.

Apparently through dynamic range our goal is to set different spatial locations into relationship (e.g., brightness of star A vs. brightness of star B or spot of a nebula) and if we want to capture that relationship properly w.r.t. intensities, then our measurement system (sensor + all down-stream electronics and software) needs to be able to 1. capture both without clipping anything and 2. have the weakest signal at a level such that it can be appropriately measured against the noise. (obviously, for nice color images, stretching the image is throwing the "exact" measurement out of the window unless you'd know the intensity transformation and calculate backwards… which I wouldn't).

Another way to think about it: why do you expose for several minutes? Because you're hunting the faint stuff (gas shell of a planetary nebula). Since you may only have a few electrons per hour (!!) generated in your pixel looking at the faint part of the field of view, the noise contributions (e.g., read noise) may drown your signal (not to mention all the other things drowning faint signal like light pollution and full moon).
Unfortunately the pixels looking at the bright star near by got filled with photons and electrons. So if your goal was to get the faint signal (which fights against the noise "at the bottom of the well") and still put it in a "realistic" relationship (intensity wise) with the bright star it only helps to make the well deeper.

The alternative approach would be to use exposures which are short to not saturate the star and capture an enormous amount of frames. Nevertheless, due to the many readouts, the SNR for the faint object will take more integration time to reach the same SNR that you would achieve if you stick with the long exposure times.

Regarding spectroscopy: A spectrometer will disperse the signal of the object over the sensor and by that even a saturated star can produce a dim spectra. Also here the dynamic range plays a role as one want's to measure the relative intensities of the spectrum which is now dispersed over the sensor, leading to basically the same idea as in the second paragraph.

Björn

I almost never care about saturated stars unless I want to do photometry.  I (and arguably many other astrophotographers) want to image faint nebulas, and this requires extreme contrast stretching during the processing.  Under such stretching, even stars that are far away from saturation in the raw files can have white cores in the final picture.  So why bother?

Some people worry that saturated stars will not show color. This is rarely the case. A saturated core will not have color, of course. However, for every saturated core pixel, there are many more pixels around it that are not saturated. Those outer pixels can still show beautiful color.  Indeed, a star with colorful outer pixels and with less color in the core, nearly white pixels is how our eyes perceive color (when brightness becomes very high, color fades away).  Some processing try to make core of bright stars as colorful as their outer fainter pixels.  That's never my cup of tea.

BTW, CMOS pixels are getting smaller and smaller.  A natural consequence is that their full well also decreases.  However, the full well per unit area does not necessarily decrease.  When a pixel is smaller, the number of photons it receives under fixed exposure time also decreases.  So it doesn't necessarily become easier to saturate. Indeed, from photometry's point of view, small pixels with shallower full well may help to confine saturation to a smaller area (e.g., a saturated 9 um pixel replaced by a saturated 4 um pixel).  Then the larger unsaturated area of a star can be used to reconstruct the information in the saturated area, making a saturated core less a problem.  I know the OP is comparing different setting of the same camera.  Here I just want to remind people that when comparing different cameras, don't just look at the face value of full well potential. Pixel size also matters if saturation is a concern.

Hello,

maybe I missed to point out that I only do "pretty picture" astrophotography and no photometry - probably like most of us.
And of course my results only apply to my individual setup - although your setup might be somewhat similar, deriving similar conclusions.

No intention to draw a general conclusion for everybody!


@Joe Linington : You might want to use a dual-approach:
1. Take long subs at high gain for the faint nebula (long meaning as long as your mount, guiding and light pollution can handle). Remove the stars from that image during post-processing.
2. Use the short subs for the stars only and insert those into your deep image.
Maybe 2 minute subs are too short for getting the IFN around M81, unless you are using a hyperstar f/2 telescope.
You might want to check out my Leo Triplet wiith tidal tail: https://www.astrobin.com/2vwq9x/
I used even 10-minute subs here with my f/5.6 setup (no saturation of the galaxies at this exposure length and High Gain Mode / Gain 56!).


Another - purely practical - thing about short exposures: You'll end up with a terribly large amount of data. One sub of my QHY600 comprises 120MB and for standard deep sky I shoot somewhere around 30x 2-minute exposurse R, G and B each plus 60-100 3-minute exposures L. So alone with that I'll end up with 22GB of data. Multiply that by 3 for a short exposure approach and you will fill up a 1TB hard drive within a year.
Not to mention the much longer processing time in stacking / pre processing....


CS
Chris

Chris White- Overcast Observatory:
but how are you determining that a star is saturated?

Chris,

in Fitswork - which is a free software - you can draw a rectangle over a star or certain area and ask for pixel evaluation.
It provides the minimum and maximum pixel values, as well as a mean and the standard deviation.

So I did not look for a certain portion of the stars' PSF to be clipped, but I simply checked for the maximum value in my area of interest.
Since I had calibrated all my frames before, there were no hotpixels left.

Values are provided in the original 16bit scheme - maximum possible value 65353.

CS
Chris

Nice experiment, Christian. Always very informative to see actual data on topics that seem at the surface pretty straightforward.

From your first sheet, it appears that Gain 0 takes 4 times as long to saturate than Gain 56. This is in line with expectations. The pixels don't reduce in capacity, they are just reaching their maximum ADU value sooner, because the signal gets amplified. That's good news for weak signals, bad news for star centers. But the most important parameter is Dynamic Range. Modern sensors switch to different analogue circuitry on the chip at the High Conversion Gain mode, which causes a sudden drop in read-noise. DR is a function of FWC and read-noise, and the result is an almost equal DR compared to gain 0. So with almost no penalty you can switch between Gain 0 (good for high signal) and Gain 56 (good for low signal). That is what makes these modern sensors so interesting. Which gain to use depends on your target. For star clusters I typically use Gain 0, whereas for a faint galaxy I use the equivalent of Gain 56.

Things get different when you're comparing different systems.

When I started this hobby I bought the ASI1600. That has FWC of 20k. My much newer ASI6200 camera has FWC of 51k. That is a 2.5 fold increase in FWC, and that is a pure win, no trade-offs. So all the values you mention get better by a factor of 2.5.

Another aspect to take into account is pixel-scale. Your system has a pixel scale of almost 1.5. If you put the same camera on a 1000mm scope, the pixel scale is about 0.8. Your scope concentrates about 4 times as much sky on the same pixel and thus 4 times more pixel-saturation. That is why it's called a 'faster' scope. That's good for faint signal, not so good for star centers.

So an ASI6200/QHY600 on a 1000mm scope at Gain 0 will saturate stars 4*4*2.5 = 40x slower than an ASI1600 on a 550mm scope at Gain56. Both are very common scenarios, but with very different outcomes.

A whole different question is how bad it is to have a center pixel of a star being clipped. The brighter the star, the more pixels it will cover and they will show a gaussian distribution in brightness around the center point. The way we experience that star is through the whole disc. So as long as the majority of pixels in that disk are not saturated, we're good and we can process, deconvolve, colorise, etc.

So to answer your question, yes FWC does matter. More is always better, but a clipped star center is difficult to completely eliminate. But not all systems are created equal and clipping does not have to occur in seconds.