Originally posted by duncan1 I've done some long exposure testing with my K10D and a lot of long exposure work with a Nikon D50. In theory if you get good results at 10 minutes with ISO 400, you should get better results - lower noise per image with no loss in bright areas by decreasing ISO to 100 and increasing exposure time per sub. Here's some experimental work that I've done with my K10D:
https://picasaweb.google.com/102361289894036464287/Astrophotos
Images 3, 4, 5, and 6 (left to right) were done with a K10D, the rest with a Nikon D50. These were done at an extremely dark location.
The 14bit K5 should always use ISO100, IMHO because it gives the best possible dynamic range and SNR.
I understand your thoughts about the lower ISO setting giving better headroom. I tried this for a while as well. Used 200 ISO for a long while and then finally moved to 400 ISO for my purposes after I did some testing.
I did a
range of tests with my K10D where I checked response and noise via a script in PixInsight. The procedure is as noted by the script developer Georg Viehoever to follow "the method described in described in "The Handbook of Astronomical Image Processing" (HAIP) by Richard Berry and James Burnell, Second Printing, Errata for chapter "8.2 Basic CCD Testing" (see
http://www.willbell.com/aip4win/Errata%20to%202nd%20%20Printing%202nd%20Edition%20of%20HAIP.pdf). All you need are 2 bias frames, 2 flat frames and 1 dark frame."
This was in the attempt to understand where unity gain should be (.7 e/ADU). The closest I was able to show was at 1600 ISO(!).
What was more interesting was the read noise. For a CMOS chip, it's very important to note the point where the signal will be higher than the readout noise. This difference becomes important once you stat stretching the image. The readnoise for the K10D remains fairly constant across the whole range at 3-4 e. Consider what happens when this noise floor is stretched along with signal. A much less aggressive stretch will be required of a higher ISO setting vs a lower ISO setting. The aggressive stretch will pull this noise floor up as it competes with the signal.
I've also tested another concept:
The sub length is the critical component as the amount of light that falls on the sensor is the same no matter what ISO setting. The test that I did compared a shot of a specified length at ISO 100 through ISO 1600. Then I stretched the results just as I would in astrophotography. What I found was that the noise levels were pretty much the same.
Look at how much the bias shows up in the stretched 100 ISO vs the 1600 ISO. The bias signal is the source of the blue streaks.
It's important to note that it's unknown what kind of optimizations are performed on the raw file as it passes through the camera's software.
Canon has been well-tested and shown that they are messing with the raws before being written. DSLRs are meant to be used in bright conditions so they are setup to give good results rather than looking at the blackness of space.
I have shot at 100 and 200 ISO before and it does seem clean. However, I feel strongly that a lot of signal is "lopped off" the low end at 100 ISO in noise reduction either in the camera or in post. Much of the headroom of the lower ISO is wasted except on bright stars. The background skyglow doesn't climb out of the lowest parts of the histogram unless using longer subs.
This leads to another point which is part of the limits of the rest of the hardware: mount capability.
I shoot 10 minute subs because my mount can handle them very well. It could handle 20 minute subs once I fix the remaining flexure issues. When I shoot widefield, there is no flexure problem so I could use a longer sub.
With all this in mind, there are other limits to the camera other than read noise and ISO setting. The K10D is plagued by strong amp glow that is non-linear in temperature response. Thus, dark frames only go so far in removing this glow. Software that is not able to handle the non-linear glow has significant problems with removing this glow. Deep Sky Stacker has known problems with amp glow and is only able to scale darks a little ways. Because the amp glow is non-linear, scaled darks can't touch it.
I finally stopped using DSS for calibration because of this amp glow. Instead, I've found that Maxim does a much better job of calibrating. Unfortunately, Maxim does not do a great job of debayer (and I've yet to get Pix Insight to do a decent job of it). So I'm still stacking in DSS of the calibrated FITs from Maxim to give a debayered FIT with good hot pixel rejection and mosaic.
FWIW, the QHY10 and Starlight Xpress SXVR‐M26C astrophotography cameras use the same Sony ICX493AQA SuperHAD chip as the K10D. There are further tests about readout and gain levels for these cameras that help to shed light on the way this chip works. Eventually, once I have exhausted the capabilities of my K10D -- or it breaks -- I'll get one of these devices to replace it.