Telescope: Dobson Orionx XX16g 16" with 1800mm focal length and coma corrector
Capture: 19000 full frame RAW of 1 second duration with Sony A7s at 6400 ISO
Siril processing:
FITS compression: Used 32Bits float Rice FITS compression with a quantification factor of 32: the 18800 full frame RAW use 600Gb on the SSD disk instead of 2700Gb…
Preprocessing: 1000* darks 1s 6400 ISO with dark optimisation, offset and flats
Alignment: Global star alignment to handle field rotation of Dobson mount
Stacking: Linear fit using 18000 images from a total of 18800 (using 10000 increased the resolution but the SNR was not good enough )
Crop of the stacked image (field rotation)
Image was finalized with PixInsight
Please find below an example of one of the 1 second frame used for the stack to see the improvment of the stacking with a low read noise sensor and high number of frames:
I just noticed that the exposures are 1sec each, so this could probably be done in a couple of nights. By any measure, that’s a LOT of data to register, stack and process… hat’s off to @fab4space!
Thanks all for your comments!
I shoot the images during 4 different nights with approximately 1 to 1.5 hours for each shooting session (night starts too late in June which prevents me to have a long shooting session )
The RAW images are 12Mb each, when converted to 32bits FITS without compression the FITS file is 135Mb so for 18800 images it would have taken 2700Gb on disk without compression!
With the FITS compression feature that I have implemented in Siril, the FITS image is 30Mb using Rice compression with a quantisation factor of 32, taking 600Gb for all the images (see https://gitlab.com/free-astro/siril/-/commit/15532560e3de1e483b199b715f5ccc4263f02c98)
The stacking with normalisation and linear fit clipping took 14 hours to complete with a Xeon E3 1270V3 and a Samsung SSD EVO 860 for the data, but most of the tests were done first with sum stacking without normalistion which took around 1 hour.
Feel free to ask if you have other questions.
Thanks!
Wonderful to see Fabrice! @fab4space
Thanks for explaining your procedure as well. One question I have: in terms of quality, could you have done the same with fewer images?
Fewer image means less signal, more noise. Consequently it is more difficult to reveal details. Signal-to-noise ratio evolution is proportional to the square root of the number of images.
@lock042 Thank you for the reply. However, even with an increasing SNR, I assume there is a point where you can hardly see the difference between e.g. 13.000 and 14.000 images. So my question remains, could you work with fewer images and still get the (visibly) same result?
Impressible results! you you show us the raw stack file, just before the PixInsight treatment begins? I’m curious to see how this 18.000x1s images looks like.
The “usual” stack I’ve seen described at DPR’s astro forum numbers in the dozens to hundreds of captures, each with ~30sec exposure. So, in terms of “light on sensor”, I think 19000 1sec images would be similar to about 600 30sec images… ???
The advantage I see with regard to short exposures is less reliance on precise tracking, which I’d imagine is more challenging with longer focal lengths.
Am I hunting in the right preserve here? I live within an hour’s drive to some of the best dark sky in the US; what better social-distancing endeavor is this??
That really depends of the read noise. With the newer Sony sensors it is really close yes. With the benefits of better resolution.
But if you take a Canon for example it will be better to shoot 1min/2min each frame.
Hello, thanks for your messages.
The advantages for this technique are:
a basic mount can be used : no guiding is required and even a non equatorial mount can be used like the dobson altaz mount that I am using. The mount setup can be done quickly (alignment,…)
higher resolution can be achieved than with regular long exposures: thousands of pictures with bad seeing or tracking issue can be eliminated easily with siril stacking based on fwhm, I will show this improvment on another target (M64) which is more suited
Please find below the improvments obtained by using 18000 images compared to using respectively 500,1500,7000 images (check in fullscreen the sky background and galaxy extensions with the 7000 version compared to 18000)
@fab4space sorry if I go completely off-topic but is your avatar, by chance, a result from some kind of stratospheric balloon launch that you did or something?
@fab4space, It is a totally unexpected and counter-intuitive result. I understand the point about guiding (and tracking, I guess) and mount. Although I guess you’d get some image rotation over the course of the night, Siril would probably correct for that.
But noise. There are a number of sources of noise, Read noise, Shot noise and light pollution and they don’t all pull in the same direction. But if you are using an untracked mount, I’m guessing 1 or 2 secs is probably about right.
You’ve managed to get a fantastic amount of detail, and I imagine that’s your lovely 400 mm F/4.5 Dobsonian at work. I can’t get anything like that on my 200 mm F/5 Newtonian.
In my own images, I’ve gone up to 300 secs, and down to 30 secs. I find 180 s a happy medium, but tracked and guided. If I was unguided, I’d probably drop to 45 secs.
I’ve also experimented with ISO from 400 to 6400, and again, I find that 1600 is better.
My main battle at the moment is with fixed pattern noise (FPN) on my DSLR. I’m experimenting with dithering to remove that, but it’s a slow and painful process. Does your method help with FPN?
It’s always nice to have our preconceptions challenged and you’ve certainly done that.
