Noisy after 6.7 hours? (cross-posted to CN)

[rickwayne]

So here's an image I've been working on lately. I leaned pretty hard on it for noise reduction, so for this example I'm pulling all that out.

137x180", ZWO ASI183MM at gain 178, ZWO 7-nm H-alpha filter, f/4.6 StellarVue SV70t-IS. (It's Sh2-155/Caldwell 9, called by some the "Cave Nebula".)

I had a suggestion that 3 minutes was way too short for narrowband, and yet that's within the ballpark for my bright urban skies with the moon out (when the 183 first came out there was a good thread on it). Also that it was "overprocessed", which I took to mean stretched too hard. Perhaps.

I'm concerned that with the residual periodic error in my CEM25P, if I go much longer I'll really have monster stars, probably egg-shaped too. Your thoughts? Vlaiv, I oughta be able to go by total integration time so long as I'm not down in the read noise for individual frames, right?

[vlaiv]

There is a bit of "everything" here.

You are probably oversampling a bit at 1.54"/px with 70mm scope. Image looks like sharpened and star mask of sorts was used during processing.

Sharpening will emphasize the noise and how much you stretched - will certainly have impact. You should learn to stretch your data only to a certain point - only as much as it will let you. Once you start forcing it - it will show.

However, I think that you could do some nice noise control with this image.

I doubt that extending sub length will have major impact on resulting SNR. It will have some, but I doubt it will show in the image. I believe it is much more important how you process the image and possibly how you work with your data.

Is this image a crop or resize of your capture? If it is resize - why not use binning instead if you are ok with having smaller image? That will deal with any possible oversampling and will improve SNR.

Would you mind posting your stacked linear data, I would like to have a go at it to see what I can pull from it.

[rickwayne]

I'm usually working toward a print for the wall and so want the maximum pixel size I can extract. Although I might have to settle for smaller with this one. With a CMOS camera, binning may as well be done in processing, no? Unlike CCDs they don't bin in hardware, so you'll wind up doing some kind of software downsampling anyway.

I had to double-check, but no, no sharpening at all was applied. I did use starnet++ (the python command-line script) to eliminate the stars and 99% of the processing was done on the starless  image, recombining at the end. That definitely left some haloing artifacts, which I actually hand-edited a bit to reduce.

This FITS is a crop of the original crazy-quilt mosaic (I was messing with the rig and no two nights had the same camera orientation). This is about 55% of the original frame size.

Thanks for having a look.
 

Sh2-155_6_nights-H_Alpha-session_1_session_2_session_3_session_4_session_5_session_6--34degCCW-1.0x-LZ3-NS-mod.fits

[vlaiv]

A few findings and result.

First, most stars have more than 3.2px FWHM - which means you should really bin x2 (in software, yes, CMOS sensors should really be binned in software not at capture time as it allows more flexibility and is in principle the same thing unlike hardware binning of CCD sensors).

Second - you pushed data too much. Only bright part of this image is Cave nebula itself - surrounding Ha region is very faint. You can bring it out a bit but it needs very delicate denoising to be shown with 3 hours of exposure that you made.

Here is my result:

Stars are a little bit bloated from excessive stretch and yes, using Starnet++ to remove them would be good approach, you just need to be careful when blending them back in to make them look natural.

[geordie85]

Your HA data looks fine to me. Noise starts to show in any image if you push the data too far.

[rickwayne]

Thanks man! All good advice. (Um...3 hours? 180 x 137 is just shy of seven hours, no?)

I'll try the Photoshop stuff again and see what I get.

Thanks too to Geordie.

<sobs>"Not stretch it so hard ? But I LOVE drama!" ;-}

 

Edited November 10, 2020 by rickwayne

[vlaiv]

1 minute ago, rickwayne said:

Um...3 hours? 180 x 137 is just shy of seven hours, no?

Yeah, sorry about that - brain doing funny things - saw 180, said to myself 3 (minutes / exposure) - jumped to 3h total for some reason

[ollypenrice]

Having just had a play with this data I don't think it's all right, I think it ought to be significantly better. The noise floor is very high. I've imaged this target twice with my old school Atik 11000 (50% QE) at two apertures, 140mm and 106mm, and my S/N is considerably better than this. The noise also has an odd look to it. It's a slightly elongated 'grain' to my eye.

My first suspicion is that calibration may not be working as it should. I would go through everything you can, particularly darks. Right settings, right temp, etc. Then I'd check the stacking parameters again.

Olly

Edited November 11, 2020 by ollypenrice
typo

[vlaiv]

6 minutes ago, ollypenrice said:

Having just had a play with this data I don't think it's all right, I think it ought to be significantly better. The noise floor is very high. I've imaged this target twice with my old school Atik 11000 (50% QE) at two apertures, 140mm and 106mm, and my S/N is considerably better than this. The noise also has an odd look to it. It's a slightly elongated 'grain' to my eye.

My first suspicion is that calibration may not be working as it should. I would go through everything you can, particularly darks. Right settings, right temp, etc. Then I'd check the stacking parameters again.

Olly

Pay attention to the fact that you imaged it with camera that has 9µm pixel size with 106mm of aperture. My guess is that you used FSQ-106ED with 530mm of focal length so your sampling rate was 3.5"/px.

If we calculate some sort of "speed index" - in form of resolution * aperture = 106mm * 3.5 = 371, we can compare to above setup in some ways. On the other hand, here we have 1.54"/px with 70mm of aperture so "speed index" will be 107.8.

Ratio of the two is ~3.44, or if we use square values (and we should since we want aperture area and pixel area) - it is ~11.84.

Above image is equivalent of about 35 mins of imaging with 106mm and 9um camera (it would be a bit more if we take into account QE of both cameras at 656.8nm as ASI183 is likely to win there). On the other hand, sky brightness has some impact since this is NB, so I guess your dark skies would win there.

Would you expect noise level in this image to be that of less than one hour of ATIK 11000 and FSQ-106ED for Ha on this target?

[rickwayne]

I know for  sure that at least one night was terminated by clouds. Probably should have looked at the SNR more carefully but I was going exclusively by the APP "quality" score and eyeballing any frame that had  a low score.

Pretty sure the darks were spot on. I only use two gain settings these days, one offset, one temperature. So if I pick the right  gain and exposure time -- and I did -- there's not a lot to go wrong there. The darks  were created with same setup, 100 frames. I shot them and the bias (500  frames) in May of this year. The bad-pixel map is older, from last year, probably ought to replace that one.

Oh, and I reprocessed the image last night. I still leaned a little heavy on the "drama" side but what can I say, I like bold images. I'll just attach the full-res image, pardon if the 9 MB file is annoying.

[ollypenrice]

1 hour ago, vlaiv said:

Pay attention to the fact that you imaged it with camera that has 9µm pixel size with 106mm of aperture. My guess is that you used FSQ-106ED with 530mm of focal length so your sampling rate was 3.5"/px.

If we calculate some sort of "speed index" - in form of resolution * aperture = 106mm * 3.5 = 371, we can compare to above setup in some ways. On the other hand, here we have 1.54"/px with 70mm of aperture so "speed index" will be 107.8.

Ratio of the two is ~3.44, or if we use square values (and we should since we want aperture area and pixel area) - it is ~11.84.

Above image is equivalent of about 35 mins of imaging with 106mm and 9um camera (it would be a bit more if we take into account QE of both cameras at 656.8nm as ASI183 is likely to win there). On the other hand, sky brightness has some impact since this is NB, so I guess your dark skies would win there.

Would you expect noise level in this image to be that of less than one hour of ATIK 11000 and FSQ-106ED for Ha on this target?

Understatement of the year?  😄  I hear what you're saying but I recently processed a multi-gigabite mosaic from a CMOS OSC which, without specific Ha filtration, had dug deeper into the Ha signal in a region I've imaged with my Atik 11000, and the CMOS data was deeper in 90 mins OSC than mine was in many hours of mono/Ha. The CMOS data was shot with a focal reducer, mine not, both in Tak FQ106 scopes. My point is that the enormous CMOS advantage I found in the data I processes is entirely absent from the OP's Cave data.

Olly

[vlaiv]

17 minutes ago, ollypenrice said:

Understatement of the year?

Could be. If we examine the data on two sensors, IMX183 does win and it is about twice as sensitive as KAI-11002.

We have definite data on KAI-11002 sensor:

It is about 32% QE at 656.

We have only relative QE graph for IMX183 and only estimated peak QE.

Again, relative QE curve at 656 is about 78%. Peak QE is often quoted to be 84% so if we multiply the two we get 0.78 * 0.84 = ~ 0.65 = 65%

This would put IMX183 at twice the sensitivity at Ha.

On the other hand Christian Buil found on more than one occasion that published QE graph differs to measured QE graph and peak QE is less. According to his measurements, Ha QE of ASI183 is closer to 51% and peak QE is 80%.

source: http://www.astrosurf.com/buil/asi183mm/

In the end linear measures like aperture and pixel size need to be squared to translate into exposure time (surface is important), but QE is linear dependence.

If we take into account difference in QE and make it so that IMX183 is twice as sensitive - we still have factor of x ~ x5.92 exposure time in favor of Atik11000 + FSQ. This is without taking into account any LP noise that will be roughly x3.2 times larger for 2 magnitude difference in SQM reading for example (even with NB filters - since both setups use NB filters).

[alacant]

Looks good. Nice detail.

[wimvb]

On 10/11/2020 at 20:29, rickwayne said:

ZWO 7-nm H-alpha filter, f/4.6 StellarVue SV70t-IS.

... my bright urban skies with the moon out

A 7 nm Ha filter in urban skies under a bright moon is a very good recipe for noise generation. And I don’t expect budget ZWO filter to block so well off band either. I think that  differences in both gear used and sky conditions, make direct comparisons very difficult.

That being said, I also found the original image pushed a bit too far.

[Hallingskies]

14 minutes ago, wimvb said:

A 7 nm Ha filter in urban skies under a bright moon is a very good recipe for noise generation. And I don’t expect budget ZWO filter to block so well off band either.

Absolutely.  In the absence of 3 micron narrowband filters, just expect a bunch of background noise if you take long exposures in bright moonlight from hazy suburban skies.