Interesting thermal noise experiment - Forums

Topic starter

May 23, 2023

Here is a 60 second sub of M1 with the moon below the horizon. ~ Bortle 4.
Camera: Omegon veLOX 224 C Color. Sensor temp was -15.5 C. Same uncooled IMX 224C sensor as in the ZWO ASI 224 MC.

And here is a 5 minute sub of NGC 2403 with the moon at the horizon in the beginning of the session and at +18 above at the end.
Sensor temp was +3.8 C. Same telescope, location and snow on the ground in both cases.

One would think that with the sensor 19.3 C warmer, some moonlight, and 4 minutes longer exposure that there would be a lot more noise. However, as you can see it's the complete opposite, the 5 minute sub has significantly less noise.

During the season i managed to take over 40 images and the results are consistent, my 5 minute subs have less noise than the 60s ones and the sensor temperature seems to have no effect on the noise (never got warmer than about +6 degrees during the season).

It would be interesting to see a dark current graph of the IMX 224C sensor, however they only seem to exist for the cooled DSO cameras.

Any thoughts on what's going on here?

This is my first post on Astrobin, seems like a great community!

Cs
Gabriel

Helpful Insightful Respectful Engaging

Andre Vilhena

May 23, 2023

Hello,

A couple of comments:

- The stretch applied may influence the way the noise is visually perceived or, in other words, a stronger stretch will show more noise than a weaker one. If you have moonlight in NGC 2403, the stretch does not need to be so strong to reach the same background level;

- Besides stretch, gradients should be removed to properly assess SNR;

- Noise may also come from the read noise, which is a type of noise constant in each frame, regardless of the exposure time, ie, a short exposure will have proportionally more read noise than a longer one. On top of that, usually read noise is larger than dark current noise (depending on the camera), of course.

So, you should take also these factors in consideration for your assessment - I hope this helps!

André

Helpful Insightful Respectful

Gabriel Wiklund

Topic starter

May 23, 2023 · Edited: May 23, 2023

Andre Vilhena:
Hello,

A couple of comments:

- The stretch applied may influence the way the noise is visually perceived or, in other words, a stronger stretch will show more noise than a weaker one. If you have moonlight in NGC 2403, the stretch does not need to be so strong to reach the same background level;

- Besides stretch, gradients should be removed to properly assess SNR;

- Noise may also come from the read noise, which is a type of noise constant in each frame, regardless of the exposure time, ie, a short exposure will have proportionally more read noise than a longer one. On top of that, usually read noise is larger than dark current noise (depending on the camera), of course.

So, you should take also these factors in consideration for your assessment - I hope this helps!

André

Hello,

The same stretch is applied (automatic in ASI fitsview).. since the stretch does not need to be so strong to reach the background level on the NGC 2403 wouldn't that mean that the noise should be more apparent in the NGC 2403 if the same stretch is applied?

The camera has extremley low read noise (0.75e) and with the high gain at 60s it should not show up according to Robin Glovers talk on CMOS imaging.

Thank you for mentioning gradients, i had not taken that in to consideration. The strange thing though, is that gradients vary from session to session but this phenomena has been consistent through the whole season.

/Gabriel

Andre Vilhena

May 23, 2023

Hi,

So, trying to clarify:

Gabriel Wiklund:
The same stretch is applied (automatic in ASI fitsview).. since the stretch does not need to be so strong to reach the background level on the NGC 2403 wouldn't that mean that the noise should be more apparent in the NGC 2403 if the same stretch is applied?

No, the harder you stretch the more visible the noise is. The comparison below is the same photo, with two different stretches applied:

The boosted stretch on the right makes noise more visible. The issue with gradients is that some areas have more signal (unwanted) than others but the autostretch will try to reach the same median value for the photo. I have added a small pedestal to this photo on the right side to simulate a gradient and after applying autostretch to it, the comparison is as follows:

Clearly, the noise is more visible on the "gradients" free image - the stretch needs to be so harder to reach the same median value. I don't know ASI fitsview but I think it is highly likely that the autostrect acts this way and therefore stretches applied to your images are not really the same.

Gabriel Wiklund:
The camera has extremley low read noise (0.75e) and with the high gain at 60s it should not show up according to Robin Glovers talk on CMOS imaging.

You can't infer that from the talk. He mentions that you need a certain minimum exposure to "swamp" the read noise but that minimum exposure will depend on the light pollution level, quantum efficiency of the camera, f ratio of your telescope, filters used and the read noise itself. Hence, 60 sec exposure may or may not be enough (for my setup it is not enough when I shoot at Bortle 4 sky, even if just for luminance)

Gabriel Wiklund:
Thank you for mentioning gradients, i had not taken that in to consideration. The strange thing though, is that gradients vary from session to session but this phenomena has been consistent through the whole season.

I can't comment on that other than if you want to reach conclusion for the impact of a given parameter, you need to keep all others constant. For what I see in your post, that was not the case and it is difficult to assess that way.

Cheers,
André

Helpful

Gabriel Wiklund

Topic starter

May 23, 2023

Andre Vilhena:
Hi,

So, trying to clarify:
Gabriel Wiklund:
The same stretch is applied (automatic in ASI fitsview).. since the stretch does not need to be so strong to reach the background level on the NGC 2403 wouldn't that mean that the noise should be more apparent in the NGC 2403 if the same stretch is applied?

No, the harder you stretch the more visible the noise is. The comparison below is the same photo, with two different stretches applied:

The boosted stretch on the right makes noise more visible. The issue with gradients is that some areas have more signal (unwanted) than others but the autostretch will try to reach the same median value for the photo. I have added a small pedestal to this photo on the right side to simulate a gradient and after applying autostretch to it, the comparison is as follows:

Clearly, the noise is more visible on the "gradients" free image - the stretch needs to be so harder to reach the same median value. I don't know ASI fitsview but I think it is highly likely that the autostrect acts this way and therefore stretches applied to your images are not really the same.
Gabriel Wiklund:
The camera has extremley low read noise (0.75e) and with the high gain at 60s it should not show up according to Robin Glovers talk on CMOS imaging.

You can't infer that from the talk. He mentions that you need a certain minimum exposure to "swamp" the read noise but that minimum exposure will depend on the light pollution level, quantum efficiency of the camera, f ratio of your telescope, filters used and the read noise itself. Hence, 60 sec exposure may or may not be enough (for my setup it is not enough when I shoot at Bortle 4 sky, even if just for luminance)
Gabriel Wiklund:
Thank you for mentioning gradients, i had not taken that in to consideration. The strange thing though, is that gradients vary from session to session but this phenomena has been consistent through the whole season.

I can't comment on that other than if you want to reach conclusion for the impact of a given parameter, you need to keep all others constant. For what I see in your post, that was not the case and it is difficult to assess that way.

Cheers,
André

Before i answer i just want to say first that gradients seem to explain a lot. Thanks for making me aware of that!

With regards to stretching i am sorry i did not explain in a clear way. What i meant was that since the moon adds light pollution (noise) and the images are equally stretched, the sub with the moon (NGC 2403) should have more noise.

I see now that i did not understand the following: I am very familiar with stretching bringing out more noise but since the moon image requires less stretching to reach the same background level as the moonless image, the moonless image will be a more stretched image hence more noise is visible?

Nevertheless your answer solved the riddle including the following: "Clearly, the noise is more visible on the "gradients" free image".
Since the moon caused a gradient on the NGC 2403 image that would explain why it looks cleaner than the M1 image.

Looking back i have been puzzled when processing images first opening a stack from DSS looking very clean, then after running Background Extraction the noise is there, it makes sense that if unwanted signal is removed the noise is what remains.

Ps. i should have made clear in my answer that i actually did the calculations with regards to read noise.
From SharpCaps Sky Background Calculator i get a Sky Electron Rate of 0.61e/pixel/s (Bortle 4, my f/ratio etc)
Read noise at 250 gain is 1.0e

1x1 = 1
0.61 x 0.61 = 0.3721

1 / 0.3721 = ~ 2.69.

x10 = only 5% extra noise (what Robin recommends)
x25 = only 2% extra noise

10 x 2.69 = 26.9 seconds
25 x 2.69 = 67.25 seconds

So in other words a 60 second sub is pretty close to only 2% extra noise.

At last i fully agree that to draw a conclusion for a given parameter all else should be constant!

Thank you for your patience

Cheers
Gabriel

Helpful Respectful