CMOS Sensors - Comparison of Few Longer Subs and Mnay Shorter Subs

[Gina]

I'm sure this has already been discussed but please bear with me.  I know that in general a CMOS sensor wants many short subs whereas CCD wants fewer but longer subs.  The question is how far to go with this.  Also, CMOS sensors have a gain setting.

What I want to know is whether I would capture as much very faint nebulosity with say 100 2m subs as 50 4m subs, other things being equal?  Also, what gain setting for the ASI1600MM-Cool for imaging DSOs, particularly nebulae.

[kens]

Its more the case that CMOS sensors generally have lower read noise so shorter subs are possible. You want the subs to be long enough on any sensor type, that the shot noise dominates the read noise. That way, stacking more shorter subs approaches the efficiency of longer subs in terms of getting a given SNR versus integration time.

The length of the subs needed for shot noise to dominate read noise depends on gain and sky background. You want to expose so that the sky background equates to somewhere in the order of 3 to 10 times the read noise squared. When you can control the gain, lower gain gives more dynamic range but higher read noise. Once you get beyond 10 x read-noise squared you are getting more saturation with little gain in stacking efficiency. Below 3xRN squared you need more integration time to get SNR equivalent to longer subs.

For the ASI1600, the optimal minimum gain is around 76. If your sub lengths at that gain are workable then use that. For NB you may need a higher gain to get workable sub lengths but with higher gain you lose dynamic range. A maximum gain of 200 gives a reasonable dynamic range - 10.5 stops; which gives some room for stretching. If you want something in between then unity (139) is a good choice. You don't need gain to be more precise than those three values and you don't need  to be too precise with the actual values. I tend to use 70, 140 and 200 as they are easy to remember.

Another factor with the ASI1600 is that it has a 12bit ADC so quantization noise is a consideration, especially at low gain. That can be overcome with more subs - say 40 or more. You are talking 50 to 100 subs so that is not really an issue for you.

 

 

 

[wimvb]

Have you seen this article by zwo? 

https://astronomy-imaging-camera.com/tutorials/cooled-asi-camera-setting-in-ascom-driver.html

[Gina]

I hadn't seen it but not sure of it's significance to NB imaging as their tests are with L filter. 

[andrew s]

1 hour ago, Gina said:

I hadn't seen it but not sure of it's significance to NB imaging as their tests are with L filter. 

Photons are photons it matters not in this case if wide or narrow band as long as the photon creates a single photo electron,  which is the case here QE not withstanding.

Regards Andrew 

[Gina]

The point is a Luminance filter lets a lot more light through than a 3nm NB filter.  This makes a big difference.

[Allinthehead]

Hi Gina. Before i sold mine i got decent results using unity gain with 150 second subs at f/2 and 300 second subs at f/5.5 with a ha filter 5nm.

I tried 30 and 60 second subs at higher gains but i found the results to be superior with the longer subs at unity.

Edited January 13, 2020 by Allinthehead

[andrew s]

15 minutes ago, Gina said:

The point is a Luminance filter lets a lot more light through than a 3nm NB filter.  This makes a big difference.

Not really. A faint object in L is the same as one of a similar luminosity in H alpha. It's just photons per second at the detector.  Obviously lower rate takes longer to get the shot noise below the read noise. 

[symmetal]

I agree with kens post.  Here's a chart I'd previously made showing the sky background level to achieve the read noise swamping values that kens mentions. I use unity gain all the time (laziness to avoid multiple darks). Also the offset used affects the sky background level required. I use a constant offset high enough to avoid black clipping at any gain setting. The sky background is pretty much the median value of your image, if the image doesn't have large amounts of nebula in it, or hover the curser over the darkest area of your image to get the actual sky background value.

If you use different gain or offsets to those shown I can make a chart specific for you if you want. I did post the excel spreadsheet file on a previous post so you can fill in your own values on that if you have excel or OpenOffice.

These are really only useful for LRGB imaging. For narrowband, you'll never achieve the required sky background level unless you're in a very light polluted area. I would need to expose well over an hour to achieve it with Ha so I just expose as long as convenient, normally 600 seconds.

Half unity gain is actually 79 and twice unity is 199 but I've used the figures Zwo prefer. (6.0dB (or 2 x gain) is 60 ASI gain steps)

With my bortle 3 skies I find 60s for L and 180s for RG and B gets me close to 10x RN^2 sky backgrounds.

Alan

Edited January 13, 2020 by symmetal

[Gina]

Where there is a very high contrast DSO such as M42, how low a gain can I use effectively?  In other words, is there a limit where reducing the gain no longer gives extra dynamic range?

[wimvb]

1 hour ago, Gina said:

Where there is a very high contrast DSO such as M42, how low a gain can I use effectively?  In other words, is there a limit where reducing the gain no longer gives extra dynamic range?

I don't know if anyone ever investigated that. But I would follow zwo's data and use 0 gain which is supposed to have the highest dynamic range and full well. An object with such high contrast will still need a combination of exposures, if you want to go really deep.

 

[Gina]

Thanks Wim.  I'll try gains of 0 and 60 and see what, if any, the difference is.  What I wondered is that the ADC is 12 bit and the data collection is 16 bit.  Guess the gain adjustment must be in the amplifier before the ADC.

[wimvb]

The ADC is 12 bit, which means that each AD unit (ADU) is a digital number 16 (2^4) in the image file. For the ASI1600, gain setting 0 is 5 e/ADU. Since the largest digital value = 4095 (2^12 - 1), this gives a full well of 20000. The graphs on the ZWO website are worth studying.

[Gina]

Yes, I have studied the graphs and what you say agrees with what I thought.  Although the read noise comes down at higher gain, the dynamic range is still best at zero gain.  I guess the optimum gain depends on how long the subs are - you need longer subs at low gain to overcome the noise level.  That determines the sensitivity.  OTOH the more subs the better the S/N.  But I guess some things cancel out and it all comes down to total integration time.

[symmetal]

The ADC is 12 bit as you say and the gain is in the amplifier before the ADC. To give a 16 bit output the 12 bit value is just multiplied by 16, so adding 4 least significant bits of zero value to the 12 bit value to give a 16 bit output.

I need convincing as to whether using gain values less than unity has much benefit. The camera ADC can only determine 4096 levels so the 20,000 well capacity, along with its dynamic range can only be converted to digital by reducing the pre-ADC amplifier voltage gain to 1/5 of it's unity gain setting.

One 5 min exp at gain 0 must give the same dynamic range as stacking five 1 min exp at unity gain. The only difference is 5 read noises at unity verses 1 read noise at 0 gain. The read noise at gain 0 is significantly higher than at unity as the bulk of the 'read noise' is caused by the ADC itself and not the amplifier or read out circuitry before it. If the read noise is swamped by the skyglow then it becomes insignificant anyway.

The stacked read noise of 5 unity gain exposures I'm unsure about as this article states that the read noise value must be squared before adding it to the shot noise and dark current etc., and then taking the square root of the result as the overall noise. This implies the read noise is more significant. Hopefully vlaiv can explain this bit if he sees it. 

Alan

[Gina]

8 minutes ago, symmetal said:

One 5 min exp at gain 0 must give the same dynamic range as stacking five 1 min exp at unity gain. The only difference is 5 read noises at unity verses 1 read noise at 0 gain. The read noise at gain 0 is significantly higher than at unity as the bulk of the 'read noise' is caused by the ADC itself and not the amplifier or read out circuitry before it.

Alan

This is the way I've been thinking.  Increasing the gain reduces the read noise the graph says but this is read noise converted to per electron.  If the read noise is in the ADC then reducing the gain below unity can't achieve any benefit, surely?

[symmetal]

24 minutes ago, Gina said:

This is the way I've been thinking.  Increasing the gain reduces the read noise the graph says but this is read noise converted to per electron.  If the read noise is in the ADC then reducing the gain below unity can't achieve any benefit, surely?

The read noise is only expressed in electrons rather than micro volts etc., so that the graphs have the same units. The read noise from the ADC is the same, so the higher the signal input value to the ADC, the higher the S/N coming out.

Alan

Edited January 17, 2020 by symmetal
typo spotted by Gina.

[Gina]

I think you mean "so the higher the signal input value to the ADC, the higher the S/N coming out."

[symmetal]

2 minutes ago, Gina said:

I think you mean "so the higher the signal input value to the ADC, the higher the S/N coming out."

Oops, you're right . I've gone back and changed it.

Further to my previous post, increasing the amplifier gain increases the signal shot noise too, so the total noise contribution from the ADC decreases. 

Post 3 from this CN post explains it more fully.

Alan

[Gina]

Is this saying that the optimum gain is unity (139)?

[symmetal]

4 minutes ago, Gina said:

Is this saying that the optimum gain is unity (139)?

Yes, I use unity gain (139) all the time for all filters. Big advantage is a lower collection of darks needed and you don't have to keep checking what the gain setting is. 

Alan

[Gina]

That makes sense.

[kens]

6 hours ago, Gina said:

Is this saying that the optimum gain is unity (139)?

There is nothing magical or optimal about unity gain. It is just a value of amplification that causes one ADU to rcorrespond to one electron captured by the sensor. 

The charge on the sensor is converted to a voltage which is then amplified before being converted to a digital value in the ADC. Collectively, the errors introduced are quantified as read noise.

If you are aiming for maximum dynamic range then try zero gain. But if you find that for a given integration time you have only a small number of subs then quantization error could be manifested in the result. If so the using a higher gain and more subs may be beneficial - but at the expense of saturation of brighter parts of the image.

[gorann]

This interesting thread has so far only been about gain, but you also need an offset setting. Like many others I always use gain 139 and offset 50 on my ASI1600 to keep things simple and limit the number of master darks needed. Then I adjust exposure time according to the object and filter. For NB I use 10 - 15 min and for Lum I use 2-5 min (or some shorter additional ones if there are very bright stars in there to be tamed). For M42 I would certainly use at least 2 different exposure times, I have not used RGB filters since I have a double rig and collect RGB with my ASI071 OSC.

PS, On nights with many clouds passing then using many short exposures could be beneficial since less subs will have to be thrown out,

Edited January 18, 2020 by gorann

[symmetal]

On 18/01/2020 at 09:26, gorann said:

For NB I use 10 - 15 min and for Lum I use 2-5 min (or some shorter additional ones if there are very bright stars in there to be tamed).

Do you have very dark skies gorann? I reach the sky background 10xRN^2 level at around 60s for lum with my Bortle 3 skies.

For gain 139, offset 50 as you use, the sky background10xRN^2 ADU level (16 bit) is 1290. If the median of your subs is much greater than this you can take more shorter subs and so reduce star bloating and get a bit more dynamic range. 

Alan