Back to forums

585sensor at zero gain

CMOS Acquisition Deep Sky
15 replies•614 views
Tony Gondola avatar
Tony Gondola avatar
Has anyone tried shooting a 585 sensor at zero gain? I typically run my ZWO version at 255 so that HGC kicks in and drops the read noise but I do give up well depth and dynamic range. Generally this works well as I like to keep my subs short. I haven't run into any dynamic range issues at 255 so I wonder of the extra 1.5 stops of dynamic range would be helpful?

Here are the performance charts for the sensor.

dkamen avatar
dkamen avatar
You use lower gain if you are imaging something very bright because dynamic range is your main concern.

You use the higher gain if you are imaging fainter stuff, in which case read noise is your main concern.

Almost every value of gain can be optimal depending on other factors, such as your f/ratio, light pollution levels, exposure time your gear is comfortable with.

I think outside of specific cases such as very short exposures or when you want to capture as much star color as possible (note that big stars will almost always saturate at the center, the question is how large will the saturated radius be), the point where HCG kicks in is just fine (the difference between 255 and the 250 given by the manufacturer is too small to matter).

You say you do not have DR problems so if you were to deviate from 255, it actually makes more sense to try a higher gain than a lower one, assuming you cannot increase exposure time due to mount limitations.
Helpful Insightful
Tony Gondola avatar
Tony Gondola avatar
I'm in Bortle-8 shooting at F/6. Exposing for as long as needed isn't a problem although I do like to keep the subs as short as possible. I know I'll have to go longer than I like at zero gain in order to swamp the read noise. I am curious as to why going with a gain higher than 255 would be advantageous? I wouldn't gain much in read noise and I would loose dynamic range.
Rainer Ots avatar
Rainer Ots avatar
I guess going to even higher gain would only make sense in case of using very narrow bandpass filters and especially in very low bortle areas as then the sky background electron rate gets so low that suddenly read noise is very important.
Especially as the time required to swamp it is dependent on the square of read noise. So going from 1.2 to 0.75 (guessing from the chart) would mean a time decrease of 2.6x.
And of course nothing wrong with using gain 0, but then again swamping read noise will mean time increase of about 29x (square of 6.5/1.2). But in bortle 8 the sky background is so high that it just means going from a few seconds to a few minutes so … whatever smile
Helpful Insightful
Tony Gondola avatar
Tony Gondola avatar
I did a zero gain test session last night on M82 with 180 sec. exposures. Lots of clouds about so only got about 3 hours. The increase in dynamic range was more noticeable than I expected. Next time out I'll shoot with my normal 15 seconds at 255 gain and do an A-B.
Jean-David Gadina avatar
Jean-David Gadina avatar
I did one attempt with zero gain on Comet C/2023 A3 in October, precisely because I was interested to see the difference in dynamic range.
Unfortunately, I had many issues during this session with the HyperStar on my C6.
My data was really bad. Huge reflections and a lot of walking noise. So I cannot really tell if it can be useful.

My previous attempt on the same comet at gain 252 was way better, so I never tried again and I now stick with 252, unless I'm doing lunar or planetary imaging.
Looking forward to see your comparison, this is going to be interesting.
Helpful Engaging Supportive
dkamen avatar
dkamen avatar
Tony Gondola:
I'm in Bortle-8 shooting at F/6. Exposing for as long as needed isn't a problem although I do like to keep the subs as short as possible. I know I'll have to go longer than I like at zero gain in order to swamp the read noise. I am curious as to why going with a gain higher than 255 would be advantageous? I wouldn't gain much in read noise and I would loose dynamic range.

In brief: That little you would gain in noise reduction can matter a lot for the faint stuff, while loss of dynamic range matters for the bright stuff. It's a tradeoff. But from the moment there are no DR issues (nothing gets clipped), you can try increasing the gain. If the highlights still don't clip, it's a win. If they clip, you go back to the lower gain. I mean obviously it is not mandatory.

Why care about a tiny read noise reduction? Because SNR=signal/noise is not the same all over the picture, despite being reduced uniformly by integration. Neither is the impact of read noise. I will give an example with completely fictional and somewhat exaggerated numbers, without loss of generality.

Let's say read noise is 5 data units and ignore the thermal aspect and other types of fixed noise for simplicity.

An area with an average signal of 10,000 data units has average signal noise proportional to sqrt(10,000) = 100. An area with an average signal of 100 has average signal noise of 10.

For the bright area, the 5 electrons of read noise are not too significant because they represent less than 5% of the noise. But for the faint area, read noise is one third of the total.

SNR of the bright area is 95.2. SNR of the faint area is 6.7. Now you know why dark nebulae are such difficult targets  

Anyway, let's say you reduce the read noise to 4 data units. Bright area SNR becomes 96.15. A 0.05% improvement, practically nothing. But dark area SNR becomes 7.2. A 7% improvement, 100 times more impactful! You would need 0.2% more exposure time to improve the bright areas as much, and 50% more exposure time to improve the dark areas as much!

So 0.1 electrons less read noise can go a surprisingly long way, depending on how faint is your signal (which in turns depends on exposure time, f/ratio, sky background and of course the target itself).
Helpful Insightful Engaging
Tony Gondola avatar
Tony Gondola avatar
As with a lot if engineering questions, it depends looms large! I do get what you're saying though. When the signal is very close to the noise floor, small changes in the noise floor make a big difference.
JoepsAstronomy avatar
JoepsAstronomy avatar
I have tried at 100 gain with 5 minute exposures various objects. It worked fine for me, but didn't do a comparison yet with 252 or 0 gain. I hope to do more testing in the near future. It's a bummer though that ZWO fixed the HCG at 252, would be much nicer if it was variable like Touptek. At 252 stars do get pretty large pretty fast.
Tony Gondola avatar
Tony Gondola avatar
I've been around and around on this and I'm back to the old reliable gain of 255 and short subs. While the increase in dynamic range at lower gain is useful for some objects the need to shoot long subs to swamp the read noise really negates that advantage for me. Here's a quick render of M106 shot last night, no star removal. 6.2 hours in near IR of 15 sec subs shot at a gain of 255. This is where I seem to get the best results from this sensor.

Helpful Insightful Concise Engaging
JoepsAstronomy avatar
JoepsAstronomy avatar
Can you clarify which wavelength you mean with Near IR?
Well Written Respectful
Rick Veregin avatar
Rick Veregin avatar
Tony Gondola:
Has anyone tried shooting a 585 sensor at zero gain? I typically run my ZWO version at 255 so that HGC kicks in and drops the read noise but I do give up well depth and dynamic range. Generally this works well as I like to keep my subs short. I haven't run into any dynamic range issues at 255 so I wonder of the extra 1.5 stops of dynamic range would be helpful?

Here are the performance charts for the sensor.


Hi Tony, I don’t have the same camera, but I am in Bortle 8, so I can share my experience and calculations. 

Read noise only matters if it adds significantly to your sky background noise, because your background noise is as good as it gets, you can’t get noise lower than that.

In Bortle 8, for an RGB no filter 4 s exposure my sky background is 32 electrons. I run my camera at 1.4 e read noise. So doing the calculation for noise:
Total noise without read noise is: sqrt(32)=5.66 e, as the Poisson noise is the sqrt of the signal.
Total noise with read noise is sqrt(32 + 1.4^2) = sqrt(32 + 2)= 5.83
% noise due to read noise: 100*(5.83-5.66)/5.83= 3%, so read noise is insignificant even at 4 s.

Now if you run a gain=0, 7 e read noise, your total noise is sqrt(32 + 7^2)=9, so your % read noise would be 100*(9-5.66)/9 = 37% noise!
Running at Gain=0, and 7 e read noise if you run 10X longer subexposure, which is only 40 s, you would get: Total noise = sqrt(320+7^2)= 19.2 vs Sky noise = sqrt(320)=17.9, and the read noise is then contributing 100*(19.2-17.9)/19.2=6.8%. Not too bad, but you would probably want to run at least 1 minute to reduce the impact of read noise. 

With my narrow band filters in B8, I need 1 to 4 minutes exposure to get down to about 5% read noise contribution. The narrower the filter, the longer exposure I need. 
So at 7 e read noise, now you are talking about at least 10 to 40 minute exposures to reduce the effect of read noise. That may be more than you are willing to run, that needs excellent guiding and a very stable setup.

My own opinion is that in most cases there is too much worry about the camera dynamic range for deep sky imaging. The important thing is that your image dynamic range is using the most of the sensor dynamic range in the sub. 

So at gain=0 you have 40 K dynamic range. But if you are taking an image of a faint nebula, the nebula itself might be only using 5 K of that range. So your whole nebula image has 5 K levels to define the dynamic range of the image. Suppose you increase your gain so that 5K image now goes to 40 K. Now you have 40 K levels to define you image. Your image smoothness is going to increase dramatically, which will enable you to stretch much faithfully to the input image.

Now that setting to optimize the nebula might very well blow out the stars. The other option is to increase your subexposure, which of course risks too saturating stars. One option if your nebula is faint, is to set high gain or longer subs for the nebula. Then take a second set of images with lower gain/expsure for the stars, this doesn’t need so much exposure. Then with SW today you can separate stars and nebula and combine them both--you get the best of both worlds. In the end it is a trade off, so you will likely need to accept some level of star core saturation if you want to optimize for the nebula. And often bright stars are saturated anyway.

Finally, note at gain=0, you have nearly a 10 electrons/adu detection. That means that 0 to 9 electrons in, gives you zero  signal out! A signal of 10 to 19 e, gives you a signal of 1 out. So you are losing "resolution" of the number of electrons. On your camera I would recommend no lower than a gain of 198, which is your unity gain. That is one electron in gives 1 unit of signal out. And there is benefit to running even higher gain, again to give you more levels to capture the weakest signals. This is because while we always talk about discrete electrons, the signal out is actually a continuous voltage with noise, so having more levels to capture fractional signals can be important for weak signals.

I would not run personally less than unity gain. And I would suggest you look at your subexposure histogram to judge how well you are using the dynamic range of your sensor by using it as much as possible with your image dynamic range. If your input histogram shows no pixels at all saturated, you are probably running at shorter exposure or lower gain than you should be at. 

I hope this helps
Rick
Helpful Insightful
Tony Gondola avatar
Tony Gondola avatar
Can you clarify which wavelength you mean with Near IR?

Sure, the filter opens up at 800nm and passes down from there, at least to 1100nm as that's where the graph stops.
Well Written Concise
JoepsAstronomy avatar
JoepsAstronomy avatar
Rick Veregin:
<long story about noise and gain >

Hi Rick,

great story which I sadly don't understand completely yet. Here are two images. 

The first image is M33 with IR Cut filter at 252 gain with 60 seconds exposure.



The second image is M81/82 with IR Cut filter at 100 gain with 300 seconds exposure



Now based on your story in using the most of the range of your sensor, which one is better?

Based on the histogram for Image 1 could I have used also 2 minute exposures?

You determine the skyglow with a 4 seconds exposure. Is there a reason for this? What is this number based on?

You speak of levels for using your image (from 5K to 40K), but how can you do this if the FW of the camera at 252 gain is only 4K and it's a 12Bit camera?
Engaging
Tony Gondola avatar
Tony Gondola avatar
Rick Veregin:
Tony Gondola:
Has anyone tried shooting a 585 sensor at zero gain? I typically run my ZWO version at 255 so that HGC kicks in and drops the read noise but I do give up well depth and dynamic range. Generally this works well as I like to keep my subs short. I haven't run into any dynamic range issues at 255 so I wonder of the extra 1.5 stops of dynamic range would be helpful?

Here are the performance charts for the sensor.


Hi Tony, I don’t have the same camera, but I am in Bortle 8, so I can share my experience and calculations. 

Read noise only matters if it adds significantly to your sky background noise, because your background noise is as good as it gets, you can’t get noise lower than that.

In Bortle 8, for an RGB no filter 4 s exposure my sky background is 32 electrons. I run my camera at 1.4 e read noise. So doing the calculation for noise:
Total noise without read noise is: sqrt(32)=5.66 e, as the Poisson noise is the sqrt of the signal.
Total noise with read noise is sqrt(32 + 1.4^2) = sqrt(32 + 2)= 5.83
% noise due to read noise: 100*(5.83-5.66)/5.83= 3%, so read noise is insignificant even at 4 s.

Now if you run a gain=0, 7 e read noise, your total noise is sqrt(32 + 7^2)=9, so your % read noise would be 100*(9-5.66)/9 = 37% noise!
Running at Gain=0, and 7 e read noise if you run 10X longer subexposure, which is only 40 s, you would get: Total noise = sqrt(320+7^2)= 19.2 vs Sky noise = sqrt(320)=17.9, and the read noise is then contributing 100*(19.2-17.9)/19.2=6.8%. Not too bad, but you would probably want to run at least 1 minute to reduce the impact of read noise. 

With my narrow band filters in B8, I need 1 to 4 minutes exposure to get down to about 5% read noise contribution. The narrower the filter, the longer exposure I need. 
So at 7 e read noise, now you are talking about at least 10 to 40 minute exposures to reduce the effect of read noise. That may be more than you are willing to run, that needs excellent guiding and a very stable setup.

My own opinion is that in most cases there is too much worry about the camera dynamic range for deep sky imaging. The important thing is that your image dynamic range is using the most of the sensor dynamic range in the sub. 

So at gain=0 you have 40 K dynamic range. But if you are taking an image of a faint nebula, the nebula itself might be only using 5 K of that range. So your whole nebula image has 5 K levels to define the dynamic range of the image. Suppose you increase your gain so that 5K image now goes to 40 K. Now you have 40 K levels to define you image. Your image smoothness is going to increase dramatically, which will enable you to stretch much faithfully to the input image.

Now that setting to optimize the nebula might very well blow out the stars. The other option is to increase your subexposure, which of course risks too saturating stars. One option if your nebula is faint, is to set high gain or longer subs for the nebula. Then take a second set of images with lower gain/expsure for the stars, this doesn’t need so much exposure. Then with SW today you can separate stars and nebula and combine them both--you get the best of both worlds. In the end it is a trade off, so you will likely need to accept some level of star core saturation if you want to optimize for the nebula. And often bright stars are saturated anyway.

Finally, note at gain=0, you have nearly a 10 electrons/adu detection. That means that 0 to 9 electrons in, gives you zero  signal out! A signal of 10 to 19 e, gives you a signal of 1 out. So you are losing "resolution" of the number of electrons. On your camera I would recommend no lower than a gain of 198, which is your unity gain. That is one electron in gives 1 unit of signal out. And there is benefit to running even higher gain, again to give you more levels to capture the weakest signals. This is because while we always talk about discrete electrons, the signal out is actually a continuous voltage with noise, so having more levels to capture fractional signals can be important for weak signals.

I would not run personally less than unity gain. And I would suggest you look at your subexposure histogram to judge how well you are using the dynamic range of your sensor by using it as much as possible with your image dynamic range. If your input histogram shows no pixels at all saturated, you are probably running at shorter exposure or lower gain than you should be at. 

I hope this helps
Rick

It does and pretty much agrees with what I've seen from this sensor. One aspect of this that's a little different is that I'm trying to optimize for resolution over going super deep. I use such short subs because while, they might not be optimal for S/N reasons they are very useful to help improve the seeing. If you shoot say 180 sec. subs and watch the HFR graph, in average seeing it will be pretty flat with not a lot of difference between sub exposures. The seeing over that time is being averaged out. However, at 15 sec. you get a lot of variation in HFR from sub to sub. Last night when M106 was imaged, I was seeing a variation range of 2.8 to over 4 HFR, nearly a factor of two and that's something you can take advantage of. The above stack was using all the data. When I process out this image in a serious way I'll look at the HFR graph and decide where the cutoff will be. Yes, I'll loose integration time but will gain resolution, my 6.2 hours might end up being 5 hours but the result will be improved.

It's just using the precepts of lucky imaging and applying that to what we do. It's not a full on application of the technique because that would require 0.1 sec subs but it still does help. Depending on how that looks I might run another 6 hours just to get more subs with low HFR values. Along with all the usual rules it overlays a different set of constraints.
Helpful Insightful Engaging
Rick Veregin avatar
Rick Veregin avatar
Rick Veregin:
<long story about noise and gain >

Hi Rick,

great story which I sadly don't understand completely yet. Here are two images. 

The first image is M33 with IR Cut filter at 252 gain with 60 seconds exposure.



The second image is M81/82 with IR Cut filter at 100 gain with 300 seconds exposure



Now based on your story in using the most of the range of your sensor, which one is better?

Based on the histogram for Image 1 could I have used also 2 minute exposures?

You determine the skyglow with a 4 seconds exposure. Is there a reason for this? What is this number based on?

You speak of levels for using your image (from 5K to 40K), but how can you do this if the FW of the camera at 252 gain is only 4K and it's a 12Bit camera?

 

Hi Tony
Looking at your 252 gain image at 60s, you are using the histogram really well, and only saturating a few pixels. I use SharpCap and it shows a histogram with the number of saturated pixels, so one can make a judgement call. My judgement is there will only be a few pixels of bright stars saturated, so this is fine. At this gain you are at 0.5 e/adu, which is great, you are above unity gain. And given you are in Bortle 8 (like me)  with just an IR filter, this exposure is more than enough to make read noise be insignificant vs. sky noise. I would not go to longer at this gain for this target. Doubling to 2 minutes means that anything > about 38000 adu will be >2x38000 is > 66,000, so saturated. Similarly if you half your exposure, you will have virtually no signal >38000, except perhaps some of those pixels that were in saturated cores. So you will not be using all of your dynamic range.

 

Looking at your 100 gain image at 300 s, you have very little in the upper half of the histogram, >3800 is pretty much empty of data, and you still have some saturated cores. So the histogram here is not looking quite as good. Your read noise here is around 4.5 e, but again based on my experience in Bortle 8, you are not getting much contribution of read noise at 300 s.  That is great. Note here you are at 3 electrons/adu, so you are loosing some electron resolution, you can’t tell 0 from 2 electrons, or 3 from 5 electrons, etc, in your output. Now unfortunately, while increasing the gain from 100 to 150 will increase your signal 2X or 1 stop you will also lose dynamic range, so likely saturate too much. And doubling the gain will not get you to unity gain. You could increase exposure, but that will not obviously get you to unity gain either, and 600 s is getting a bit longer than you might want? I would increase gain to get to unity gain, this gives you two stops or 4 X bigger signal. So you could drop your exposure to 150s, bringing it down 2X in signal. So now your histogram will be better filled. The advantage of going from 100 to 198 gain is that there is virtually no loss in camera dynamic range.

I know this is all quite complicated, there is a lot to think about.

General rules of thumb:

Unity gain is a good place to be, but if you are not filling the histogram you could increase gain, but carefully as you now will lose dynamic range with higher gain. But if you are far away from filling the histogram, higher gain can be good. Of you could increase subexposure, but in Bortle 8, without NB filters, 60 sec is going to be a long enough sub to subdue the read noise, going longer is not helpful, particularly if you have to drop below unity gain. 

 

Regarding your other questions

 

You determine the skyglow with a 4 seconds exposure. Is there a reason for this? What is this number based on?

Actually I took a longer exposure, measured the background signal, then with the calculations as I showed, calculated that 4 s was sufficient that read noise with my settings would be insignificant compared to sky noise. This means anything more than a 4 sec sub is not helping the image quality, though obviously at 4 s that would be a lot of subs!!

You could easily do you own background calculation. I point my telescope overhead as that is the darkest part of the sky and I do it in the middle of the night when it is darkest. So this is the lowest background I can get. I take any arbitrary length exposure at the gain I want to use. I use DeepSkyStacker to stack, it shows the background % signal for each image when you register the image. You need to run a bias or dark frame as well, so that is subtracted in your stacking software (note I'm not actually stacking, just registering one sub). So say DSS says 1% background. My full well for that gain is say 10,000 electrons. Then my background is 10000 x 1% = 100 electrons and my background noise is sqrt(100) = 10 e. Your total noise is sqrt (10^2 + Readnoise^2). I calculate the noise result and compare to the value without read noise. Your background signal is proportional to exposure, so if I find I need to double my background signal to get to a good spot, I know I just need to increase my subexposure. So in this way you can calculate what exposure you need so Read noise doesn’t contribute significantly. This is what I did.

 

You speak of levels for using your image (from 5K to 40K), but how can you do this if the FW of the camera at 252 gain is only 4K and it's a 12Bit camera?
 

I was referring to a case where you would have the 40 K FW. And unfortunately, I fell into a trap that is easy to fall into, since I was thinking of my 16-bit camera, where the FW in electrons and levels is the same. My apologies—let me correct this below.

At zero gain, your camera has 40 K electrons full-well, not 40 K levels. As a 12-bit camera it only has 4096 levels. So 40,000 e/4096 levels = 9.77 e/adu, which is what your chart shows. So I could have said you want to fill the full-well up to 40 K electrons, or you want to fill as 4096 adu units. But this is where it gets complicated, because for a 16-bit file output, we have a problem because we only have 12-bits of output. The actual histogram should go from 0 to 4095, not to 64K. It depends both on the camera and what software you use how this problem is dealt with for cameras with less than 16 bits of data. Sometimes you will get the real histogram, other times the data is adjusted in some way, as your histogram clearly is.  So one cannot use the histogram in this case to do any calculations, unless you know what the camera and your software is doing to the data to fill the 64k histogram. All you histogram can show you is if you are using the full range of possible levels, and if you are saturating.

Unfortunately, the more you dig into this, the more complex it gets. But I hope that the general rules and approach will enable you to optimize your imaging. And by the way, none of the histograms show a real problem, they are fine--though in my opinion going to less than unity gain is not the best. If you really need that wide a dynamic range just decrease your exposure a bit.
Rick
Helpful Insightful