Narrowband Exposure Lengths vs. Swamping Read Noise ... help!

I just shoot 180 sec exposures for all my subs - RGB, luminance and narrowband. Seems to work for me.

John Stone:
Can someone help me out here a little. It's my understanding that you want  to square your read-noise at the gain you're shooting at and then multiply it by a swamp factor then convert that to ADU units and then add that to the median ADU value of your bias frame to get the desired median value of your light frames. 

Is that right?

For example for my IMX533 sensor at gain 100 offset 70 I lookup a read noise of 1.5e- and lets say I want a swamp factor of 5x:  (1.5 * 1.5) * 5 = 11.25e- of "sky glow" in my subs to swamp the read noise. For an offset of 70 ADU and gain of 100 (1e-/ADU) that would mean a "sky glow" of floor(70 + 11.25) = 81 ADU, right?

So I'm shooting F7 with my 8" SCT reduced to 1450mm on an IMX533 through Antlia 3nm narrowband filters and in 300 sec subs I'm getting these median ADU values:

Ha: 73
Oiii: 76
Sii: 73 

That's WAY less than 81! 

So, okay, I take 900sec subs and get a median ADU of:

Ha: 77 

So 7 ADU units above the bias median work out to be a swamp factor of:  7 / (1.5 * 1.5) = 3.11x

It looks like I'd need to take something like 1200sec subs just to get to about 5x "sky glow" above the read noise. 

That's crazy, I'd need to buy a $12k mount to track well enough for those subs lengths.

Is there something wrong with my camera or filters or something?

This seems like a CRAZY long exposure time for B7 skies, (measured 18.5 mag/arcsec2).

Is this comparable to what other people are seeing?

😴😴😴 don’t bother with all that, just enjoy taking pictures, 1-2 min subs for broadband (depending on target brightness) no filters, and 3 -5 mins for Narrowband with 3nm filters (again target brightness dependant) works fine for me and noise is barely noticeable…

In this presentation by Dr. Robin Glover (Sharpcap), https://www.youtube.com/watch?v=3RH93UvP358, at about the 50min mark he shows an exposure length vs f-ratio chart for different Bortle values at a given read noise. The shown exposures lengths are for luminance, but it's noted that exposure for RGB filters is x3 and NB, x25.

Cheers,
Scott

To get a rough ballpark figure I use a calculator for that: https://smallstarspot.com/shiny/StackSNR/

There you can calculate how deep you can go for a given time, additionally to that it calculates the swamp factor for you.

Tobiasz:
To get a rough ballpark figure I use a calculator for that: https://smallstarspot.com/shiny/StackSNR/

There you can calculate how deep you can go for a given time, additionally to that it calculates the swamp factor for you.

Can you explain the chart?  Where is the swamp factor?

John Stone:
Can someone help me out here a little. It's my understanding that you want  to square your read-noise at the gain you're shooting at and then multiply it by a swamp factor then convert that to ADU units and then add that to the median ADU value of your bias frame to get the desired median value of your light frames. 

Is that right?

For example for my IMX533 sensor at gain 100 offset 70 I lookup a read noise of 1.5e- and lets say I want a swamp factor of 5x:  (1.5 * 1.5) * 5 = 11.25e- of "sky glow" in my subs to swamp the read noise. For an offset of 70 ADU and gain of 100 (1e-/ADU) that would mean a "sky glow" of floor(70 + 11.25) = 81 ADU, right?

So I'm shooting F7 with my 8" SCT reduced to 1450mm on an IMX533 through Antlia 3nm narrowband filters and in 300 sec subs I'm getting these median ADU values:

Ha: 73
Oiii: 76
Sii: 73 

That's WAY less than 81! 

So, okay, I take 900sec subs and get a median ADU of:

Ha: 77 

So 7 ADU units above the bias median work out to be a swamp factor of:  7 / (1.5 * 1.5) = 3.11x

It looks like I'd need to take something like 1200sec subs just to get to about 5x "sky glow" above the read noise. 

That's crazy, I'd need to buy a $12k mount to track well enough for those subs lengths.

Is there something wrong with my camera or filters or something?

This seems like a CRAZY long exposure time for B7 skies, (measured 18.5 mag/arcsec2).

Is this comparable to what other people are seeing?

I haven't checked your math but note the 533 is 14 bit, so check your ADU conversion.

Additionally, you will not be able to fully swamp read noise with 3 nm narrowband filters at f7, even with the wide aperture . That's how NB filters work   For very dark sites, or for smaller aperture/ bigger focal ratio, you might not be able to shoot long enough to meet the calculated exposure requirement. So...

instead, you need to shoot as long as you reasonably can. I shoot 6 min exposures in B5 conditions, 4.5nm filters, sky brightness 19.5, 90mm APO at f4.6, i still need a lot of subs for oiii and sii, because signal is down in the noise floor. That's why we stack ;)

There's a spreadsheet online somewhere that calculates minimum exposure for the 533 sensor, scope size and various sky brightness/ filters, by Steven Bellavia(?)

Putting your numbers in, I get an exposure around 22 minutes for a 3nm filter at f7 on a 200mm aperture at 100 gain, with a swamp factor = 10 skyflux/RN2 (=3.14 in other units). So yes, you need a long exposure to meet the swamp factor. You can check the median ADU of a test sub to see if you are in the right ball park. For my 533mc, using ASI Air, 70 offset is around 2800 ADU (converted to 16 bit in the Asi Air app)  so I go for a target ADU of around 3000 to swamp read noise, if achievable 

https://www.cloudynights.com/topic/918793-help-with-choosing-sub-exposure-length-zwo-asi-533-and-asiair/

Thanks for the interest in my SNR tool.  I think it can help with this thread so I'll go through what I think is the situation for the OP's conditions - but I'm not sure of some of the numbers.  There appears to be a related thread in CN on the same topic - so I'll use what I think are the best numbers.

The situation is f/7 with 3.76um pixels, 3.12e read noise (due to gain setting being a bit low) and a sky background flux, combined with dark current, resulting in 14e in 900s, or 0.016 e/s total background signal.

I'm not a fan of defining thresholds or "optimal" settings for something like subexposure time because, in terms of the noise model, longer is always better and shorter is worse.  The real question is how much difference going from 300s to 600 or 900 might actually make.  And the way I characterize the difference is in terms of how much deeper you can stretch the image - in terms of magnitude per arc-sec squared - to achieve a given minimum SNR target.

The minimum SNR target is entirely up to the user in terms of how much they want to stretch and reveal background noise.  For technical measurements a min SNR of perhaps 10 is common - but when imaging faint structure you can see details with SNR as low as 1.  So I use a target of 3 for these assessments.

First a check on the sky background value.  I image with 3nm filters from light pollution of around 18.5 mag/arc-sec^ and at f/10 with 3.78um pixels my background is around 0.028 e/s.  If I scale this by the f/ratio factor of (10/7)^2 I get 0.056 e/s if I were at f/7.  The OP's value of 0.016 is about 3 times smaller, or about 1.5 mags darker - corresponding to a dark sky of around 20 mag/arc-sec^2.  This is all just rough estimates but as long as the OP's sky is very dark at around that level, it serves as a sanity check on what looks like a fairly small sky background signal.

Assuming 10 hour total imaging time and dead time between exposures of 5s, the resulting magnitude limit for SNR=3 improves with subexposure time according to the following plot:




The plot is all relative to the physical limit of a perfect imaging system with no loss and limited only by shot noise, operating at f/7 with 3.76um pixels.  That is the very best you can do, and the vertical axis represents loss relative to that limit.  At 1200s exposures you are losing about 3.2 mag, while at 600 you are losing about 3.4.  So you can concretely say that going from 600 to 1200 will gain you 0.2 mag - if you can do it.  So this isn't showing things as an "increase in SNR" - it's showing an increase in limiting magnitude - which is more meaningful.

The different lines in the plot show different terms being added in the noise model - and the big jump from the upper bluish line to the final orange line is due entirely to read noise - so you can see the big role it plays.  If the gain were turned up a bit so the read noise dropped to about 1.4 it would help a lot.  But this gives an idea of just how much benefit you get from longer exposure.  There is no point where suddenly you are doing about as well as you can.  And it also serves as a sanity check on the sky background flux.

Frank

John Stone:
Frank,

Thanks for your reply.   

In trying to understand all this I've noticed a discrepancy in how different people actually calculate the  "swamp" number.

The formula is:  <scale-factor>*RN^2/<sky-flux>

The discrepancy is in the order operations.   I've seen a lot of people calculate like: 

((<scale-factor>*Rn)^2)/<sky-flux>

and others like you calculate like:

(<scale-factor>*(Rn^2))/<sly-flux>

I'm not sure which one is correct.

Can you comment?

Hi - 

My main point is not to rely on such factors because there is no clear "knee" in the plot.  The expression I used is what I thought was the standard convention - but I won't vouch for any of them.

If you want the "swamp factor" to refer to the ratio of noise terms, then it would be sqrt(background)/rn.  If you want the background noise to be X times the read noise - that is the expression to use.

But 10x the read noise is huge and in terms of the total noise, the read noise is a negligible contribution.  That's why sqrt(10) as a factor is more reasonable - hence the choice I made to leave it outside the ^2.

If you had no read noise the total noise would be sqrt(b) where b is the background.  If the read noise is 1/sqrt(10) times the background noise, then the total noise is sqrt(b + b/10) or sqrt(b)*sqrt(1.1) = 1.05 sqrt(b).  Hence the read noise is only adding 5% - which is fairly negligible.

But there is nothing magic about 10 or sqrt(10) or 5%.  You can image well with read noise even equaling the background noise as long as you expose enough.  So I would just look at the sky background relative to the read noise and say - Oh - I should use at least 5m exposures for this.  Whereas for RGB imaging you might say - the background is huge so I don't need long exposures - but I might as well use 5m anyway.

Ideally the exposures would be long enough that read noise is at least less than sky background noise.  If the two things are equal, read noise is increasing the noise by 40%.  If read noise is about 3x smaller than background noise, it is raising it by 5%.  You can image anywhere in that range and beyond as long as you expose enough.

For your example the read noise is high compared to sky background unless you use very long exposures - but fortunately you are able to change gain so the read noise is much lower.  If the resulting "swamp time" is still too long to be practical - at least strive for read noise to be less than background noise.

Another way to look at this is to view it as a "good problem to have."  It's hard to go long enough to swamp the read noise - but that's because you have such low background noise in the first place.  The 3nm is doing its job - and you are in great shape.

Frank

John Stone:
Frank,

Thanks for the clarification.  I'll continue to use your approach when running my numbers.

I am a little worried that something's off since I own a SQM meter that reads 18.5 mag arcsec-2 and as you say my numbers don't support that sky; there is 3x less flux than expected.

Next time I get clear skies I'll go and check my other 3nm Antlia filters;  maybe I got an defective Ha filter.

J.

Oh!  In that case I suspect there is a 14/16 bit issue here and your actual background rate is 4x larger (2 bits).  Either that or the gain is wrong by a similar factor.  I did want to confirm you really have dark skies and apparently you don't.

Maybe just take new data with the gain set higher so the read noise is lower.  If you can create two bias, two flat, one dark at 900s and one 3nm Ha sky exposure at 900s that would allow everything to be calculated with certainty.

My ASI camera is internally 14-bit but they shift the bits left by 2.  But I just view it as a 16-bit camera and I measure the gain in terms of that 16-bit adu range.  So my gain is a small value at 0.095 e/adu - but there is nothing wrong with that.  The fact that the "real" (14-bit) gain is 4x higher is irrelevant since the actual adu's in the image are 16-bit.  This way there is no confusion about 14/16 bit.

So for you the "good" news is you don't need to worry about super long exposures.  The bad news is that you are stuck with higher noise from the sky that you can't do much about.  At the same time, you probably still would want at least 5m exposures and longer may be worthwhile - once you know the numbers.

Frank

Its worth noting a couple of things.

1 - If your mount is capable of tracking accurately for 10x180s subs without guiding errors ruining any of the subs, then your mount is good enough to do 1800s subs. You don't need crazy expensive gear to shoot 20min ~ 30min subs. I used to run 1200~1800s subs routinely on a belt modded HEQ5 guided with a modified philips webcam and a 50mm finderscope, back when CCD's had 9e read noise, and it was FAR better to shoot 3x20min subs than 60x1min subs.

2 - All the math and theoretical minimums etc are fantastic, and yes, there are optimal settings for all filter/sky condition/camera configurations, but its more important to not overthink it and weigh your head down unnecessarily with all these recommendations.


With my IMX294 camera, I run 180s subs with UV/IR filter in bortle 5 skies, and I run 180s subs with a 6nm Ha/OIII filter in the same sky conditions. Both with the same gain and offset settings… Is it optimal? probably not. Are my results good enough for me? yep. I have run 600s Ha/OIII subs, and I do think that in SOME WAYS it was better, but in others it was worse… 

With my 1600MM, I run 600s narrowband subs, 120s L subs and 180s RGB subs, largely because the well depth of the camera saturates stars quickly.