Modern CMOS, multiple sub length?

[DaveS]

I know it has been SOP with wide DR targets like M42 and M31 to take multiple sub lengths and use layer masking to hold back the bright cores while bringing out the faint outer details.

However, looking at some of the newer QHY CMOS cameras with 14+ stops DR and huge Full Well Depths I wonder if it's still necessary to capture multiple sub lengths to avoid blowing out highlight detail?

I know DDP can do wonders with pulling out faint detail without blowing out bright areas, perhaps combine this with a masked stretch for the very faint stuff?

[vlaiv]

You can build very high dynamic range image with even 8bit camera - you don't need large DR in single exposure - that is what the stacking is for.

It is feasible way of doing things with modern CMOS cameras because they have low read noise and therefore can utilize short exposures in LP areas.

Otherwise you loose more than you gain by not using multiple exposure lengths.

Whenever you stack certain number of frames - you extend dynamic range by log base 2. Stack two frames - your data range went up by one bit. Stack 16 subs - it went up by 4 bits. Stack 256 subs - it went up by 8 bits and so on.

If you want to capture very big dynamic range - trick is to stack many short exposures. Each short exposure won't saturate sensor and enough of them will make final image very high dynamic.

Problem with this approach is read noise - which is per exposure - you inject a lot of it. It has to be really low and smaller than some other noise source (usually LP noise is a good candidate) in order not to matter much.

 

[DaveS]

OK.

My sky is SQI 21.66, so not dark dark, but usefully dark even so.

Will have to think about how short is short. Currently I use 10 min subs.

[vlaiv]

2 minutes ago, DaveS said:

OK.

My sky is SQI 21.66, so not dark dark, but usefully dark even so.

Will have to think about how short is short. Currently I use 10 min subs.

What is your background value in electrons from that sky?

Although it is very dark, if you are shooting wide field / low resolution - large part of the sky will be covered by single pixel and pixel values from the sky will still be higher. High resolution work on the other hand spread sky background over more pixels thus lowering each value.

[DaveS]

10 minutes ago, vlaiv said:

What is your background value in electrons from that sky?

Although it is very dark, if you are shooting wide field / low resolution - large part of the sky will be covered by single pixel and pixel values from the sky will still be higher. High resolution work on the other hand spread sky background over more pixels thus lowering each value.

TBH, I don't know the answer to that. I am / will be imaging at 910mm FL and around the 0.85"pp level.

It will be a while before I can afford to buy one of the new CMOS cameras so plenty of time to work out how I will run it.

[Ken82]

I’ve always taken similar sub lengths with my cmos cameras. I haven’t found any reason to take multiple Exposures so far. 
 

Using HDR in pixinsight brings out better detail in over exposed areas. I’m sure PS or others have similar ?? 
 

[ollypenrice]

Vlaiv's is an informative post regarding dynamic range. Most of the people I know using modern CMOS cameras are using fine minute subs. It's easy to see if you're saturated the moment you look at the linear stack and, if you suspect an area might be saturated, you just mouse over to read its values. My guess is that multiple sub lengths will be as rarely needed as with CCD imaging.

SQM 21.6 is dark!  We occasionally hit 22 here but it's usually 21.6 or so. Guests who've been to Nambia say the zenith is similar there to here but that the sky is dark to lower elevations, which is to be expected.

Olly

[DaveS]

Thanks guys

I'll have a look at a linear stack of my data, probably a Luminance stack, and I'll do one using Sigma average rather than add. ATM I'm only using a fairly rubbish ASI 1600 with 12 bit ADC and miserable FWD.

Thinking of one of the upcoming QHY 268 as that will avoid the expense of 50 mm unmounted filters, though a full-frame OSC has a certain siren attraction.

[Seelive]

So far, I've only used multiple exposures on M42 in order to avoid saturating the core and then I used the DSS Entropy Weighted Average (High Dynamic Range) stacking method to produce the stacked image (although targets like M31 and M94 could well benefit).  This is an enlargement of the core using normal Average stacking of just the 180s exposures, the brighter stars and core are saturated.

This is the result using the Entropy Weighted Average (High Dynamic Range) stacking method with 180S, 60s, 20s and 7s exposures, the Trapezium stars are now just visible:

The shorter exposures on their own did not bring out the fainter nebulosity captured in the 180s exposures.

[ollypenrice]

4 hours ago, DaveS said:

Thanks guys

I'll have a look at a linear stack of my data, probably a Luminance stack, and I'll do one using Sigma average rather than add. ATM I'm only using a fairly rubbish ASI 1600 with 12 bit ADC and miserable FWD.

Thinking of one of the upcoming QHY 268 as that will avoid the expense of 50 mm unmounted filters, though a full-frame OSC has a certain siren attraction.

Dave, remember that, if you have an RGB image as well, this may give you all the 'short' exposures you need. I quite often use the RGB-only as a less saturated stack to cover regions saturated in the luminance layer. I say 'quite often,' by which I mean, 'Quite often in the rare cases where I need less exposure.' You can extract a synthteic luminance from the RGB if you like but it will work in the same way as a luminance layer in its natural form.

When layer masking three exposure lengths for M42 I don't shoot the shortest in luminance at all. Why bother? If you're trying to reduce signal there is no point in capturing luminance in order to capture more!

Olly

 

[vlaiv]

4 hours ago, DaveS said:

Thanks guys

I'll have a look at a linear stack of my data, probably a Luminance stack, and I'll do one using Sigma average rather than add. ATM I'm only using a fairly rubbish ASI 1600 with 12 bit ADC and miserable FWD.

Thinking of one of the upcoming QHY 268 as that will avoid the expense of 50 mm unmounted filters, though a full-frame OSC has a certain siren attraction.

Why do you feel your ASI1600 is rubbish?

[DaveS]

1 hour ago, ollypenrice said:

Dave, remember that, if you have an RGB image as well, this may give you all the 'short' exposures you need. I quite often use the RGB-only as a less saturated stack to cover regions saturated in the luminance layer. I say 'quite often,' by which I mean, 'Quite often in the rare cases where I need less exposure.' You can extract a synthteic luminance from the RGB if you like but it will work in the same way as a luminance layer in its natural form.

When layer masking three exposure lengths for M42 I don't shoot the shortest in luminance at all. Why bother? If you're trying to reduce signal there is no point in capturing luminance in order to capture more!

Olly

 

Thanks Olly

I'll investigate further, but I've just rebooted the computer. Came in from working on the obsy to find we had a power cut. It's back on now but I've been running around making sure everything's back to normal.

[DaveS]

1 hour ago, vlaiv said:

Why do you feel your ASI1600 is rubbish?

Well, perhaps not rubbish, but not up to the latest CMOS cameras.

12 bit ADC and amp-glow are the biggest negatives. Plus I just had to replace a dead fan.

Could do with more Full Well Depth too.

[vlaiv]

Just now, DaveS said:

Well, perhaps not rubbish, but not up to the latest CMOS cameras.

12 bit ADC and amp-glow are the biggest negatives. Plus I just had to replace a dead fan.

Could do with more Full Well Depth too.

Your amp glow does not calibrate out? Mine does without any issues.

Whole ADC / FW thing and dynamic range is thing of perspective.

What if I told you that ASI1600 is camera that has 64K full well capacity, 16bit ADC and 6.8e read noise?

Suddenly it does not look nearly "as bad" anymore, right?

But how to turn ASI1600 in such camera? Let's say you want to do 15 minute subs with such camera, what should you do? Well, take 16 frames 56.25s long (unity gain) and stack them (sum stack) and result will be the same as if you used 64K FW, 16bit ADC, 6.8e read noise camera and took single 15 minute sub. If you dither, there will be added bonus with calibration and noise "smoothness".

Here is very simple explanation for what is going on:

Imagine you have number in range 0-3. You need two bits of information to store that value.

Now look what happens when you add two such numbers: if you add 0 and 0 you get 0, but if you add 3 and 3 you get 6. Lowest number that you can get as a result is 0 while highest is 6. Result is therefore in range of 0-6. That is 3 bits of data (0-7).

Now, if we add 4 such numbers, result will be in range of 0-12, which is 4 bit value (0-15).

If you add 256 such numbers, result will be in range of 0 - 768 - which is 10 bit value (0-1023).

And so on ....

(there is small under utilization of the full bit range that depends on how many values you add).

Noise adds in quadrature, so read noise of 1.7 added 16 times will be sqrt(1.7^2 * 16) = 1.7 * sqrt(16) = 4 * 1.7 = 6.8e.

 

[DaveS]

Thanks, yes the amp-glow calibrates out, but I wish it wasn't there to begin with.

Reading the rest of your post suggests the "hundreds of short subs" regime, with 16 subs for every single sub (Usually 600 sec) that I currently take.

I currently dither between subs, and use Sigma Add stacking, though with only in the low tens of subs made at the Max DR setting.

Will give this more thought, though the QHY 268 (Hopefully colour in Oct) is looking very attractive.

 

[vlaiv]

6 minutes ago, DaveS said:

Will give this more thought, though the QHY 268 (Hopefully colour in Oct) is looking very attractive.

Indeed, I wish for larger sensor as well since ASI1600 is only 4/3 format.