Need to check my sub exposure length

[ianaiken]

Hey all,

I hope someone can provide some advice. With the current clear sky, I'm doing some imaging of IC1396. Guided 10 min subs at ISO800 with modded Canon 40D on a SW 100ED DS-Pro at ~f7.65

Can I get a longer sub, is this too long because of light pollution? What are your thoughts?

Linky: http://www.ianaiken.co.uk/sgl/IC1396_LIGHT_600s_800iso.JPG

Many thanks for looking and providing any advice.

[rfdesigner]

once you've swamped the readout noise with dark current / sky glow.. then stop.

Canon 40Ds have roughly 7e readout noise, 0.13e/s dark current @~20C

so dark current noise = readout noise at roughly 6 minutes. By 10 minutes readout noise is nicely smothered.. especially if you have sky glow.

Also I think ISO400 has similar noise performance at ISO400, but more headroom.

take a peak at: EOS40D / EOS50D comparison

clear skys

Derek

[ianaiken]

Derek,

I need to do some more reading to fully understand this, but at 20 C your saying 6 mins is the max exposure length?

Temp current reading +15 C on CMOS sensor so I guess this alters it to maybe 8 mins at a rough guess.

I will be taking darks, which I assume compensates for the readout noise you refer too, or is it no good if your readout noise is nicely smothered?

[rfdesigner]

Not quite.

At 6 minutes with a 22C sensor the dark current noise equals the readout noise. By 10 minutes the dark current noise will be somewhat more.

Have a look at the excel file I've bashed together to illustrate.

Remember to work out noise:

frame noise = sqrt(readout_nosie^2 + dark_current_rate*time)

To this you will end up having to add skyglow, with a lot of that then it means the readout noise is even more swamped and you can afford even shorter subs.

clear skys

Derek

EDIT: sorry, I mislabelled my units in the dark current column which should be (e) not (e/s) as the column takes account of the sub lenght.. tiny detail but if you're trying to get your head around this it could throw you off.

Total noise vs sub length.xls

[rfdesigner]

Derek,

Temp current reading +15 C on CMOS sensor so I guess this alters it to maybe 8 mins at a rough guess.

Silicon noise reduces by half every roughly 6~8C temperature fall... if you're running at half dark current then you can look at double frame lengths (assuming you're not skyglow swamped)

I will be taking darks, which I assume compensates for the readout noise you refer too, or is it no good if your readout noise is nicely smothered?

No.

Readout noise is the noise in the output amplifiers that read the CCD/CMOS sensor. Thus it is the same (more or less) for every pixel and incidentally changes little with temperature.

Dark current is leakage current into each pixel on the sensor, and does not build up smoothly, the difference between your dark frame and the current that's actually built up is what we're calling noise here. Look up "shot noise": Shot noise - Wikipedia, the free encyclopedia

Derek

[ianaiken]

I've updated my local copy of the sheet. Thanks for this, I am trying to get my head round this. Wish you were sitting next to me, but I think I am getting there.

So dark current noise impacts my SNR, and cannot be removed by using darks. Why does dark current noise need to be higher than read out noise? I understand that my sky glow needs to be higher than read noise so my faint target is not lost in the read noise, but where does dark noise fit into this?

Appreciate your time and patience with me :-)

Edit: 22:40 is probably not the best time to try to understand this, but it's interesting and has me hooked on getting my head around this.

[rfdesigner]

Dark current noise is REDUCED by using darks, but there is a limit:

the uncertainty in the dark current = the square root of the total dark current... this is the noise.

The difference between your dark current in your 'light' frame and your 'dark' frame SHOULD equal the squre root of the total dark current... on average, over your entire frame.

i.e. dark current = 0.1e/s so after 1000s dark current = 100e The uncertainty or nosie in that is sqrt(100) or 10e

Remember that the dark current in each pixel is unique (the average may be 0.13e/s but each pixel will be different), which is why it is advantageous to remove a dark frame, and not just take away a fixed offset from the image.

Derek

[dph1nm]

Why does dark current noise need to be higher than read out noise?

It doesn't have to be, It is just a more efficient use of your imaging time if it is.

If it isn't, your final stacked image will end up with more noise in it than it could have had for that total exposure time (if you see what I mean!).

NigelM

[ianaiken]

I think that relates to the total noise then in the spreadsheet. I think I need to work out my sky glow also. Not tonight though, I'm too tired from last night's stopping up late and up and down to attend my baby.

Thanks for all the input

[rfdesigner]

If you want skyglow please have a copy of this:

It took me a couple of days to put together but it seems to have some accuracy about it, use is as is, build you own version and see how it compares... I'm sorry the maths is a bit long winded and hard to follow, but the estimate of the skyglow is not empirical, it works from physical calculations of the number of photons incident on our atmosphere so that part at least should be accurate.

I've got three outputs:

1. limiting magnitude / pixel : just looks at a single pixel.

2. limiting magnitude / star : tries to account for the fact that a single star does not fall on a single pixel.

3. limiting magnitude of Emission Nebula : takes account of the fact that it is both narrow band and is a disperse object.

There may be errors I think I've got rid of most of them, and I've cross checked it against what data I have and it seem robust.. but that checking isn't by any means exhaustive.

Derek

photon counting.xls

[ollypenrice]

What about some simple cooling for your camera, even just physio gel packs around it? Some of my DSLR using guests do this.

Olly

[ianaiken]

Olly - I'm making my third peltier cooling box :-) just need a new sheet of metal. That should significantly reduce noise and gives me something to do while its cloudy plus this version should allow me to control temp a bit.

[ianaiken]

@Derek

Canon 40Ds have roughly 7e readout noise, 0.13e/s dark current @~20C

Since I'm shooting at ISO800 I think the readout noise is roughly 5 rather than 7. Is that right (I'm using the link you provided).

Let's take Sky Glow out of the equation for a minute and pretend it does not exist. A couple of questions which I think may help with my understanding.

1. A 60 min sub would be the ideal in the excel since total noise is the lowest for 60 mins worth of data?

2. 3 min subs would not be advantageous since dark current noise is below readout noise (I adjusted my sheet to have a readout of 5 instead of 7)?

3. A 60 min sub would not allow stacking, therefore it is advantageous to either get 50+ 60 min subs, or 50 hours worth of 4 min subs (at 4 min dark current noise is greater than readout noise) but then total noise is greater, so it's a trade off?

Appreciate your time in helping me understand this. Of course, when you add sky glow, this will then impact the max sub length that can be taken. Is it beneficial to go beyond swamping out readout noise, by swamping out do you mean as far as possible before sky glow takes over?

[ianaiken]

To this you will end up having to add skyglow, with a lot of that then it means the readout noise is even more swamped and you can afford even shorter subs.

So the key is getting as many subs as possible to the point where dark current noise/sky glow is just above readout, therefore not swamping the sensor with too much sky glow. But then, some targets are so faint that you need longer subs to be able to get any detail out of them. So you could in theory go for 60 min subs if we ruled out sky glow, which won't happen in my LP area.

Just trying to understand the non physics part first, if I can understand the objective then then the physics bit I can work out easier since I know what the aim is.

[rfdesigner]

@Derek

Since I'm shooting at ISO800 I think the readout noise is roughly 5 rather than 7. Is that right (I'm using the link you provided).

correct

Let's take Sky Glow out of the equation for a minute and pretend it does not exist. A couple of questions which I think may help with my understanding.

1. A 60 min sub would be the ideal in the excel since total noise is the lowest for 60 mins worth of data?

Yes.. concidering purely noise, a single 60 minute sub will outperform two 30 minute subs, but only by a small fraction of a magnitude.

2. 3 min subs would not be advantageous since dark current noise is below readout noise (I adjusted my sheet to have a readout of 5 instead of 7)?

Yes. The readout noise in each frame would be significant compared to the dark current noise in each frame, 5 seems fine for ISO800.

3. A 60 min sub would not allow stacking, therefore it is advantageous to either get 50+ 60 min subs, or 50 hours worth of 4 min subs (at 4 min dark current noise is greater than readout noise) but then total noise is greater, so it's a trade off?

ahh lets try not to talk about anything other than 60 minutes of data collection for now, then we're comparing apples with apples. However if you take two hours of data (say) then you will end up with 40% more noise.. but also 100% more signal, so the signal to noise improves.

Appreciate your time in helping me understand this.

no probs.. no doubt others will be following this as well, we all try our best to help, I've already benefitted quite a bit, so it's nice to give a bit back.

Of course, when you add sky glow, this will then impact the max sub length that can be taken. Is it beneficial to go beyond swamping out readout noise, by swamping out do you mean as far as possible before sky glow takes over?

Sky glow is cumulative just like dark current, so in the situation where readout noise is dominated by dark current after say 10 minutes, then if you add skyglow that is like increasing the dark current, which means your readout noise is dominated in an even shorter time interval.

Sky glow may be more or less than the dark current, but as both occur at the same rate, that ratio is fixed regardless of sub length.

clear skys...

Derek

[rfdesigner]

So the key is getting as many subs as possible to the point where dark current noise/sky glow is just above readout, therefore not swamping the sensor with too much sky glow. But then, some targets are so faint that you need longer subs to be able to get any detail out of them. So you could in theory go for 60 min subs if we ruled out sky glow, which won't happen in my LP area.

Just trying to understand the non physics part first, if I can understand the objective then then the physics bit I can work out easier since I know what the aim is.

Aim:

Noise to be as close to minimum as you can get with short exposures. What we're aiming for here is a guess at just how short an exposure we can get away with. To see deeper with these short exposures, take more exposures of the same length

With a readout noise of 5e, that means once the dark current + skyglow come to 25e (the noise of which is sqrt(25e) or 5e) then

readout noise = dark&sky noise. When the dark + sky noise level comes to 100e then the readout noise is pretty much swamped.

Try making an exposure and measuring the background level, then compare against a bias frame. (remember 1 ADU reading is not 1e, there is a gain, the data I pointed you towards before has that gain in it.)

Derek

[ianaiken]

Derek,

Thanks. Going to read this in detail tomorrow. What about diminishing returns with a high number of subs, like over 50? Using shorter subs this will become.an issue? Maybe not if SNR is.actually overall higher in the shorter subs.

[dph1nm]

But then, some targets are so faint that you need longer subs to be able to get any detail out of them.

Tricky one this. In theory no - once you are well above the read-noise then adding subs should be no different from taking one long exposure. This, after all, is how the Hubble Ultra Deep Field was done (adding lots of 20min subs). Under certain circumstances (depends on the gain I guess), there may be issues with quantisation noise, which can alter the simple theory, although how it alters it is not straightforward.

NigelM

[ianaiken]

Nigel,

Thanks. I will keep away from quantisation noise for now, and stay with the "simple" theory. What about diminishing returns, or is this only for darks? I've read over 50, and certainly 100, that your not getting benefit.

I've got some subs at different lengths that I am going to look at the background level and comparing it against a bias sub as Derek recommended. I've also seen a few tutorials on how to work it out that I am going to look at.

There is also a small section of reading in Jerry Lodriguss PS for Astrophotography that I am going to read over again. One of his main points, which is discussed in this thread with the addition of the noise, is exposing to the sky fog limit, under and detail is lost, over and faint detail will be lost in the sky fog.

I don't think this is something that I will understand overnight being away from any kind of Physics and Maths for a long time, but it is very interesting and I think it will click soon once I develop some fundamental understanding during daylight hours (I've only been doing any kind of Astro work since March this year, so new to the field for sure).

Going to keep at the theory for now, I want to maximise my imaging and produce some good images, and I think getting this right will help significantly without taking away any enjoyment. I like to throw myself in the deep end.

[dph1nm]

What about diminishing returns

Returns generally go as the square-root of whatever is increasing. So 10,000 subs are 10x better than 100 subs, which in turn are 10x better than 1 sub (assuming the subs are all the same length). So adding 99 subs to 1 you already have is great (10x improvement) , but adding 99 if you already have 100 only gets you about 1.4x times improvement. I think this is what folks mean by 'diminishing returns'. This works for darks, lights, biases, even flats.

exposing to the sky fog limit, under and detail is lost, over and faint detail will be lost in the sky fog

Hmm - don't understand that. Exposing longer than the sky fog limit will lose *bright* details (through saturation), not faint.

NigelM

[ianaiken]

Nigel, indeed when you put thought against the saturation that makes sense. Not sure what Jerry is talking about then, unless I've read it wrong, or it was specific to film although I would have thought same principals apply.

[rfdesigner]

sky fog limit sounds like film.

expose film too long and you just increase the background to a point where it's very noticeable, and saturate all the bight areas, each film was different, nightmare!.. no subtraction of a dark frame or gradient adjustment, other than by very messy means which I never tried.

Derek

(boy am I glad I don't have to worry about reciprocossity failure and manual guiding any more!)

[ianaiken]

Derek, Nigel, any chance you could check my understanding here?

1. Canon 40D gain at ISO400 is 0.84e So gain at ISO 800 is (400/800)*0.84 = 0.42 e/ADU ??

2. My camera is 14 bit sensor so 2 to the power of 14 = 16384 max ADU?

3. For calculating sky background in Maxim, so I need a calibrated light frame or one processed with Bias and Darks. When you say compare with a Bias do you mean remove the Bias level from the resulting average ADU of the sky background? If not, what?

4. Will Maxim show me ADU higher than 16384 with my RAW file or is it a case that if the ADU value of the bright stars are 16834 then they are likely saturated?

Some observations I have made so far. I think my ADU on an uncalibrated frame is somewhere around 1600. However, mistakenly the sub I used does has IC1396 in the background so I think I need to do this in a part of the sky where there are no targets. Also, I guess dec will have a play in this, more sky glow the lower I get? In an uncalibrated light my brightest start is 15000 @ ISO800, 600s sub, 15C.

I did a measurement of a calibrated frame with bias and darks applied, however, it only seems to have done a quarter of the image, bottom right. ADU here is around 800. Again with a light with IC1396 in the background, so not a true measurement.

Am I on the right track?

[ianaiken]

Aim:

With a readout noise of 5e, that means once the dark current + skyglow come to 25e (the noise of which is sqrt(25e) or 5e) then

readout noise = dark&sky noise. When the dark + sky noise level comes to 100e then the readout noise is pretty much swamped.

Derek

So at 100e total noise (dark + sky) and readout noise is 5e which is the recommended 5% readout rule some adhere too.

[dph1nm]

When you say compare with a Bias do you mean remove the Bias level from the resulting average ADU of the sky background? If not, what?

You need to measure the noise per pixel in a single bias frame. This will give you the read noise. Then do the same on a light (without any processing, and away from objects). Subtract the square of the read noise from the square of the light noise and square root the answer. This will give you the dark+sky noise. Do all this in elections by the way. Then you can compare the two. Personally I think 5% is unnecessarily small, but it is all a matter of choice.

Strictly speaking, on a colour CCD or DSLR, you should do this for each channel (RGB) separately before the debayering process. I don't use Maxim so I have no idea if it will allow this.

Your gain looks about right. There are some figures here.

NigelM