Which would you choose?

[Demonperformer]

If you wanted to get (for example) 6 hours of OIII data on an object (and assuming your tracking/guiding were up to the longer subs) how would you split the time - 72*5m, 36*10m, 24*15m, 18*20m, or 12*30m?

I understand the basic theory that the longer the subs the more data they collect, which is good. But balanced against that you have: (1) the fewer the number of subs, the greater the noise in the finished result, & (2) the greater the loss if a sub is ruined by (for example) an airplane flying through it.

Would anything be gained by mixing the exposure lengths (for example, 24*10m+4*30m, so getting lots of subs for noise reduction, but also some long-exposure data)? Or would this be introducing unnecessary complications?

Thanks.

[swag72]

This is really interesting for me ......

...until now I have been a real advocate of long exposures, certainly for narrowband and I do nothing less than 30 minutes. However, I have recently been given cause for thought and I am doing a little investigation.

I've got some data on NGC1499.

8x1800s (4 hours)

17x600s (2hrs 50m)

I stacked both of them along with 6x1800s (3 hrs) ..... when I put the stack through PI and produced some graphs, the 10m subs won every time! The 10m subs gave me a smaller FWHM, rounder stars and less noise.

So if I only had 6 hrs to gather data - I'd go for 10m subs I think ........

[Kropster]

I suppose that more shorter exposures will produce a lower noise result, as the average noise will drop the more subs used.

Would you lose some of the faintest detail in an object by shortening the exposures? ...maybe not reaching the threshold at which the sensor produces an observable output?

[swag72]

Would you lose some of the faintest detail in an object by shortening the exposures? ...maybe not reaching the threshold at which the sensor produces an observable output?

When I stretched and looked closely at the difference in a stack of 10m subs and 30m subs, there was no difference in the faintest detail that I could see. What I saw in the 30m exposures was equally there in 10m...

[opticalpath]

This is really interesting for me ......

...until now I have been a real advocate of long exposures, certainly for narrowband and I do nothing less than 30 minutes. However, I have recently been given cause for thought and I am doing a little investigation.

I've got some data on NGC1499.

8x1800s (4 hours)

17x600s (2hrs 50m)

I stacked both of them along with 6x1800s (3 hrs) ..... when I put the stack through PI and produced some graphs, the 10m subs won every time! The 10m subs gave me a smaller FWHM, rounder stars and less noise.

So if I only had 6 hrs to gather data - I'd go for 10m subs I think ........

Well that throws the cat amongst the pigeons, Sara!  I've always believed the theory - that fewer-longer will always be at least as good as more-shorter ........ assuming good tracking is maintained for the longer exposures and brighter areas are not saturated.  The smoothing-out effect of stacking many short exposures certainly reduces noise, but the signal-to-noise ratio is poorer to start with in a short exposure compared with a long one.  Stacking a large number of low SN images should not produce a higher SN result than stacking a smaller number of high SN ones.  We've captured the same total photons and the same noise in both cases if total exposure time was the same (apart from some extra read noise in the many-shorter case).  That's just theory of course.  I've recently tried a 1 hour exposure but did not think it any better than 2 x 30 min., but that wasn't much of a test.

 A few things I'm wondering about:

Do small guiding errors accumulate 'worse' in longer exposures and produce stars with higher FWHM compared to shorter exposures? 

Are longer exposures more prone to changing seeing conditions or slight loss of focus during the exposure - affecting FWHM adversely?

Is dark subtraction somehow adding disproportionately more noise in longer exposures?

Is 8 exposures enough for sigma-reject stacking to work effectively - can that increase noise?

Very interested to hear if you come up with any more thoughts on this. 

Adrian

[Galaxyfaraway]

Is the exposure time not target - and filter-dependant? Would you still see that squid thingy with only 10 mins?

[opticalpath]

I just had a rummage through my old archive and found an image from 2003 (  ) taken with a modified video board-camera.  140 x 12 sec plus 140 x 20 sec. exposures, unguided, from very L-P'd suburban garden shows stars to magnitude 20 ....  so it can work.  But read noise becomes a significant factor with very many short exposures and that can set the practical limit.

Adrian

[ollypenrice]

Sara's doing this comparison carefully whereas I've only ever done it 'seat of the pants' and anecdotally. When going after the structured shape of the outer parts of M31 I found I could get it using 30 minute subs but not (at all) using a longer overall set of 15 minute ones. I would not shoot Ha in less than 15 minute subs. I'm still a long sub person but Sara and I have been talking about this off the forum and I'm certainly going to do some testing.

The other thing is that it may well vary from camera to camera and with focal ratio. My cameras have huge well depth but low quantum efficiency. This may make them favour long subs.

What's this about an aircraft ruining a sub? No no no! A good sigma stacking routine in a stack of more then 10 should 'disappear' it perfectly. If it doesn't, do this.

- Make a full stack with the plane.

- Make a stack without it.

- In Ps paste the plane stack twice on top of the clean stack, so two upper layers with planes.

- Stretch the top layer so you can the plane. Leave it visible.

- Make the middle layer active.

- Run the right sized eraser down the middle layer plane trail. You won't see it change anything but that middle layer now has no plane in it.

- Delete the top (stretched) Layer.

- Flatten the planeless full stack onto the shorter stack and you now have a planeless (painless!) full stack.

I can see no point whatever in mixing sub lengths except in dealing with high dynamic range and, apart from with M42, I don't do this.

Olly

[ollypenrice]

 Would you still see that squid thingy with only 10 mins?

Personally I doubt it. Sara? Attagirl!

The best OIII squid I've seen was Julian's, taken from his place not far from me. I don't recall his sub length but he used an Atik 460 in Bin 2.

Olly

[swag72]

I'm not sure I'm ready for the squid yet Olly!! This is only a work in progress so far looking at sub lengths (well that is the optimum for my camera). 

[andrew s]

When judging which is better of many short or a few long exposures adding to the same net exposure you need to consider the total noise added not only from the exposure but also from the processing.

While a single long exposure is potentially optimal (assuming you don't saturate the image) compared with say 10 shorter images it depends on the processing and the noise added there. All images (target and calibration) have read noise, thermal electron statistic noise and if exposed to light photon statistics noise and possibly non-linearity due to saturation. If you don't reduce the noise by combining many dark frames then just dark subtraction can dominated the noise. As could dividing by a noisy flat field. Getting sufficient dark frames when long exposures are involved can take a considerable time - a day or more. Some flat fields through filters, especially narrow band filters, can prove a challenge if the illuminating source is weak in certain spectral bands. It is best to check the noise levels by measuring them using your processing software and not just by eye.

You need to look at the net S/N ratio to compare the two strategies.

Regards Andrew

[Demonperformer]

Thanks for all your responses.

I seem to have asked this question at an "interesting" time (as in the old Chinese curse!) and I shall certainly be keeping an eye out for future developments of Sara's & Olly's testings.

Some interesting thoughts there, Adrian. I would think that longer exposures are going to put greater "strain" (possibly the wrong word) on the system, requiring smaller tolerances (even if aircraft can be made to disappear - thanks, Olly).

Talking of the squid, surely the question is not would it appear on a 10m sub (which it certainly wouldn't, or it would have been discovered a lot earlier than 2011), but would it appear on a big enough stack of them. Granted the cumulative readout noise would be greater on more shorter subs and so you might need a greater integrated time. But you are surely going to get it eventually? (There was an image [not NB] on SGL a few years back taken of the Horsehead nebula, consisting of a ridiculously large number of 2-second subs.)

Andrew, I hadn't even started to consider the additional time to get all the calibration subs for longer times, which would seem to be another reason to limit them.

Thanks.

[Martin Meredith]

Interesting challenge! I'm willing to bet you could get the Squid using (lots of) subs of less than 10m (and if using a low read noise camera, using even shorter subs). I think its an important experiment to do: what is the lower limit on sub length for really really faint stuff?  

Just to throw in another data point: it is possible to pick up mag 20 galaxies in (stacks of) 10s subs under less than ideal skies (SQM 20.1) with a moderate read noise camera (Lodestar X2). Admittedly, these point-like sources are much easier than a low SB object.

Martin

[Demonperformer]

Just found http://www.cloudynights.com/topic/478773-ou4-the-squid-nebula-in-oiii-wip/... 17 hours using 30m & 60m subs ... and yet it seems to me to be fainter than http://cosmicneighbors.net/sh2-129.htm which was only four hours of data. OK, it's f/5.6 instead of f/4, but that should still only require twice the time? There's obviously more to this lark than just the number-crunching!

[Demonperformer]

Thinking a bit more about those two photos. The one with 4hrs@f/4 was binned 2*2. IIUK, that means that is equivalent to 16hrs@f/4 without binning, which would equate to 32hrs@f5/6, which is roughly twice the time combined in the first picture. So, if the first picture is taken without binning, that would make more sense.

BUT, that opens up a new question about previous responses: if people say that they take a particular exposure length NB, is that without binning or including some sort of binning. If without, then presumably a (for example) 20m sub without binning would equate to a 5m sub binned 2*2?

Thanks.