What ISO?

[Andrew*]

I know this question has come up umpteen times, having just done a forum search, but I'm none the wiser having read the answers! It seems to be a matter plagued with variety from personal preferences, experiences, and the camera model.

Some people only image in ISO 1600 while others top out at ISO 400.

So, for my Canon 450D, what ISO would you recommend I use for DSOs?

Andrew

[jgs001]

Andrew I read somewhere that the Canon iso sweet spot is ISO800 and that's generally what I use.

[Sam]

Andrew,

I would recommend ISO 800, that's what I use and I think it gives me pretty good results. I read a bunch of the tech articles that were posted a while ago about noise and that's why I settled on ISO 800. Though when I was imaging with the 8" reflector I was using ISO 1600, it seemed to give a better result but with the new 10" ISO 800 is the one I use. It's probablt best to experiment and see what works for you as many other factors will come into play, such the sky brightness, light pollution, brightness of object, moon phase etc etc.

Sam

[blinky]

I just about always image @ ISO 1600 - with the nights we have there is no time to experiment!

[Ben Ritchie]

I use 800 as well, although there are arguments for 400 or 1600.

[blinky]

My reasoning for 1600 ISO was that there will be more noise but (supposedly) more detail. The noise will be averaged out by the large number of frames and darks so in the end I will have a stronger signal and the same or less noise.

But I could be talking nonsense!

[Ben Ritchie]

Heck, after reading lengthy arguments on CN for both 400 and 1600 I settled on the middle option

[dark knight]

I'm no expert by a long shot, but here is my view for what it is worth.

The ISO also plays a part in the length of exposure used and not just the camera's sensitivity, a higher ISO enables faster shutter speeds to be used and therefore more frames can be grabbed in a shorter period of time. This may not seem too important, but if your polar alignment isn't spot on the use of a higher ISO can be beneficial, as the faster shutter speed can help alleviate star trailing in the image...Start at ISO 800 and see how it goes. Are you going to shoot in RAW mode?

[dark knight]

PS. By shutter speed I mean length of exposure in bulb setting.

[brianb]

The ISO also plays a part in the length of exposure used and not just the camera's sensitivity, a higher ISO enables faster shutter speeds to be used and therefore more frames can be grabbed in a shorter period of time.

But the frames are noisier too. To get signal to noise ratio, there's no alternative to getting more light in the big end of the lens. The ISO setting DOESN'T MATTER!!!!!

[Andrew*]

Thanks all. Things seem to tip in the favour of ISO 800.

I have heard of people using a combination of 800 and 1600 as well. What do people think of this?

Andrew

[dark knight]

But the frames are noisier too. To get signal to noise ratio, there's no alternative to getting more light in the big end of the lens. The ISO setting DOESN'T MATTER!!!!!

So you are saying ISO is not a relevant factor, surely comparable longer exposures at lower ISO's will also introduce noise as the camera generates heat etc..I did say I was no expert and maybe we would benefit if you clarified your view point as to exactly why ISO is not important. If it was not important why don't people use an ISO of say less than 400??? I would hazard a guess that the exposures would have to be too long to get the required level of detail. Apologies if I have taken your comment wrongly, but to say ISO is not important seems a bit strong.

[Aza]

800 for me. Unless it is really bright. I think I did 400 for my M13 image.

[Psychobilly]

Mainly 800 with a Dash of 1600 and the odd little embelishment of 400, 200 or even 100...

You wont go far wrong with 800...

[blinky]

It's a question I have ofter pondered! A CCD/CMOS Sensor cannot alter it's sensitivty - the ISO is how much the signal is amplified as it's read from the sensor (at least thats my understanding) so 'in theory' should we not all just use ISO100 as the sensor is just as sensitive at this setting as at ISO1600?

[Andrew*]

Craig,

I was also thinking along those lines, but I think that something is lost in the saving, even to RAW, if the histogram is too dark. By upping the ISO, you're actually saving mroe of the faint stuff.

Very basic theory of course - more than happy for folk to correct.

Andrew

[adamsp123]

I must admit I find this all very confusing, if as Brian states ISO setting is irrelevant then why all the debate on ISO settings? lets use minimum (for best "quality") and have done with it! but I think there is more to all this - in this electronic world of photon processing and the resultant image we wish to see, but I don't fully understand it.

If I took a series of subs at 100 ISO and the same at 1600 the number of photons hitting the sensor would be similar but the resultant stacked and processed images would be very different would they not?

Pete

[narrowbandpaul]

taken from my earlier post a few moonths ago. I hope to confirm with exp evidence

I believe the issue of ISO to be misunderstood. An image taken at high ISO looks brighter on the camera screen and when downloaded, but is there any more signal.

An important figure of merit for judging image quality is the signal to noise ratio. It is the ratio of stuff you want (signal) to stuff you dont (noise).

signal is produced when incoming photons are absorbed by the silicon and 'create' an electron. You can increase signal by using faster optics (lower F number) or by using longer exposures or by using a camera with a higher quantum efficiency, QE.

noise is the non-perfectness in this process. More correctly, shot noise, is the imperfections in this photo-electric process. Say you have a 3x3 grid of pixels. Say also you fire 1000 photons at each pixel. With the same efficiency of each pixel, the number of electrons generated should be the same. However when you look at the number of electrons generated you see that each pixel has slightly more or less charge than its neighbour. This is a statistical variation due to the arrival, or counting of photons, it is called shot noise. The typical variation in pixel charge is the square root of the number of photons detected. If you have a 0.4 QE then each pixel should have intercepted 400 photons and should have 400 electrons in every pixel. The typical variation is the square root of 400, which is 20. The shot noise is 20e-. That means we expect every pixel to have 400e- +- 20e-.

you will always have shot noise in your images. However you want longer exposures (more signal) to compete with this noise. If you collect more photons, then the noise will also grow. But the signal grows faster...so more data =higher S.N.R

the noise is proprtional to sqrt(S) and the Signal is S. so the ratio S/sqrt(S) is sqrt(S)...ie SNR grows with signal......longer subs give better images.

because this is a random process the shot noise is random in each image. Stacking many frames together will cancel out some of the shot noise, and the SNR will be higher. The SNR is proportional to the number of images taken. eg take 25 subs, average them, and the SNR is 5 times higher over a single sub.

there is another type of noise, that will affect your images......read noise. And this is where the ISO comes in (I firmly believe)

Read noise is generated during the signal processing step, ie where the signal in each pixel is measured. This is first carried out by a 'sense node' which is just a capacitor....which creates a voltage dependent on the signal in the pixel.

read noise will appear in your images even if there is no signal there. If you take a really short exposure in the dark, eg lens cap on, there will be a variation in the signal despite no photons ever been collected. This is read noise. A typical value may be 10e-. This number will not vary with signal, unlike shot noise. If your subs are short then it may be the case that the read noise is higher than the shot noise. This is called read noise limited.

You can reduce the read noise by stacking many frames, as it too is random.

When astroimaging it is quite common, very common in fact when doing narrowband imaging, where your subs, despite being 20mins long, are still read-noise limited. That is to say, your read noise is higher than your shot noise, which really should be the only noise source in an ideal world.

Thus read noise is important when detecting faint sources.....and now for the ISO.

Thinking about it, if I take 2 exposures, identical apart from ISO speed, what is the difference. If I use ISO 800, and view the image on a computer or LCD camera screen, then I say "that image is bright, the object stands out, I must have more signal". Now say I use ISO100. Now I say "the image is dark the object is hardly there, I have less signal".

Is this correct though?

I exposed for the same time, with the same lens and f number. How did I suddenly recieve less photons when I changed ISO. Do the photons know I changed the ISO. Of course not. Your camera is lying to you!

Lets follow the 'read-out' chain, and see what happens to electrons.

The signal in each pixel is measured by a sense node, which creates a voltage dependaent on the number of electrons it recieved from the pixel. This voltage is amplified, and next turned into a digital number, DN, by the output amplifier. Then the DN is downloaded to PC.

The moral is, you dont see the number of electrons that were collected, you see the DN, which has been amplified by an amount you dont know.

for example.

lets say two pixels each have 1000e-. The first one is read out, and amplified with a gain (DN/e-) of 5. Ie each electron corresponds to 5DN. Thus the 100-e- charge packet will be represented as 5000DN in the final image you view on a PC. Say the next 1000e- charge packet has a gain of 0.5 applied. The final DN will be only 500DN. So the DN you see is completely meaningless when it comes to image quality, and the signal you think you have.

The only thing that determines the amount of signal you will recieve is f number, exposure lenngth, QE. Definately not ISO.

ISO does not give you more signal. It just boosts it more. Infact SNR is dependent on the ISO. A high ISO may degrade SNR, and a low ISO may be better for imaging.

heres my theory.

the first stage of read out involved the sense node. This creates a voltage. The voltage created per electron is governed by a number called the sensitivity. The sensitivity is determined by the capacitance.

I believe that the ISO is applied here, and changing the ISO changes the sensitivity. A high ISO has a high sensitivity. But a high sensitivity actually produces lower read noise. But a high sensitivity means the analogue digital converter will saturate before 'full well capacity' is reached. This will result in lower dynamic range, and most astro objects have a high dynamic range, M42 for example, Amanda.

a lower sensitivity will have a higher read noise, but higher well capacity and dynamic range. The more electrons that can be collected means you will have a higher SNR.

I expect the best value will be either ISO200 or 400. But it will vary from model to model. Testing it is do-able but tricky.

But dont be fooled in to thinking that ISO1600 is best because the image is brighter........

[brianb]

A CCD/CMOS Sensor cannot alter it's sensitivty - the ISO is how much the signal is amplified as it's read from the sensor (at least thats my understanding) so 'in theory' should we not all just use ISO100 as the sensor is just as sensitive at this setting as at ISO1600?

Precisely. The sensor has a native sensitivity - usually about 200 - all the other settings are obtained by digitally multiplying the value by some constant which varies with the setting. Hence more noise at high ISO. Hence also less dynamic range at high ISO, saturation occurs when the value adjusted up exceeds 2^16-1 (16 bit values....) There is some loss of tone resolution at speeds below native, 50 & 100, this makes smooth areas smoother than they really are.

A high ISO image only looks brighter than a low ISO image with the same exposure settings because of the way the display program works. Stretch the levels & the images look very nearly identical irrespective of the ISO setting - actually identical but for random (readout & quantum) noise, clipping of overexposed areas at high ISO and tone smoothing at low ISO.

[Ben Ritchie]

Read noise is also an issue - there's a bunch of stuff online that suggests that at low ISO the read noise is higher in the Canon DSLRs. That's introduced downstream from the sensor, I think the ADU. So in pure signal/dark current terms the ISO doesn't matter as all it's doing is changing the gain, but the read noise is relevant.

[jgs001]

That all makes sense, and I guess that explains why testing in the field has produced results that indicate that ISO800 is the sweet spot for the consumer Canon's. One question about shooting lower ISO, such as 200 or 400. Assuming that your exposure time is limited, and I've also seen suggestions that SLR exposures for astro shouldn't go beyond 5 minutes, if you want to pull the details out of the stacked subs with the lower ISO, are you not going to have to stretch the image further than with the higher ISO and therefore end up with similar, if different noise ? I know when I stretch the histogram for normal photos, that the dark areas become much more noisy when I do that. I've not tried for Astro, although with the Owl I did on Saturday at ISO400, the stretch was limited by the noise in the image, I used 400 as I was pushing for 20 minutes and it's been my experience that the sky glow gets in the way beyond 10 minutes at ISO800, so I halved the ISO for double the exposure time.

[dark knight]

thanks to all for explaining, it does make sense now..

[brianb]

and I've also seen suggestions that SLR exposures for astro shouldn't go beyond 5 minutes

Why? Hot pixels, amp glow etc. can be removed by subtracting dark frames ... If you have a sky glow problem, that is less tractable but anti-pollution filtering may help. Those with serious skyglow problems usually resort to shooting with narrowband filters on a monochrome CCD camera.

[jgs001]

I don't know Brian... I did a 20 minute exposure to test out guiding, and apart from hot pixels there wasn't an issue, I used a CLS to remove the skyglow and dropped the ISO to 400. I had no amp glow. I didn't take any darks to remove the hot pixels as it was only a test of tracking.

[Beyond_Vision]

It is possible that the amplification process adds noise to the end result if it is in the analogue domain. If you take the analogy of a audio pre-amp and main amp each stage of amplification can introduce noise from for example it's own power supply. Is the ISO setting purely a multiplier of the ADU count?

Regards

Kevin