DSLR iso options

[fredman]

Hey guys

I've been playing around with my Nikon D5100 to get to know it (first DSLR) and i found that as Hi0.7 / Hi 1 and Hi 2 iso.

Every iso option as an image to show the type of ambience you can use it for, and for the three above as a night sky full of stars.

Well, i have seen that a lot of AP that use iso 800 and such. Are those three good for AP?

As anyone used this type of iso?

Thanks and clear skies

[kalasinman]

After applying Nikonhacker firmware, I get fairly good results up to ISO 3200, without it I never imaged over 800. Every camera has a sweet spot that yields the highest dynamic range. I believe for a stock D5100 this is ISO 640. Having said that, my current approach is to image at ISO 200-640 with as long an exposure as I can manage

Long exposures are the best way to bring out faint details, as pixels only become active after a certain number of photons are received, less than that they report a 0.

Increasing ISO does nothing to help this as 1000x 0 = 0. Photons received is determined by aperture and time, with objects of equal brightness.

[dph1nm]

Long exposures are the best way to bring out faint details, as pixels only become active after a certain number of photons are received, less than that they report a 0.

This is only an issue at very low ISO - once you are at unity gain then you get one count for one photon. This varies from camera to camera, and it is rather difficult to find any numbers for the 5100, but it looks like it might be around ISO800.

NigelM

[fredman]

I will be sure to try ISO 800, and play with it a bit. Thanks for the input everyone

[fredman]

Yep, it' comfirmed.

If i go higher then ISO800 the photos get to much brightness. Perhaps because i dont lower the exposure time, could that be it?

I did ISO800, f/4 (?) at 32mm if i am not mistaken, and 10 seconds.

Nevertheless, i am in the city centre so... I will be picking all the city lights

[kalasinman]

This is only an issue at very low ISO - once you are at unity gain then you get one count for one photon. This varies from camera to camera, and it is rather difficult to find any numbers for the 5100, but it looks like it might be around ISO800.

NigelM

My understanding is that even above unity gain the QE of the sensor, be it CMOS or CCD, produces less by QE % than input. Thus a sensor with QE of 50% would produce 5 counts for every 10 received, etc.

[kalasinman]

Another issue is that the D5100 and D7000, among others, use black point clipping . This means that any signal below a certain level is entered as zero. Thus all photons below that set point are discarded. This is one reason Nikonhacker developed the "True Dark Current" hack for firmware packages. This enables the very low read noise levels , below some CCDs, of the Sony sensor to be leveraged into higher dynamic range and increased sensitivity to faint signals.

Increasing ISO cannot increase the number of pixels actually received. It just multiplies what count is reported above the QE of the sensor. One issue with this  is it increases the likelihood of saturation and blooming.

Only way to increase photons received is to increase time and aperture.

[sharkmelley]

For Canon cameras, the read noise generally goes down as the ISO goes up but at the same time you lose dynamic range and the highlights will saturate earlier.  So setting the ISO is a compromise.  Usually somewhere around ISO 800 is right but it depends on the exact model.

However, the Nikon 5100 is one example of what is known as an ISO-less camera i.e. the read noise is more or less identical at all ISOs.  In this case I would recommend ISO 100 because this will give you the highest dynamic range.  You can see some approximate read noise and DR figures for various cameras at http://www.sensorgen.info/

In that table compare, for instance, the Nikon 5100 with the Canon 550D and you'll see the varying read noise effect.  Don't put too much trust in Sensorgen figures because better testing techniques give more accurate results.  But they do show the general trend.

My argument above applies to multiple long exposure deep sky imaging but if you are simply taking single exposures of star trails and constellations and you want to avoid heavy post processing then simply select the ISO that gives you the best exposed image previewed on the rear LCD!

Mark

[kalasinman]

Thanks Mark. The guy with the chart must like the Nikon D70.. a QE of over 200% ....???? HMMMM. Is that possible? Or desirable?

The developers of Nikonhacker have measured read noise for the D5100.of 1.7 e at room temp after the firmware mod.

[sharkmelley]

Thanks Mark. The guy with the chart must like the Nikon D70.. a QE of over 200% ....???? HMMMM. Is that possible?

It's not physically possible.  Some of the figures on that page are complete nonsense but for most cameras they seem to make sense.  They need backing up with proper testing such as that done by Christian Buil.

Mark

[dph1nm]

Thus a sensor with QE of 50% would produce 5 counts for every 10 received, etc.

OK, at unity gain you get one photoelectron per detected photon. The fact that 50% get lost somewhere in the silicon is taken care of by the shot noise.

In this case I would recommend ISO 100 because this will give you the highest dynamic range

And horrrible quantisation?

NigelM

[kalasinman]

Not here to argue Nigel. You seem to have modified from your original statement that 1 in =1 out at some point. Clearly that's not possible if you're talking actual photons hitting the sensor.

The level of the sky fog is what it is. Just getting the signal histogram above the leading edge of it by 15% or so of histogram width seems to be an effective technique. I'm not sure 100 ISO is the answer, although the developers of the firmware imply that it may be useful.

Seems to me that read noise of 1.7 e is out of the experience of most and that  looking at altered techniques may be suggested.

With a stock D5100, I'd agree with the OP conclusion to image at ISO 800.

[sharkmelley]

And horrrible quantisation?

No, I really don't think that is a problem.  On that camera at ISO 100, the gain is just over 2 electrons per digital unit.  Even if quantisation were a problem with a single exposure, once you stack exposures then the statistics makes the problem disappear i.e. the sum of digital units statistically converges to the sum of electrons.

Mark

[dph1nm]

Even if quantisation were a problem with a single exposure, once you stack exposures then the statistics makes the problem disappear i.e. the sum of digital units statistically converges to the sum of electrons.

This all depends on the ratio of the quantisation steps to the noise, in much the same way as for spatial resolution. If you undersample the noise you will lose information.

NigelM

[dph1nm]

Not here to argue Nigel. You seem to have modified from your original statement that 1 in =1 out at some point. Clearly that's not possible if you're talking actual photons hitting the sensor.

If you have a low ISO such that gain is 2e- per ADU, then if you never receive more that 1 photon per exposure you will never detect a signal, even if the QE is 100%. On the other hand, if the QE is 50% and you have 1e- per ADU then you will detected the expected signal (i.e. 50% of all photons  reaching the camera).

This argument is modified by the presence of noise (see my post above) but quantisation is usually a bad thing to have.

NigelM

[kalasinman]

Just curious, if the read noise of the sensor is low, below 2, and constant regardless of ISO would that not be a little different?

[sharkmelley]

Nigel is absolutely correct that quantisation adds an extra noise contribution (quantisation noise) to the already existing read noise, shot noise, skyglow noise and thermal noise and therefore should be avoided where possible.

My argument is that for this particular Nikon camera (which is ISO-less) the quantisation caused by the gain of 2 electrons/digital unit at ISO 100 will not be visible for all practical imaging because it will be swamped by the other noise sources. 

But you would not use such a low ISO on Canons and most other Nikons because the read noise increases to high values at low ISOs (because they are not ISO-less).

Mark

[MikeODay]

For Canon cameras, the read noise generally goes down as the ISO goes up but at the same time you lose dynamic range and the highlights will saturate earlier.  So setting the ISO is a compromise.  Usually somewhere around ISO 800 is right but it depends on the exact model.

However, the Nikon 5100 is one example of what is known as an ISO-less camera i.e. the read noise is more or less identical at all ISOs.  In this case I would recommend ISO 100 because this will give you the highest dynamic range.  You can see some approximate read noise and DR figures for various cameras at http://www.sensorgen.info/

In that table compare, for instance, the Nikon 5100 with the Canon 550D and you'll see the varying read noise effect.  Don't put too much trust in Sensorgen figures because better testing techniques give more accurate results.  But they do show the general trend.

My argument above applies to multiple long exposure deep sky imaging but if you are simply taking single exposures of star trails and constellations and you want to avoid heavy post processing then simply select the ISO that gives you the best exposed image previewed on the rear LCD!

Mark

Am I understanding this correctly - the read noise is a constant value per exposure regardless of exposure length?

Wouldn't that mean that unless the read noise is extremely low at high ISO we would be better using low ISO and increasing exposure?

eg. consider the following:

For my camera (Nikon D5300) sensorgen.info states the read noise at ISO100 as 3.4 and at ISO1600 it is 2.

- for the sake of the example, assume on average 1 electrons are landing on each pixel per second

- consider 2 exposures with the same EV: 1 sec at ISO 1600  or 16 sec at ISO100

- this means at ISO1600 the camera would collect 1 e / pixel which is less than the read noise at ISO1600 and hence we are missing out on available signal

- and at ISO100 the camera would collect 16 pixels or over 4 times the read noise floor

If the above is correct (and ignoring tracking errors) then it would imply that low ISO / long exposures are better unless the read noise is drastically lower at high iso (in my example it would need to be 0.2 e- in order to achieve the same margin).

Have I misunderstood?

[sharkmelley]

Correct, the read noise is not dependent on exposure length but it does tend to change with ISO.

For low-light photography you are always better off increasing exposure because you increase the number of photons collected.  So 1 sec at ISO 1600 can never be considered to be equivalent to 16 sec at ISO 100 in terms of image quality.  In your example, the 16 sec exposure will give you a better signal to noise ratio than a 1 sec exposure for both ISO 100 and ISO 1600 because in both cases the signal is greater then the read noise.  Your 1 sec exposure will be very noisy because it is difficult (but not impossible) to see the signal within the read noise.

Mark

[MikeODay]

Thanks Mark

[fredman]

Hi all,

I subscribe what was said here

The other day i was doing some AP of Jupiter and i think (i have to review the photos again and post here) the photos taken with ISO100 are less grainy/noisy than the ones with higher ISO.

Of course the exposure time was lower than 5secs. No tracking mount [emoji14]

[MikeODay]

Hi All

I am still struggling with what is the best ISO to use.  Based on this thread and some external links it would seem that the best ISO is the one that gives the widest dynamic range whilst also being near to unity gain (so I guess in my case this would likely be somewhere in the 100 to 400 ISO range).  I understand the part about dynamic range I think but the importance of unity gain still eludes me.

Forgive the possibly stupid question but these concepts are new to me -

If ISO100 has gain of 2e- per ADU and ISO200 has gain of 1e- per ADU then for the same EV don't you get the same recorded signal?  That is, you will have half the gain at ISO100 but to maintain the same EV you would also have to double exposure time so the total number of electrons / the gain is the same.  Have I miss understood?

Also, if one is imaging in semi-dark/light polluted sky (as I am) then one would normally increase the exposure so that the sky is clear of the extreme left of the histogram and hence aren't the low signals we are speaking about likely to be overwhelmed by the sky with the first usable deep-sky signal resulting from many hundreds of photons?  Would this tend to dilute the importance of having unity gain?

I was thinking of using the following guide:

- determine the longest exposure time that my setup can still reliably produce reasonably shaped stars (~= tracking time limit) (for me this is currently somewhere in the range 2 to 4 mins and I'm still working to increase it)

- using this exposure time, select the lowest gain (ISO) that will result in the background sky being exposed such that the histogram is shifted to the right and well clear of the extreme left

For me, with an exposure time of 4 minutes and on a moonless night this means I would be looking at an ISO of around 200 to 400 (@ f4).  If I can improve my DEC tracking performance I might get this is as low as 100 to 200 with tracking time of 8min.  What do you think of this approach to finding a good compromise?

Thanks

Mike

[sharkmelley]

If ISO100 has gain of 2e- per ADU and ISO200 has gain of 1e- per ADU then for the same EV don't you get the same recorded signal? 

No.  The gain of 2e- per ADU has the effect of rounding the number of electrons.  For instance if adjacent pixels collect, 0, 1, 2, 3,4 electrons then in your example at ISO 200 it will give an ADU counts of 0, 1, 2, 3, 4 (ignoring read noise) but ISO 100 will give counts of 0, 0, 1, 1, 2 which means in half the cases, the extra electron is ignored e.g. it cannot distinguish between 2 & 3 electrons - both give the same ADU.  "Miscounting" electrons is not ideal - it is called quantisation error.

There are generally 2 good reasons for choosing a higher ISO:

1)  Read noise generally reduces with higher ISO

2)  You can reach unity gain, where quantisation error disappears

Both the above help reduce overall noise and make it easier to extract faint signal.  However if you record so your histogram is not stuck at the left hand side then other sources of noise i.e. sky background noise) tends to be larger, making the above effects less important.  But I would still not want to record at 8e- per ADU nor at high levels of read noise.

However the original poster asked about a Nikon camera whose read noise does not increase much at low ISO and whose gain only increases to 2e- per ADU at low ISO.  With such a camera then ISO 100 should work well. 

In my opinion dynamic range is overrated for deep sky imaging.  If your dynamic range is low (which saturates highlights) then simply take some additional shorter subs without saturation.

The answer to the best ISO to use is very dependent on exact model of DSLR.

Mark

[MikeODay]

No.  The gain of 2e- per ADU has the effect of rounding the number of electrons.  For instance if adjacent pixels collect, 0, 1, 2, 3,4 electrons then in your example at ISO 200 it will give an ADU counts of 0, 1, 2, 3, 4 (ignoring read noise) but ISO 100 will give counts of 0, 0, 1, 1, 2 which means in half the cases, the extra electron is ignored e.g. it cannot distinguish between 2 & 3 electrons - both give the same ADU.  "Miscounting" electrons is not ideal - it is called quantisation error.

...

Thanks Mark, much appreciated.

Not arguing - just still not really understanding... my problem probably comes from my long history with photography and very short experience with astrophotography.  In photography I never really had to worry about the deep shadows. 

I'll try harder to explain the source of my confusion...

Following on from the previous example, suppose four adjacent pixels are registering e- at the rates of 1,  2, 3 & 4 e- per minute respectively.

At ISO200 with gain of 1e- per ADU and an exposure of 1 minute this would mean these pixels would register ADU counts of 1,2,3,4 

and at ISO100 with a gain of 2e- per ADU and the same exposure of 1 minute this would result in ADU counts of 0, 1, 1,  2, - ok I see that but really all we would be doing is halving the effective exposure so it is not unexpected that we would miss out on information in the deep shadows.

So really, by halving ISO one would also need to double the exposure duration to maintain the same EV, so the ADU count  at ISO100 with a 2 minute exposure would be: 1,2,3,4   - or exactly the same as ISO200 with 1 minute exposure.

I'm obviously missing something ... I'm sure it's a lot more complicated than my simple analysis.

I do understand your point about dynamic range though.  If we can simply use HDR processing to combine multiple exposure durations then the dynamic range of individual subs is less important.  I still need to work on this - my efforts with HDR on astro images have so far been quite poor.  For now I think I might need to maximise the dynamic range in each sub while I keep working on my processing techniques.

Cheers

Mike

[sharkmelley]

Let's go back to a 2min exposure at ISO using your example of 2e- per ADU.  Some pixels may indeed have electron counts of 2,4,6,8 which gives ADU of 1,2,3,4 just as you say. 

But another set of pixels may have electron counts of 1,2,3,4 which gives ADU of 0,1,1,2  which give us our original quantisation problem.

Another set of pixels may have electron counts of 101,102,103,104 which gives ADU of 50,51,51,52  which again is the quantisation problem being manifested.

So quantisation problems keep re-occurring at ISO 100.  It's a fault of the gain at ISO 100.  To a certain extent, longer exposures can help mitigate this, though a difficult statistical analysis is needed to prove this is the case.  It is better to avoid the problem in the first place if you want to extract faint detail from the image.  Things get far worse if the gain is 8e/ADU or 16e/ADU as I have seen on some cameras.

What camera are you using?  It might then be possible to give a specific ISO recommendation that both avoids severe read noise and severe quantisation.

Mark