[dosdan]

Smeggypants, let's look at altering ISO.

First off, a read of this section of Emil Martinec's noise treatise is useful: Noise, Dynamic Range and Bit Depth in Digital SLRs -- page 2

Assuming that the various noise sources in the camera are not correlated, they are added in quadrature i.e squaring each noise value, adding the squares together and then square-rooting the total.

For a camera with a single-stage PGA (programmable gain amp situated between the sensor & the ADC - used to vary the ISO sensitivity):

R^2 = (G * R0)^2 + R1^2

where:
R = total read noise
G = ISO gain
R0 = noise before the PGA e.g. sensor read noise, thermal noise & PGA input noise
R1 = noise after the PGA e.g. ADC input noise & quantisation noise.

The noise can be expressed relative to the input as equivalent electrons, or output-referenced as either ADU (analogue-to-digital output units) or DN (digital numbers).

As a working example, for the Nikon D3:

R0 = 1.22 x 10^-2 ADU
R1= 4.2 ADU

So at ISO100, total_read_noise^2 = (1.22 x 10^-2 * 100)^2 + 4.2^2
=1.22^2 + 4.2^2 = 1.49 + 17.64 = 19.3.
The total read noise is the square root of that = 4.4 ADU

Now consider ISO400: total_read_noise^2 = (1.22 x 10^-2 * 400)^2 + 4.2^2
= 4.88^2 + 4.2^2 = 23.81 + 17.64 = 41.5.
The square root = 6.4 ADU

And at ISO1600, (1.22 x 10^-2 * 1600)^2 + 4.2^2
= 19.52^2 + 4.2^2 = 381.03 + 17.64 = 398.6
The square root = 20 ADU

Next, let's assume that a signal of interest above this noise floor was at a level of 250 ADU. Ignoring photonic noise, let's look at the effect on the SNR of varying the ISO.

SNR
ISO100 = 250 / 4.4 = 57:1
ISO400 = 1000 / 6.4 = 156:1
ISO1600 = 4000 / 20 = 200:1
ISO6400 = 16000 / 78 = 205:1

As we increase the ISO 4x each time, we are also increasing the signal of interest by the same amount. So relative to ISO100, the 250 ADU (equiv.) signal from the sensor after the ISO400 PGA stage, comes out at an 1000 ADU level. Looking at the SNR, you can see here the benefit of boosting ISO by analogue amplification in the D3. Eventually the amplified combined sensor output & PGA input noise (the first term) dominates the constant ADC input noise (the second term) and the SNR improvements, with increasing ISO boost, flatten out. Also the DR is decreasing with each ISO boost step. A 14-bit ADC, without zero/black offset will top at at about 16,383 ADU. So in this example we could not apply another 4x boost, because we would overload the ADC input. (If we could, at ISO25600 the SNR would still be 205:1).

Photonic noise is ignored in this comparison since it related to original light level, and after conversion to an electrical signal, does not change relative to the signal of interest, because we are not boosting the light level, and hence not increasing the number of photons, when we boost ISO - we are increasing the electrical level after photo-electrical conversion. Quoting from Emil Martinec from the link at the top of this post:

Photon noise, being a property of the light itself, couldn't care less what ISO is set in the camera; only the read noise enters into the question of whether the data amplification is as well done in software post-capture as it is by hardware during image capture.

Consider now the K-5. I'll use Falconeye's derived values. He lists these in this post:

https://www.pentaxforums.com/forums/pentax-k-5-forum/135603-isoless-sensor.html#post1416601

His read noise formula is slightly different from Emil's in that he uses R0 for post-PGA noise & R1 for pre-PGA noise, whereas Emil uses them the other way around. Also Falconeye appears to use the reciprocal of gain compared to the way Emil uses it. I've followed Emil's usage in these matters to match the previous section.

R0= 2.3 e- (noise before the PGA e.g. sensor read noise & PGA input noise)
R1 = 2.5 e- (noise after the PGA e.g. ADC input noise & quantisation noise)

To convert from equivalent electrons to ADU/DN, divide the saturation "Full Well" capacity at base ISO (the effective FWC will be less at higher ISOs because the ADC will now overload well before the sensel has reached saturation) by the number of steps in the ADC. From Sensorgen - digital camera sensor data, the K-5 FWC at base ISO is 47,159 e-. Assuming the number of steps used to reach saturation in the ADC is 16,383, the conversion factor is 47159 / 16383 = 2.9 e-/ADU. (The 14-bit ADC may not reach all 16,383 steps before the sensel saturates at Base ISO, but it's close enough.)

R0 = 2.3 e- / 2.9 e-/ADU / 70 (divide by 70 because the actual base ISO is ISO70 and Emil uses an ISO1 value for specifying the gain) = 0.79/70 ADU = 1.13 x 10^-2 ADU

R1 = 2.5 e- / 2.9 e-/ADU = 0.86 ADU

(If a noise value is converted from output-referenced to input-referenced, remember to divide by any ISO gain applied so that the equiv. input electron value will be correct.)

ISO100 = 1.13^2 + 0.86^2 = 1.28 + 0.74 = 2.02 . The sqrt is 1.42 ADU
ISO400 = 4.52^2 + 0.86^2 = 20.4 + 0.74 = 21.1. The sqrt is 4.60 ADU
ISO1600 = 18.8^2 + 0.86^2 = 327 + 0.74 = 328. The sqrt is 18.1 ADU
ISO6400 = 72.3^2 + 0.86^2 = 5230 + 0.74 = 5231. The sqrt is 72.3 ADU

SNR for an 250 ADU signal of interest is:

ISO100 = 250 / 1.42 = 176:1
ISO400 = 1000 / 4.6 = 217:1
ISO1600 = 4000 / 18.1 = 221:1
ISO6400 = 16000 / 72.3 = 221:1 (actual SNR may be higher due to on-chip NR).

With the k-5 you can see that even at ISO100, the sensor output read noise is greater than ADC input noise and the total read noise plateaus much quicker than with the D3.

Now, if no analogue amplification was applied, and only digital multiplication is used, whether in camera or in the raw converter software, the noise formula changes from

R^2 = (G * R0)^2 + R1^2

to

R^2 = (R0^2 + R1^2) * G^2

So regardless of the digital multiplication applied, the EIN (equiv. input noise) value will remain constant and that's what we see with the D7000 where the EIN between ISO100 and ISO1600 seems constant.

What would be so significant about the D7000 if it's not using analogue amplification at all for ISO adjustment? Well no one's complaining about its performance when operating away from base ISO which, if digital multiplication/amplification is actually being used, could just as well be performed in raw converter software afterwards. And in these situations where digital multiplication is used, it will not be operating as ETTR as it's known, regardless of what the histogram is showing, as the digital gain is added after the ADC stage, and yet the camera seems to take good pictures. This would be a significant proof that fixed base-ISO operation is possible. Which is a stepping stone to a full "ISOless" implementation.

As has been argued in the posts in this "ISOless" thread (actually a thread about operating at fixed base sensitivity), the concept of ISO sensitivity becomes less meaningful when the total read noise is low at base ISO. Furthermore, with the total read noise, when ADC input noise is less than the sensor read noise, because analogue gain before the ADC provides less of a SNR benefit. Also, as long as ADC is dithered sufficiently and does not have a problem with linearity towards the LSB end, it will not make a big difference on the output result.

What it does require, for convenience, is a "gain" (EV boost) metatag in the raw file and embedded JPEG preview so that the the LCD backscreen display and the initial view shown in raw converter software or in other "default" raw conversion situations, are in the ball-park EV boost/comp region as determined by the camera's metering.

When read noise drops further, the advantage of a fully-implemented "ISOless" approach will be much greater headroom in situations where you'd would normally use higher ISO on a conventional camera.

Dan.

---

Last edited by dosdan; 10-21-2011 at 11:02 PM.

[simico]

This theoretical "ISOless sensor" discussion is entertaining to some degree (well, not those ugly equations), but I can't really see a practical benefit for the photographer. I don't really get the "preserve highlights" thing either - meter and set exposure carefully, then you won't have blown highlights regardless of ISO.

Anyways, approx 1 year ago I snapped two shots with my K-m to see how much difference there is between shots taken the regular way (using ISO1600 for the pre-set aperture and shutter) and intentional underexposure (using ISO100 with the same aperture and shutter as for the ISO1600 shot) & pushing it in PP. There wasn't much of a difference at 100% pixel peeping. Yesterday I tried the same with K-5 and the two shots are practically identical noise wise, hardly anyone could tell which was pushed by me in PP and which by the camera gain.
I still prefer the old, regular way though , it's just easier for me to shoot that way.

[Class A]

SNR for an 250 ADU signal of interest is:

ISO100 = 250 / 1.42 = 176:1
ISO400 = 1000 / 4.6 = 217:1
ISO1600 = 4000 / 18.1 = 221:1
ISO6400 = 16000 / 72.3 = 221:1 (actual SNR may be higher due to on-chip NR).

...

Now, if no analog amplification was applied, and only digital multiplication is used, whether in camera or in the raw converter software, the noise formula changes from[/SIZE]

R^2 = (G * R0)^2 + R1^2

to

R^2 = (R0^2 + R1^2) * G^2

[FONT=Verdana][SIZE=2]So regardless of the digital multiplication applied, the EIN (equiv. input noise) value will remain constant and that's what we see with the D7000 where the EIN between ISO100 and ISO1600 seems constant.

Another way to look at this is to abstract from a particular signal strength (was 250) and express SNR in terms of ISO:

SNR (G) = S * (G/100) / R (G)

Set S=250 and replace G with values 100-6400 and you'll get Dan's list I quoted above.

So after some algebra

SNR (G)
= S * (G/100) / R (G)

= S / (R(G) / (G/100))

= S / (sqrt((G * R0)^2 + R1^2) / (G/100))

= S / (sqrt((G * R0)^2 / G^2 + R1^2 / G^2) * 100)

= S / (sqrt(R0^2 + (R1/G)^2) * 100)


In the final form, you see that as R1 approaches zero, the influence of G (the ISO setting) becomes insignificant. SNR (G) = S / (R0 * 100) is the plateau approached by increasing the ISO value and/or by letting R1 approach zero.

In other words, when R1 (e.g., ADC input noise) approaches zero, there is no point in raising ISO values anymore. The K-5 has come impressively close.

N.B.: The above illustrates that a constant SNR (i.e., one that does not depend on an ISO gain) can not only be explained by the absence of a variable ISO gain but also by (R1/G0)^2 (with G0 being some constant gain) having become insignificant.

[falconeye]

Sensorgen.info data for Pentax K-5
Sensorgen.info data for Nikon D7000

Consider now the K-5. I'll use Falconeye's derived values. He lists these in this post:
https://www.pentaxforums.com/forums/pentax-k-5-forum/135603-isoless-sensor.html#post1416601

Hi Dan,

interesting contribution to the discussion. Thanks!

The D7000 figures and K-5 figures and remaining uncertainties however are such that the conclusion (D7000 fixed at analog gain around ISO 100) is likely but not failsafe. It is all based on a single electron read-out noise difference.

I prefer a point of view where all of this doesn't matter and it all boild down to one statement:

The K-5/D7000 only add 2-3 electrons noise to the signal whatever be the ISO setting (where a fully saturated signal is around 40000 electrons).

I can't really see a practical benefit for the photographer. I don't really get the "preserve highlights" thing either - meter and set exposure carefully, then you won't have blown highlights regardless of ISO.

I tried to explain the practical benefit for me as a photographer in a couple posts above.

With my K-5, when I see that the scene contrast is high, I dial in -1, -2 or even -3 EV exposure compensation without hesitation. I check that the ISO wasn't upped (it shouldn't because underexposure of course allows for lower ISO in an Auto ISO setup). Even fill flash against the sun isn't mandatory anymore. Just keep the ISO down.

Then, in LR (Raw now is a must), I use exposure, levels and eventually fill light (don't forget to set blacks to zero first) to recover what I would have seen at a higher ISO step -- but without the blown highlights.


This is NOT true: "meter and set exposure carefully, then you won't have blown highlights regardless of ISO". Except if you allow for EV compensation to avoid blown highlights. But then, you already use the sensor in an ISOless style (w/o realizing obviously): if you use -EV compensation then you lower ISO (in Auto ISO or you do so manually if you care about all three settings).

In lesser cameras (than the K-5), when using -EV compensation you had to care about resulting shadow noise (because EV-boosted shadow noise was higher than the normal noise you would have obtained at a higher ISO setting). Not anymore with the K-5.

You may call it ISOless like the thread title says.

I prefer to call it increased dynamic range and increased photographic freedom.

[jbrothman]

I am baffled by the equations on this post but it seemed a natural place to ask an ISO question (especially since every time I start a new thread on this forum someone just points me to another thread and so maybe posting here will preempt that).

I shoot sports and want the flexibility of shooting inside (e.g. volleyball) at high shutter speeds but assumed that I will be forced to suffer with noise using my K20D.

This weekend, at my daughter's bat mitzvah, our photographer was lamenting trading her Canon 5D Mark II (she was using a 5D) because of the high iso on that body. But I said, with high iso, you get high noise, and she said nope, on the 5D Mark II there was practically no noise even at the highest iso settings.

So I came to Pentax forums wondering whether this was true for the K7 or K5 as well, and found this post.

So my question: Do all these equations above mean that I can shoot indoor volleyballers with my F* 300mm f4.5 without noticable noise if I get a K5? How about K7? Or should I wait for the next body that replaces K5?

Thanks in advance.

[Smeggypants]

What would be so significant about the D7000 if it's not using analogue amplification at all for ISO adjustment? Well no one's complaining about its performance when operating away from base ISO which, if digital multiplication/amplification is actually being used, could just as well be performed in raw converter software afterwards. And in these situations where digital multiplication is used, it will not be operating as ETTR as it's known, regardless of what the histogram is showing, as the digital gain is added after the ADC stage, and yet the camera seems to take good pictures. This would be a significant proof that fixed-base ISO operation is possible. Which is a stepping stone to a full "ISOless" implementation.

So you don't know if the D7000 isn't using any analogue gain. Why not just ask Nikon?

What it does require, for convenience, is a "gain" (EV boost) metatag in the raw file and embedded JPEG preview so that the the LCD backscreen display and the initial view shown in raw converter software or other in other "default" raw conversion situations, are in the ball-park EV boost/comp region as determined by the camera's metering.

I certainly wouldn't want to import a shoot of RAWs into Lightroom and have most of the pics 'looking black' and having to adjust exposure for each shot either. LR only allows 4 stops of gain anyway

Regardless of gain being before or after conversion or a combination of both the ISOless concept seems pointless to me. If you want to preserve your highlights then underexpose for those particular shots.

I'm with simico on this.

[dosdan]

I certainly wouldn't want to import a shoot of RAWs into Lightroom and have most of the pics 'looking black' and having to adjust exposure for each shot either. LR only allows 4 stops of gain anyway

If there was a gain/EVboost metatag written in the raw file by the camera and the raw converter understood it, the image in the converter, e.g LR, would not be dark.

Say you shoot in a camera in a fixed base ISO/"ISOless" mode. Say the fixed/base/native ISO is ISO100. Now you take a shoot in a situation that in a normal camera you would use ISO400. So in this fixed-ISO mode you've now taken a shoot that requires a digital multiplication of +2 EV (digitally, a simple two bit shift). The camera records the fact in a raw metatag that this image requires +2 EV boost and displays it correctly when you review the image in the camera.

When you load it in LR, the program reads the metatag and automatically sets the Exposure/EV setting to +2 EV and you see the picture correctly. You do a default conversion from raw to TIF or JPEG - the converter automatically applies +2 EV.

But, if you're not happy with the camera's metering - its choice of EVboost value - you can easily change it in PP. But you don't lose any clipping headroom, unlike in an ETTR ISO400 shot. It would be equivalent of having an extra 2 stops (in this particular case) more highlight recovery range.

If a future version of LR was aware of this metatag, I'm sure the max EV offered would be greater than +4 EV - you can get greater than that now using UFraw or DCraw.

Dan.

---

Last edited by dosdan; 04-06-2011 at 04:51 PM.

[Smeggypants]

And you'd lose shadow footroom. I've successfully recovered 5 stops with the K-5, when inadvwertendly underexposing ( usually becuase flash didn't fire ( ), but there's no more room to boost shadow areas ( fill light ).

It's always better to sample an analogue signal with a higher bit resolution IMO

However when I get time I will do some tests with my K-5 and compare between under exposing and correctly exposing the same scene and see what impact that has on PP.

Did you ask Nikon whether they apply analogue gain between sensor and ADC in the D7000 yet?

[dosdan]

Did you ask Nikon whether they apply analogue gain between sensor and ADC in the D7000 yet?

No, but I've posted this question in the DPR D7000 forum though. Hoping for interesting replies (hopefully without any flaming).

Does the D700 use digital amplification at low to medium ISOs: Nikon D90 - D40 / D7000 - D3000 Forum: Digital Photography Review

Dan.

[garyk]

I don't know squat about this, but i do know you can underexpose and recover nicely in raw. i only shoot raw and it has saved mannnnnny photos. I am getting where i underexpose regularly on difficult subjects. Rather than blow the hignlights on parts of the subject. Very usful indeed. And if you use the camera much at all. This is quite apparent very fast.
This is where pentax forums earn there keep for Pentax shooters. good thread.
Good reading here some times.

[dosdan]

I don't know squat about this, but i do know you can underexpose and recover nicely in raw. i only shoot raw and it has saved mannnnnny photos.

Here's an unintentional ("failed flash") 6-stops push with a K-5:

Noise and DR comparision: Canon 5D vs 5D2 vs 7D vs Pentax K5

Dan.

[simico]

With my K-5, when I see that the scene contrast is high, I dial in -1, -2 or even -3 EV exposure compensation without hesitation.
...
This is NOT true: "meter and set exposure carefully, then you won't have blown highlights regardless of ISO". Except if you allow for EV compensation to avoid blown highlights. But then, you already use the sensor in an ISOless style (w/o realizing obviously): if you use -EV compensation then you lower ISO (in Auto ISO or you do so manually if you care about all three settings).

I do the same if I care about highlights (because sometimes blowing highlights doesn't matter, other times it matters) and this is exactly what I meant with "meter and set exposure carefully, then you won't have blown highlights regardless of ISO". In my book it's the photographer's natural task to override the exposure set by the camera (by using EV comp, manually changing ISO or aperture or shutter, using flash EV comp, etc) to avoid crushed shadows or blown highlights or to put the subject in the desired zone.

In lesser cameras (than the K-5), when using -EV compensation you had to care about resulting shadow noise (because EV-boosted shadow noise was higher than the normal noise you would have obtained at a higher ISO setting). Not anymore with the K-5.

Yes, and this is one of the many things why I "fell in love" with K-5

You may call it ISOless like the thread title says.

I'm under the impression that "ISOless sensor" really means a fixed ISO value and absolutely no additional gain (neither analogue, nor digital) in the camera. That is, a camera with fixed ISO100 and nothing else - so the "holy triangle" of exposure is reduced to aperture and shutter speed only, all the shots are taken at ISO100 regardless of aperture and shutter and light.

Now this is what doesn't give any benefit to the photographer, I think.
Even with today's and tomorrow's brilliant sensors (and NR softwares), an image pushed 6-10 EV always have more noise than one pushed only 2-3 or not pushed at all. I, as a photographer, want to control how much noise I'll get and I want to be able to know that before I even take the shot. By being able to see and set ISO on the camera I have this freedom. With a truly ISOless camera I don't.

[garyk]

Dan, I wasn't expecting anything like that. That is just un real..

I shoot birds and many time underexpose and loose no detail.. Just an amazing camera or sensor...

Good thread by the way.

[Jeff Charles]

Now this is what doesn't give any benefit to the photographer, I think.
Even with today's and tomorrow's brilliant sensors (and NR softwares), an image pushed 6-10 EV always have more noise than one pushed only 2-3 or not pushed at all. I, as a photographer, want to control how much noise I'll get and I want to be able to know that before I even take the shot. By being able to see and set ISO on the camera I have this freedom. With a truly ISOless camera I don't.

An ISOless camera would not preclude what you are asking for. The camera could indicate the amount of gain that its meter determined should be applied, without having to lock that gain into the raw file. I can imaging replacing the ISO value displayed in the viewfinder with a value for Gain. It would serve as a "quality gauge."

Jeff

[Smeggypants]

No, but I've posted this question in the DPR D7000 forum though. Hoping for interesting replies (hopefully without any flaming).

Does the D700 use digital amplification at low to medium ISOs: Nikon D90 - D40 / D7000 - D3000 Forum: Digital Photography Review

Dan.

Cool. Lets hope someone knows