Jack,
This is a re-reply to your
http://www.dpreview.com/forums/post/52518108
I think that I am getting there. My photography will tell whether or not "I've got it"!
Had a couple of nice days. Managed to sneak away for a bit to capture some of it.
http://www.dpreview.com/forums/post/52542334
This has been a very instructive session for me. Thank you very much.
I am left with two burning questions, though.
- Hsat in the table, measured in Lux-seconds appears to be the exposure (luminance, f/, ss) that will generate full sensors at the appropriate ISO settings. So there must be a relationship between H (in lux seconds) and photons (per area?). Is that relationship available on the internet.
- eDR eludes me. Again, can you please point me out to a link?
Many thanks, for your help!
Tom
Jack Hogan wrote: 3) If the mean number of arriving photons is unchanged, wouldn't photon noise also be unchanged?
Yes. I (re)read
Shot Noise and recalled that Shot Noise is significant with small signals, small numbers of photons being collected, therefore in the shadows. If shot noise is amplified (along with signal) by increasing gain, it becomes significant in the image.
Hi Tom,
The reference for noise in DSCs is this treatise by
Professor Martinec, read it at your leasure.
Working through it. Great stuff!
There you will find that light is statistical in nature and random noise is measured as the standard deviation of the mean signal. Noise inherent in natural light (photon/shot noise) follows poisson statistics, where the standard deviation is the square root of the mean (of the number of photons, the signal). So if the number of photons hitting a sensel (i.e. Exposure) is unchanged, so will its square root, shot noise.
4) If ISO has only to do with processing the captured photoelectrons from the given unchanged Exposure, what happens to the full well count in stops when raising ISO from base to 400? To 800? Refer to the chart of the GX7 above.
Base ISO is said to be ISO 200 (I don't know why.)
ISO 200: FWC= 13501 electrons
ISO 400: FWC= 6663 electrons Diff= 6838
ISO 800: FWC= 3375 electrons Diff= 3288
Increasing the ISO by a stop decreases the FWC by ~1/2 .... which seems to be a pattern all of the way up the gain scale. Does that mean that since the capacity of the well is halved by each stop increase in ISO that a constant gain factor is applied at the sensel level as the ISO is increased stop by stop?
Right. Here is the table again for easy reference.
Recall that Exposure is linear with the number of photons hitting a photosite, which is linear with the number of photoelectrons generated by it, which is linear with the Raw value written to the Raw file. The relationship is:
Raw Value Written (ADU) = NumberofPhotonsArrivingatphotosite x EffectiveQE x CameraGain. So if 20k photons arrive at the photosite during the given Exposure (determined by scene, ss and aperture only), about 2780 electrons (20k*13.9%) will be captured by it. What value will be written to the Raw file?
Preliminary Approximate Values
That depends on the gain applied by the electronics downstream of the sensor, which is controlled by the ISO dial on your camera. If the camera was at base ISO (I believe 125 for the GX7) the inverse of gain would be 4.67 e-/ADU (see table, igain counterintuitive but so by convention). So the Raw value written to the file would be approximately 595 ADU (= 2780e- / 4.67 e-/ADU). If it was at ISO 800, 3390 ADU. If at ISO 1600 it would be clipping at this exposure, since the GX7 is a 12 bit camera and its maximum Raw value is 2^12-1.
OK
The column mis-indicated as FWC represents the number of electrons that will result in clipping for each camera gain. It goes down because as you raise gain you reach clipping sooner, just like in the example above.
Ok, now!
As you know in photography we tend to think in stops: 1 stop = a doubling/halving of the signal (Light/Exposure/number of photons/number of electrons aotbe). For example doubling exposure time from 1/100s to 1/50s will result in 1 additional stop of light aotbe [log2(2)]. Twice the aperture diameter = four times the area = 2 stops more light [log2(4)]. Hence the linear and quadratic relationship of EVs and time and f/# respectively.
Ok
So if at camera ISO125 19126e- result in clipping, but at ISO400 only 6663e- do, that means that by increasing ISO from 125 to 400 the maximum signal we can record was lowered by log2(19126/6663)= 1.52 stops.
OK
If the noise 'floor' has remained unchanged, it follows that DR has dropped by that much.
OK. But it doesn't, as seen in the above table. It only dropped by (11.9-11.3 =) 0.6 stops. So the noise floor has changed by ~.9 stops?
5) So you have lost that many stops at the top end. What has happened at the same time to your engineering eDR?
6) And if it has dropped so much less, doesn't that mean that your deep shadows have improved by the difference?
But has the noise floor remained unchanged when increasing camera ISO from 125 to 400? It hasn't, for reasons that are the subject of another post:
The noise floor is affected by the change in Read Noise in going from ISO125 to ISO400.
Read Noise (that is the noise added by the sensor and camera electronics during capture) has improved from 5e- to 2.7e-, or +0.89 stops at the higher ISO.
5/2.7 = .89stops (this is the improvement in read noise between 125:400 ISO)
125 ISO Read Noise of 5 leads to a DR of 19126/5 = 3825.2 =11.9stops
400 ISO Read Noise of 2.7 leads to a DR of 6663/2.7= 2467.7 = 11.27stops
So raising ISO has resulted in a lower potential maximum signal recordable, but also a lower noise added and hence a lower minimun acceptable signal.
I'm trying to visualize comparing the ISO125 wells with the ISO400 wells:
The
upper level of the well has been DECREASED from 19,126 to 6663 e- (a total of -1.52stops)
The
floor of the well has been ALSO DECREASED LOWERING THE NOISE by 5e- to 2.7e- (a total of +.89stops)
resulting in a net (due to the effects of clipping + effects of decreased Read Noise)
change of 0.63stops.
Low signals that were below the floor of the
ISO125 well are now detectable ... since the floor of the
ISO400 well is (.89stops) lower.
Therefore DR has not dropped by -1.52 stops (looking at clipping only), but by +0.89 stops less, or a total of -0.63 stops as you can see in the relative column*
19126/6663=2.87= -1.52stops due to clipping only
Noise floor has improved by +0.89stops so
The Net Drop in DR is (-1.52 + 0.89) or -0.63stops
[ If the read noise proportion of FWC, (e- ), were to stay constant as ISO changes, the only effect on DR would be the effect of clipping. So there would be no advantage in increasing ISO (from the point of view of SNR, only.) This would be the ISO-less or ISO invariant camera, where changes in ISO don't change read noise. I suppose that the contribution of noise from the analogue amplifier would be minimal, leaving all of the noise generation to the sensel and other electronics independent of the gain amplifiers.]
7) So if by raising ISO that way no desirable highlights were clipped, would your IQ maximizing good self feel compelled by his conscience to do it?
The question is why you would not do it if there are no desirable highlights that would clip: you lose nothing and you have lower noise in the deep shadows.
With a NON ISO-less camera like the GX7:
Sooooooo. Trade-off (as is usual):
IF I am
concerned about highlight clipping I would minimize ISO (to minimize clipping due to gain).
If I am
not concerned about highlight clipping but more about noise in the deep shadows, I would increase ISO to lower the floor of the sensels as low as possible. I would be trading off DR for a low noise floor. If I didn't need DR I would give it up for a lower noise floor.
BUT (subject to motion blur, DOF)
increasing exposure (f/,ss, scene luminance)
up to the point of oversaturation of the sensels by the highlights would lead to optimal exposure.
A no brainer. Of course the answer would be different if the maximum desirable signal were higher, and increasing gain/ISO would cause it to be clipped.
From that moment on, in that exact setting, forget about camera settings and shoot away with abandon concentrating on composition and capturing the moment - confident that you are capturing the best quality information possible from the scene. If the scene conditions change, re-evaluate in light of the new situation.
AND THAT IS THE BEST PART!!!!! (it brings out the lil bit of artiste in this old f@rt engineer's soul!
Jack
* With your GX7, should you be increasing ISO past 800 given the table I posted earlier? Probably not, because the advantage you gain in lower noise is minimal compared to the loss of DR.
Makes sense!
Should you stop at ISO400? In non DR critical situations I would. Note that this last paragraph has nothing to do with Exposure, which a Manual shooter always evaluates and sets independently of ISO.
I always use exposure to be affected by scene luminance, f/, shutter interval. INDEPENDENT OF ISO.
So this last paragraph has to do with GAIN, the amount of boost the ADC is giving to the signal from the sensels. The more in-camera gain, the less gain required in later post raw conversion brightening.
Right?
Right!
So the next question becomes: assuming we start at base ISO with the max possible Exposure given artistic constraints, how high should one raise the ISO as long as no desirable highlights are clipped?
I am visualizing a low light condition where I don't need a lot of DR. I can't open up f/ due to DOF, and I can't lengthen Shutter interval due to motion blur. No flash/lighting. Shooting RAW.
Looking at the histogram I have a big hump on the left and a very important highlight spike that is well left of the right edge.
I leave the exposure as is, in (M)anual mode and gradually increase ISO. That moves the histogram spike to the right, increases the brightness of the EVF. When I hit (say) ISO 1600 the highlight spike hits the edge. I back off to ISO 800, realizing that the "relative to FWC" Read Noise are not that different (2.3e- vs 2.1). So I shoot at ISO800, chimp to make sure that RGB channels are all ok, and take my image home to make a masterpiece in LightRoom.
If I have lots of light and/or need lots of DR I will be shooting at BASE ISO.
If I have lots of light and need more DR I will shoot a bracket using exposure bracketing.
If I have little light (limiting an increase in exposure) and need more DR, I can only bring out the shadows by using ISO bracketing, I think. Haven't ever had the need to do that, probably because motion blur could be better stopped by a tripod.
The answer is as high as the read noise keeps dropping meaningfully (it's for you to decide what meaningful is according to the situation): no more and maybe a little less because you don't want to come home and realize that you should have kept an additional stop of highlights which instead you gave up for a measly 0.2 stops better SNR in the deep shadows - what I think Russel was driving at earlier ;-)
I am pretty sure that I get it now.
Jack
PS. N, Y, Y, -1.5, -2.5, -0.8, -1.6, +0.7, +0.9, Y, heck yeah
*My question in the previous post was about eDR which is more correct, but I realize that's an unneeded complication at this stage. In that case the correct answers are above.
Many thanks, Jack.
Tom
