D3 dynamic range?

Started Nov 10, 2007 | Discussions
dave steinberg Regular Member • Posts: 315
D3 dynamic range?

Have any D3 users got a feeling for the dynamic range? Up till now, I have stayed with my fuji S5 because of it's excellent DR.

Dave

JohnnyRX7 Senior Member • Posts: 1,853
Re: D3 dynamic range?

dave steinberg wrote:

Have any D3 users got a feeling for the dynamic range? Up till now,
I have stayed with my fuji S5 because of it's excellent DR.

Dave

I think 14 bit will give maybe +1 in DR to the D3 meaning it will probably be around 9-10 .. Fuji is what... 12.. I think the fuji's still have a stop or 2 on the competition.
--
Johnny

Jarkko Haarla Jr Senior Member • Posts: 1,055
14 bit

You think wrong.

The A/D bit depth has no effect on the dynamic range of the sensor. If a system can handle 50 lux to 50000 lux in one exposure without the blacks blocking or the light greys burning, it can do it undepending on whether you slice the 50-50000 lux scale into 100 or 100000 slices.

The Fuji Super CCD has a larger DR as it uses two sets of photosites for one pixel; a sensitive one for the 50 lux end and a less sensitive for the 50000 lux end of the exposure. So, practically having two systems in one they widen the DR as neither end of the range gets saturated or dead so easily.

The newer sensors with lower noise performance probably makes it possible for the electronics to fish more details from the dark end of the range.

50 and 50000 lux here only as example values.

Bill Janes Senior Member • Posts: 1,925
Re: 14 bit--Wrong!

Jarkko Haarla JR wrote:

You think wrong.

The A/D bit depth has no effect on the dynamic range of the sensor.
If a system can handle 50 lux to 50000 lux in one exposure without
the blacks blocking or the light greys burning, it can do it
undepending on whether you slice the 50-50000 lux scale into 100 or
100000 slices.

The ADC does have an effect on the dynamic range. DR can be limited by quantization or noise. In the first case, digital capture is linear and the scale is linear. Therefore, 14 bits can represent values in the range of 1..16384 and for 12 bits the range is 1..4096. These values correspond to 14 and 12 f/stops respectively.

In most practical situations DR is limited by noise and electronics engineers define DR as:
full well capacity/read noise, both expressed in electrons.

Read noise can be affected by the ADC (analog to digital converter). The ideal SNR of an ADC equals 6.02N+1.76 dB, where N is the number of bits. Going from an ADC of 12 bits to 14 bits can give 2 additional stops of DR. Of course, using a 14 bit ADC in front of a sensor with a low DR will not help, since the DR is limited by the sensor in this case rather than the ADC.

-- hide signature --

Bill Janes

Michel Jarry
Michel Jarry Contributing Member • Posts: 573
In theory, you're right but...

...pratically, you will not get a noticeable gain in DR because of the marginal gain consequent to luminesacence values being expressed with 14 rather than 12 bits.

The practical and perceivable difference comes from having larger wells (photon collectors) whatever bit deph is used to express the wider range of their luminescence values.
--
M. J.

D2x, SB-800, 12-24DX, 17-55DX, 70-200VR, 105 Macro, R1C1 Macro Flashes, TC-17, NPS Member

 Michel Jarry's gear list:Michel Jarry's gear list
Nikon Z7 Nikon AF-S Nikkor 70-200mm F4G ED VR Tamron SP 150-600mm F5-6.3 Di VC USD G2 Nikon Z 35mm F1.8 Nikon Z 14-30mm F4 +2 more
Jarkko Haarla Jr Senior Member • Posts: 1,055
Re: 14 bit

heheh...

there are numerous posts in this forum about bit depth vs. dynamic range. It wouldn´t have been any issue for a camera manufacturer to make a 32 bit procesing unit - thus getting 32 stops of DR?

In the only USD 250.000 imaging systems we sell, so far there has been 0.0 stop difference in DR between going 8 bits, 10 bits or 12 bits with a camera. This is about having thinner or thicker slices, but the cake is still the same. Black is 0%.0%.0% and white is 100%.100%.100% undepending on how you convert it.
ADC has no influence on the null and saturation ranges of a sensor.

The 12 bit D2x doesn´t make 10 stops and certainly would not make 10 stops with 14 bits either.

fjp Veteran Member • Posts: 5,145
Re: 14 bit--Wrong!

Bill Janes wrote:

The ADC does have an effect on the dynamic range. DR can be limited
by quantization or noise. In the first case, digital capture is
linear and the scale is linear. Therefore, 14 bits can represent
values in the range of 1..16384 and for 12 bits the range is 1..4096.
These values correspond to 14 and 12 f/stops respectively.

In most practical situations DR is limited by noise and electronics
engineers define DR as:
full well capacity/read noise, both expressed in electrons.

Read noise can be affected by the ADC (analog to digital converter).
The ideal SNR of an ADC equals 6.02N+1.76 dB, where N is the number
of bits. Going from an ADC of 12 bits to 14 bits can give 2
additional stops of DR. Of course, using a 14 bit ADC in front of a
sensor with a low DR will not help, since the DR is limited by the
sensor in this case rather than the ADC.

This is just a bunch of math gobbledegook, a smokescreen to cover up fuzzy thinking, that has nothing to do with Jarkko's argument. However you slice the dynamic range of the sensor, it doesn't change its dynamic range. Jarkko's explanation is sound, and I might add, clear, unlike whatever it is you were saying in your post.

-- hide signature --

http://www.pbase.com/fjp
FJP, Software Developer

bobn2
bobn2 Forum Pro • Posts: 66,297
Re: 14 bit--Wrong!

fjp wrote:

This is just a bunch of math gobbledegook, a smokescreen to cover up
fuzzy thinking, that has nothing to do with Jarkko's argument.
However you slice the dynamic range of the sensor, it doesn't change
its dynamic range. Jarkko's explanation is sound, and I might add,
clear, unlike whatever it is you were saying in your post.

No, he's correct. However, a Jarkko points out, current sensors don't have the DR to make use of even a 12 bit A/D, so it's the sensor DR which is currently the limit. The DR of the system includes the DR of the sensor and the DR of the A/D. When sensors get better, the A/D DR will be significant, and it may be significant with the D3, if that shows 1 or 2 stops extra DR. The current fad for 14bit A/D came because the latest generation of Analog Front End chips are 14 bit, presumably in preparation for better sensors.
--
Bob

ejmartin Veteran Member • Posts: 6,274
Crude DR measurements

I just had a look at the raw samples posted here:

http://aaronlinsdau.com/gear/articles/d3.html

The latest build of the raw converter dcraw can understand D3 raw files; analyzing a dark patch in the upper right of the ISO200 file, the standard deviation of the raw levels in the deepest shadows appears to be about 5 to 6 raw levels. Could be a little higher, since Nikon clips their blackpoint and so the noise is not a Gaussian distribution but a half-gaussian. The raw levels clip at 16383, and so the dynamic range seems to be about 16383/5.5=2980, or about 11.5 stops.

This is a very rough measurement. A proper measurement would use a blackframe image, shot at a high shutter speed with a lens cap on the body. That will take a lot of sloppiness out of the above analysis, but I would be surprised if the dynamic range is more than 12 stops. A similar analysis at ISO 1600 yields about 10 stops DR. Thus it would appear that the D3's dynamic range is about the same as Canon 1 series DSLR's.
--
emil
--

http://theory.uchicago.edu/~ejm/pix/20d/

Bill Janes Senior Member • Posts: 1,925
Re: infinite DR?

Jarkko Haarla JR wrote:

heheh...
there are numerous posts in this forum about bit depth vs. dynamic
range. It wouldn´t have been any issue for a camera manufacturer to
make a 32 bit procesing unit - thus getting 32 stops of DR?

In the only USD 250.000 imaging systems we sell, so far there has
been 0.0 stop difference in DR between going 8 bits, 10 bits or 12
bits with a camera. This is about having thinner or thicker slices,
but the cake is still the same. Black is 0%.0%.0% and white is
100%.100%.100% undepending on how you convert it.
ADC has no influence on the null and saturation ranges of a sensor.

You did not read my post carefully enough. The ADC bit depth does not determine the DR of the system, but a low bit depth can limit the system. BTW, are you using European notation? $2.5*10^5? Tell us a but about your systems.

Just to show how fuzzy your thinking is, you just reported an infinite dynamic range: 100% 0% = ∞

Bill Janes

Bill Janes Senior Member • Posts: 1,925
Re: Crude DR measurements

ejmartin wrote:

I just had a look at the raw samples posted here:

http://aaronlinsdau.com/gear/articles/d3.html

The latest build of the raw converter dcraw can understand D3 raw
files; analyzing a dark patch in the upper right of the ISO200 file,
the standard deviation of the raw levels in the deepest shadows
appears to be about 5 to 6 raw levels. Could be a little higher,
since Nikon clips their blackpoint and so the noise is not a Gaussian
distribution but a half-gaussian. The raw levels clip at 16383, and
so the dynamic range seems to be about 16383/5.5=2980, or about 11.5
stops.

This is a very rough measurement. A proper measurement would use a
blackframe image, shot at a high shutter speed with a lens cap on the
body. That will take a lot of sloppiness out of the above analysis,
but I would be surprised if the dynamic range is more than 12 stops.
A similar analysis at ISO 1600 yields about 10 stops DR. Thus it
would appear that the D3's dynamic range is about the same as Canon 1
series DSLR's.
--
emil
--

http://theory.uchicago.edu/~ejm/pix/20d/

Emil,

A good attempt at scientific analysis according to the engineering definition of DR: full well capacity/read noise, both expressed in electrons. The ADC data number is proportional to electrons, so the method seems valid.

However, as you note, you do not have a black frame image, and flare light from the lens or from internal reflections in the camera box could have a marked effect on the black point.

A small point, but isn't the distribution is the extreme shadows Poisson rather than Gaussian?
--
Bill Janes

ddt332 Contributing Member • Posts: 760
Re: Crude DR measurements

If D3 has a DR around 11 EV, that would be my dream come true! I would certainly return my Fuji S5 and just wait for my D3.

However, I seriously doubt that D3 can has such a high DR. I would expect 9~10 EV will be reasonable. That's why I also ordered a S5.

ejmartin Veteran Member • Posts: 6,274
Re: Crude DR measurements

Bill Janes wrote:

A good attempt at scientific analysis according to the engineering
definition of DR: full well capacity/read noise, both expressed in
electrons. The ADC data number is proportional to electrons, so the
method seems valid.

Yeah, I'm aware of the definition, and the shortcomings of the file I used for the analysis. Until someone gets a D3 and does controlled tests we won't have a proper measurement of the DR of the new Nikons. But I thought I'd give it a try; I believe one can regard my measurement as a lower bound on the DR.

However, as you note, you do not have a black frame image, and flare
light from the lens or from internal reflections in the camera box
could have a marked effect on the black point.

Indeed these might have an effect, such that the true noise is a bit less than what I measured.

After writing this post, I did some more sampling of the image I used and got a more accurate measurement of the read noise -- 4.9 in 14-bit raw units, resulting in a DR of 11.7 stops.

It also turns out that the raw files were shot in 12-bit mode on the D3. It will be interesting to see if the readout methodology of the camera differs enough in 14-bit mode to change the DR appreciably. If it is not, then the extra two bits are going to be as useless as they are on the Canon 1DMk3.

A small point, but isn't the distribution is the extreme shadows
Poisson rather than Gaussian?

My understanding (but I'm no expert here and will defer to anyone more qualified) is that read noise is more or less Gaussian; it is on Canon cameras, see for instance

http://www.pages.drexel.edu/~par24/rawhistogram/Mk3Test.html

(I'm not sure what the funny dips are on the sides of the distribution there, but it's roughly gaussian). Canon imparts a bias voltage so that their blackpoint is at some nonzero raw level (1024 on the 14-bit 1D3 and 40D, 128 on most earlier 12-bit cameras); this is good because the deep shadows are not distorted by clipping. In Nikons the blackpoint is set to zero, and the negative voltage side of the noise distribution is clipped, distorting the deepest shadows. In my more accurate second measurement, I evaluated the noise a little above zero exposure at various points, plotted the results and extrapolated to zero exposure. This avoids the problem of clipped black point which underestimates the noise.

As an aside, the photon shot noise that dominates at higher exposure levels is Poisson distributed, but as soon as it's a few hundred photoelectrons it's also for all practical purposes Gaussian, due to the central limit theorem.

-- hide signature --
ejmartin Veteran Member • Posts: 6,274
Re: Crude DR measurements

ddt332 wrote:

If D3 has a DR around 11 EV, that would be my dream come true! I
would certainly return my Fuji S5 and just wait for my D3.
However, I seriously doubt that D3 can has such a high DR. I would
expect 9~10 EV will be reasonable. That's why I also ordered a S5.

Just to clarify, the definition of DR being used here is the engineering definition -- the maximum signal (the point at which the raw data clips in highlights), divided by the noise when there is no signal (the so-called read noise). In practice this is a rather liberal definition, since the lower end of the dynamic range stops when the signal/noise ratio is one. Most photographers wouldn't want to use that part of the shadow end.

For reference, the DR according to the definition I used is pretty much the same as the Canon 1D Mark III.

-- hide signature --
cloudbuster Regular Member • Posts: 456
DR thoughts

An interesting debate, but would it be right to think that whatever the scientific DR of a sensor is, what really matters is the useable range, and how the sensor transitions from the data areas to the clipped

It is no good for many photographers if the shadows are just noisy blocks. If the image lighting stays the same - which it will unless we move into more challenging areas - By increasing bit depth by definition means that the bottom 1/4 of the image will have more data and therefore more image potential. (More bit data in the shadows is the reason behind the expose to the right theory - which works well IMHO)

So even if DR is the same, more shadow detail should be extractable and the transition to 255/255/255 should hopefully be a little smoother.

thx1138
thx1138 Veteran Member • Posts: 7,162
Re: D3 dynamic range?

Johnny Gossett wrote:

dave steinberg wrote:

Have any D3 users got a feeling for the dynamic range? Up till now,
I have stayed with my fuji S5 because of it's excellent DR.

Dave

I think 14 bit will give maybe +1 in DR to the D3 meaning it will
probably be around 9-10 .. Fuji is what... 12.. I think the fuji's
still have a stop or 2 on the competition.
--
Johnny

I think you are pessimistic. The 40D was measured by Phil to be around 9 EV in jpg and 11 EV in RAW. If the D3 can't better that by 1EV at least somethings fishy.

Steve Bingham
Steve Bingham Forum Pro • Posts: 27,590
Who's definition of DR are we using?

My noise tolerance (noise floor) might be quite different from somebody else's. Phil's raw tests have had DR from Canons, Nikons, Sony, and Fuji down around 10.5. Five years ago my S2 had 11.6. See these charts (scroll down). http://dustylens.com/extended_range.htm

My guess, with the FF having a somewhat lower noise floor, might well end up with a DR somewhere between 11 and 12. More importantly, this extended DR would tend to lose less as it climbed up the ISO ladder. Maybe even 10 at ISO 3,200. Now that would be exciting. It would be nice if someone actually tests this.

Years ago, the lowly Fuji S2 out resolved Canon's flagship in resolution at ISO 1600. A little known fact at that time - but shown with tests. I imagine the new Fuji is right up there around 11.5. (I still love my Nikons for a zillion other reasons)
--
Steve Bingham
http://www.dustylens.com
http://www.ghost-town-photography.com

 Steve Bingham's gear list:Steve Bingham's gear list
Nikon Coolpix 990 Nikon D810 Nikon D7200 Nikon AF Nikkor 85mm f/1.8D Fujifilm XF 60mm F2.4 R Macro +32 more
ejmartin Veteran Member • Posts: 6,274
Re: DR thoughts

cloudbuster wrote:

An interesting debate, but would it be right to think that whatever
the scientific DR of a sensor is, what really matters is the useable
range, and how the sensor transitions from the data areas to the
clipped

There are really only a few sensor characteristics that are of importance here. Noise comes in two types for most applications (neglecting thermal noise, which only kicks in for very long exposures): read noise and photon shot noise. The former is a fixed characteristic of the sensor; the latter rises proportional to the square root of the light intensity, with the constant of proportionality being a measure of the efficiency of the pixels in capturing light. Since the shot noise rises with signal, it dominates over the read noise in midtones and highlights; the read noise dominates in shadows. As far as visual characteristics go, the shot noise is much less obnoxious in appearance -- it just looks grainy; the read noise has a component that is patterned which we see as banding.

The amount of read noise (both random and patterned), and the efficiency (in terms of photoelectrons per raw level) determine how much shadow noise there is, and where shot noise takes over from read noise with rising signal, and completely determine the characteristics of the camera and what you can achieve imagewise. There are of course other factors in the imaging pipeline but those are dependent on the programming of the raw converter or jpeg engine, not the intrinsic properties of the raw data.

It is no good for many photographers if the shadows are just noisy
blocks. If the image lighting stays the same - which it will unless
we move into more challenging areas - By increasing bit depth by
definition means that the bottom 1/4 of the image will have more data
and therefore more image potential. (More bit data in the shadows is
the reason behind the expose to the right theory - which works well
IMHO)

This is a common misconception. One exposes to the right because it increases the signal-to-noise ratio, not because it increases the bit depth. You only notice the quantization of signal when the quantization step exceeds the noise level. To demonstrate that I made a little example. I generated a smooth gradient from 128 to 144 on the usual 0-255 scale, to which gaussian noise with a spread of four levels was added. I then truncated the bit depth -- at the top, 8-bit, then 7-bit, etc down to 3-bit; finally I stretched the histogram with a levels adjustment so that the result filled the whole 0-255 scale. Here is the result:

http://theory.uchicago.edu/~ejm/pix/20d/posts/tests/noisegradient4.jpg

You will see that the 8- 7- and 6-bit depth samples are virtually indistinguishable. For these the noise exceeds the quantization step in tonal jumps; it effectively dithers the transitions. You can maybe see a little degradation in the 5-bit sample, and posterization finally crops up at lower bit depth.

The conclusion is you only have more useable data by using higher bit depth when the noise level is sufficiently small. Quantization step smaller than the noise level has no effect on assessing tonal transitions and the extra bits used to record it just waste storage space without adding to image quality. So, a 14-bit file with noise level of four levels is no better than a 12-bit file with noise of magnitude one level.

So even if DR is the same, more shadow detail should be extractable
and the transition to 255/255/255 should hopefully be a little
smoother.

I hope I've cleared up that misconception.

-- hide signature --
Leon Wittwer Forum Pro • Posts: 13,479
Unsure about that...

ejmartin wrote:

(I'm not sure what the funny dips are on the sides of the
distribution there, but it's roughly gaussian). Canon imparts a
bias voltage so that their blackpoint is at some nonzero raw level
(1024 on the 14-bit 1D3 and 40D, 128 on most earlier 12-bit cameras);

If the black point is at 128 on a 12 bit camera, I would expect the dynamic range as measured by Phil and others who use the step wedges, to be no more than 5 stops, that is, 4096/128. Am I missing something?
--
Leon
http://homepage.mac.com/leonwittwer/landscapes.htm

ejmartin Veteran Member • Posts: 6,274
Re: Unsure about that...

Leon Wittwer wrote:

ejmartin wrote:

(I'm not sure what the funny dips are on the sides of the
distribution there, but it's roughly gaussian). Canon imparts a
bias voltage so that their blackpoint is at some nonzero raw level
(1024 on the 14-bit 1D3 and 40D, 128 on most earlier 12-bit cameras);

If the black point is at 128 on a 12 bit camera, I would expect the
dynamic range as measured by Phil and others who use the step wedges,
to be no more than 5 stops, that is, 4096/128. Am I missing
something?
--

Raw converter software sets the black point at 128 rather than 0. Then saturation is at 4095-128=3967, which is 11.95 stops. You lose essentially nothing, but don't distort the noise spectrum at the bottom end. Nikon should do this biasing, it's good engineering.

-- hide signature --
Keyboard shortcuts:
FForum MMy threads