Production D800E Photograhic Dynamic Range and Read Noise

[bclaff]

Thanks to Andy E. of Andy E.'s Photo Blog ( http://nikonandye.wordpress.com ) I have full Nikon D800E Photographic Dynamic Range (PDR) and Read Noise data.

Read noise looks slightly but measurably lower than the D800 with an almost imperceptible increase in PDR.

I suspect that without the AA filter just a little more light is getting to the sensor.

If you haven't looked at the charts recently, I have been trying to add more Canon data and I've added Read Noise in Electrons.

See the updated PDR chart ( http://home.comcast.net/~NikonD70/Charts/PDR.htm#D800E ) and read noise charts ( http://home.comcast.net/~NikonD70/Charts/RN_ADU.htm#D800E_14 and http://home.comcast.net/~NikonD70/Charts/RN_e.htm#D800E_14 )
--
Bill (visit me at http://home.comcast.net/~NikonD70/ )

 

[DigVis]

Read noise looks slightly but measurably lower than the D800 with an almost imperceptible increase in PDR.

Thank you for sharing the data. Would you consider the read noise difference to be significant compared to an expected sample variation. Have you got sufficient data from different cameras of the same model to quantify typical sample variations? Do other factors, such as temperature, significantly affect the read noise figures?

 

[BobYIL]

Thank you for these insightful charts. Your efforts so far have been very valuable for us in satisfying our inquiries.

(Amazing to note what a tiny change in the OLPF can cause some measurable change in read noise but almost nothing in DR...)
--
BobYIL

 

[NektonFi]

Thanks a lot! I am more eagerly waiting for my D800E now, eventhough I doubt I will see any noticeable difference in SNR or dynamic range compared to D800 (which I currently own)

 

[Robin Casady]

Thanks very much for sharing this information.
--
Robin Casady
http://www.robincasady.com/Photo/index.html

 

[rhlpetrus]

Thanks Bill, nice job. It seems the 5D3 uses a lot of analog processing up to ISO1600 in order to keep DR relatively flat, compared to D4, for example. D800/E, as expected, show Sony's traditional linear behavior from base ISO up. If somehow Nikon/Sony were able to delay the high ISO linear decay, they would be able to produce a sensor that would improve high ISO by at least a stop compared to the D4's performance. The architecture likely makes that impossible, Sony would love to have a camera with better high ISO than the D3-D4 line.

Now, what do you think account for the seesaw behavior on the 5D3? Intermediate ISO using digital gain only is a possibility.
--
Renato.
http://www.flickr.com/photos/rhlpedrosa/
OnExposure member
http://www.onexposure.net/

Good shooting and good luck
(after Ed Murrow)

 

[Luke Kaven]

Hi Bill,

I've been surprised by the amount of thermal noise this sensor puts out. Even more when using live view. I wonder if you've had any further insights into that, and into ways to mitigate it?

 

[Iliah Borg]

I suspect that without the AA filter just a little more light is getting to the sensor.

D800E seems to have two "AA" filters canceling each other.

--
http://www.libraw.org/
http://www.RawDigger.com/

 

[NektonFi]

I think you're correct. In fact anything isn't missing from the E compared to the D800. It is just a bit different.

I suspect that without the AA filter just a little more light is getting to the sensor.

D800E seems to have two "AA" filters canceling each other.

--
http://www.libraw.org/
http://www.RawDigger.com/

 

[lakofsth]

I think you're correct. In fact anything isn't missing from the E compared to the D800. It is just a bit different.

Well, the E has an optical flat between the two AA filter layers instead of a wave plate (birefringent 90 degree polarization rotator).

 

[TOF guy]

I suspect that without the AA filter just a little more light is getting to the sensor.

D800E seems to have two "AA" filters canceling each other.

It's plausible that the filter which cancel the action of the other takes a little more light away than one that one which adds blur. But if I'm reading the graphs properly the difference in "Photographic dynamic range" is not insignificant at all: close to 1 stop !(11.44 -10.62 ~ 0.8 at ISO "80") That does seem a bit suspicious.

--
Thierry

 

[TOF guy]

Thank you for your efforts and sharing these results.

In addition to my reply to Iliah in this thread, does it make sense that the read noise in electron (graph read noise in electron vs. ISO) may be different for 2 cameras - D800 and D800e - with the same sensor? Could you add error bars to the graphs so that we get an idea of what the accuracy of the estimates is? And is it reasonable to give read noise with 4 digits (like "4.208")? How do two cameras from the same model (say your D70 and a D70 you've borrowed) compare?
--
Thierry

 

[DSPographer]

[snip]

Now, what do you think account for the seesaw behavior on the 5D3? Intermediate ISO using digital gain only is a possibility.

The 5D mark II plot also shows a seesaw behaviour. If you check the read noise plots they show that the read noise follows the already known analog gain steps that occur at 2X multiples of 126 ISO. So, the seesaw behaviour reflects a difference in measured full scale capacity. This may not be correct. I remember previous 5D2 analysis showed that the intermediate ISO values on either side of the power of 2 multiples of 100 ISO are from digital gain only, and have the same DR as the primary ISO steps.

 

[DSPographer]

Thank you for your efforts and sharing these results.

In addition to my reply to Iliah in this thread, does it make sense that the read noise in electron (graph read noise in electron vs. ISO) may be different for 2 cameras - D800 and D800e - with the same sensor? Could you add error bars to the graphs so that we get an idea of what the accuracy of the estimates is? And is it reasonable to give read noise with 4 digits (like "4.208")? How do two cameras from the same model (say your D70 and a D70 you've borrowed) compare?

Error bars would require an analysis to show the sample to sample as well as environmental variation in the read noise. The primary environmental factor in the read noise is the sensor temperature. So to set error bars, in addition to measuring the variation in read noise between production lots to get an estimate of sample variation, you also need to measure how the read noise varies with the sensor temperature. Measuring the read noise before and after heating the sensor using live view or movie mode would help to get a rough estimate of the sensor temperature effect.

 

[TOF guy]

Error bars would require an analysis to show the sample to sample as well as environmental variation in the read noise. The primary environmental factor in the read noise is the sensor temperature.

Yes that makes sense. Thank you for the explanation.

So now comes to the O/P the following additional questions:

• Do you have measurements of the photographic dynamic range at different temperatures (say ~4°C and ~20°C)?
• Were the D800 / D800e measurements made at temperatures close enough that in light of your estimates above you can tell that the differences are not explained by temperature?
• And still to be answered the sample variations from one camera to another of the same model.

--
Thierry

 

[DSPographer]

[snip]

Now, what do you think account for the seesaw behavior on the 5D3? Intermediate ISO using digital gain only is a possibility.

The 5D mark II plot also shows a seesaw behaviour. If you check the read noise plots they show that the read noise follows the already known analog gain steps that occur at 2X multiples of 126 ISO. So, the seesaw behaviour reflects a difference in measured full scale capacity. This may not be correct. I remember previous 5D2 analysis showed that the intermediate ISO values on either side of the power of 2 multiples of 100 ISO are from digital gain only, and have the same DR as the primary ISO steps.

Sorry, previous analysis showed that the digital gain step below a primary ISO had the same DR as the primary ISO. This is because the saturation was the same and the noise was not affected by the digital gain value below 1 since 14 bits is over-kill for the 5D2 noise levels. The digital gain steps above the primary ISO steps have reduced DR though, since the digital values clip early- reducing the effective saturation value.

 

[bclaff]

Iliah,

D800E seems to have two "AA" filters canceling each other.

Yes, but also the D800E has optical glass for one component as opposed to a wave plate for the D800. The wave plate must absorb more light than optical glass.
In any case, the OLPF assembly is different.

Regards
--
Bill (visit me at http://home.comcast.net/~NikonD70/ )

 

[bclaff]

Thierry

does it make sense that the read noise in electron (graph read noise in electron vs. ISO) may be different for 2 cameras - D800 and D800e - with the same sensor?

Not to me. I don't have hyper-accurate gain figures. I suspect with more accurate gain figures that the noise in electrons would/will agree.

Regards
--
Bill (visit me at http://home.comcast.net/~NikonD70/ )

 

[bclaff]

I remember previous 5D2 analysis showed that the intermediate ISO values on either side of the power of 2 multiples of 100 ISO are from digital gain only,

For Canon cameras if the adjacent intermediate ISO are level in electrons then digital gain has been applied as opposed to the following intermediate ISO showing a slight drop.
For example the 5D Mark III versus the original 5D:

http://home.comcast.net/~NikonD70/Charts/RN_e.htm#EOS%205D_12,EOS%205D%20Mark%20III_14

Regards
--
Bill (visit me at http://home.comcast.net/~NikonD70/ )

 

[bclaff]

The primary environmental factor in the read noise is the sensor temperature.

Agreed. Dark noise is not controlled or reported in my data.

That said, I have multiple data sets and the trend is consistent and significant (in my experience, not statistically).
I think of it as a curiosity with no photographic value that gives us insight.

Regards
--
Bill (visit me at http://home.comcast.net/~NikonD70/ )