Nikon D810 read noise vs shutter speed

Like most cameras I’ve used, the D810 has an option to have the camera do processing to remove noise a for long exposures. Like most people I know, I leave if off most of the time.

I made a series of dark-field exposures with and without the long-exposure noise reduction:

Long exposure noise reduction off

Long exposure noise reduction off

Long exposure noise reduction on

Long exposure noise reduction on

Not what I expected. First off, there’s something that looks like noise reduction that starts to take place below 1/4 second whether long exposure noise reduction is turned on or off. Second, long exposure noise reduction doesn’t seem to have much effect. But there’s good news, as well. The long exposure noise appears to be quite low.

I’ll have to repeat the tests with a brighter field.

This camera has all kinds of tricks.

Nikon D810 read noise vs ISO setting

I brought the dark-field exposures of the previous post — plus a few more so that I has an exposure every 1/3 stop — into RawDigger, selected the central 90% of the frame, and measured the standard deviation of the read noise. I normalized the results to a full scale count of 16383, and here’s the result:

D810 RN vs ISO SS

The read-noise-limited dynamic range is a commendable 14 stops. The three ISO settings below ISO 64 are a shuck, as usual. You can see that there’s something funny going on between ISO 2000 and ISO 6400 that causes the noise to be lower than you’d think it should be.

Looking at it referred to the input of the variable-gain amplifier:

D810 RN vs ISO SS corr

The read noise actually rises from ISO 5000 to ISO 6400, which is passing strange.

 

Nikon D810 dark field histograms

Here are a series of D810 raw histograms of a dark field at selected in-camera ISO settings. The field I chose was the back of a Sigma 50mm f/1.4 DG lenscap. To make sure it was really dark, I set the lens to f/16 and the exposure time to 1/8000 second. I used single shot mode, which makes a difference on Sony cameras, but has not had an effect on past Nikons. I turned black point subtraction off in RawDigger to make sure we got the whole histogram. I set the camera for 14-bit operation , and used lossless compressed raw. High ISO noise reduction was off.

Comments follow each histogram.

ISO 64

ISO 64

Nice, tight grouping. All 14 bits are present and accounted for.

ISO 125

ISO 125

The dropout in the blue channel is because Nikon uses digital white balance pre-scaling.

ISO 500

ISO 500

You can see that the histogram is Gaussian, and both the red and blue channels show the effects of digital white balance pre-scaling.

ISO 1000

ISO 1000

More of same. Bored already? Just wait.

 

ISO 1250

ISO 1250

At ISO 1250, we start to lose the 14th bit, even in the un-pre-scaled green channel. Note that it hasn’t dropped out completely, though.

ISO 1600

ISO 1600

The incomplete combing in the green channel continues. Now it looks like all three channels are clipping at an ADC count of about 540.

ISO 2000

ISO 2000

The clipping continues, and will until the end of the series. Now we have completely lost the 14th bit.

ISO 2500

ISO 2500

More of same.

ISO 5000

ISO 5000

We’re starting to lose the 13th bit, and the clipping is extreme — even more than we see at higher ISOs.

ISO 8000

ISO 8000

At ISO 8000, the D810 is a 12-bit camera. The clipping level has dropped to about an ADC count of 480.

ISO 12800

ISO 12800

At the highest non-push ISO, we have some strange partial combing. The clipping level has dropped further.

All this indicates more raw processing than I usually see on Nikon cameras.

Nikon D810 self-heating

Before I start seriously testing a camera, I make a series of dark field exposures to see if there are self-heating effects that would compromise results made with rapid series of shutter activations. Of late, it’s not been a problem.

Here’s the standard deviation of each raw channel for a series of exposures of a dark field at 1/8000 second shutter sped, ISO 12800, single shot mode:

d810 self heating

The green dotted line is a least-squares fitted trend line. There is no self-heating in evidence. The reason that the red and blue channels are noisier is that Nikon uses digital prescaling on those channels in this and most if its high-end DSLRs.

Nikon D810 testing

The D810 arrived today. I’ll be testing it over the next few weeks.

Some things that I’m curious about:

  • How does the read noise compare with the D800E? Does the new Sony sensor have some tricks up its silicon sleeve, like the one in the a7S?
  • How about photon noise? Are the full well capacities of the D800 and he D810 the same, or has progress been made?
  • Has the awful, subsampled live view of the D800 been fixed, or at least improved?
  • Will the electronic first curtain shutter on the D810 result in sharper images?
  • Has the auto-focusing been materially improved?
  • Will all the D800 accessories, including the RRS plates, fit properly?
  • Now that the anti-aliasing filter is well and truly gone, instead of half there as in the D800E, are the pictures sharper? Is the aliasing worse?

Stay tuned.

How read and quantizing noise interact

The Sony a7S has raw bit depths of 12 or 13 bits depending on the way the camera is set. It also has low read noise in many circumstances. The combination of the two conditions has lead several people to ask me if, when I think I’m measuring read noise, am I sometimes measuring just the quantizing noise of the ADC.

Good question. I built a little simulator to find out:

qnvsRB Sxript dc var

It models read noise as having a Gaussian distribution, and computes the sum of the read noise and the quantizing noise when the standard deviation of the read noise varies from 1/16 of the least-significant nit (LSB) of the analog-to-digital converter (ADC) to 2 LSBs. Since the quantizing noise in this case is also a function of the average (dc) level of the read noise, the little simulation models that level as varying from 0 to 1 LSB.

Here is the family of curves that the simulation produces:

QNvs RN vasrying dc

 

And here’s a closeup of the area around the origin:

qnvsrnblowup

You can see that the curves are symmetric with respect to  a dc offset of 1/2 LSB. 1/8 LSB and 7/8 LSB produce the same result, as do 1/4 LSB and 3/4 LSB. Same with 3/8 and 5/8.

You can also see that by the time the standard deviation of the read noise gets to 1/2 LSB, the dc offset doesn’t make much difference.

If I change the script slightly to look at the mean of the total noise instead of the standard deviation:

qnvsrnmean script

We get this:

qnvsrn mean

This indicates that, when the read noise standard deviation is over 1/2 LSB, it supplies enough dithering to let the quantizer, on average, resolve signals that are less than one LSB.

 

EFCS vs silent shutter vibration

I’ve already established that electronic first curtain shutter can provide sharper tripod-mounted images with the Sony a7 than with the all-mechanical shutter mode on that camera and the all-mechanical shutter on the a7R. I wondered if using the a7S in all-electronic shutter mode — aka silent shutter — would provide any benefits.

I took the longest lens I have that will take a 77mm filter: the Nikon AFS-Nikkor 80-400mm f/4.5-5.6 G ED, and mounted it to the a7S with a Metabones adapter. Using the rotating foot, I clipped the lens, in landscape orientation into an Arca Swiss C1 head which was attached to a set of RRS TVC-44 legs.

I mounted a Heliopan variable ND filter on the lens. I lit an Imatest SFRPlus target with a Fotodiox 5500K variable output LED source. I set the ISO to 400, focused wide open, and set the aperture to f/11. Using the ND filter and the light together to control the, and made a series of exposures of durations from 1/250 second through 1/8 second at 1 stop intervals. I controlled the number of photons hitting the sensor with the ND filter.

Then I did it again.

I processed the images in Lightroom 5.6 with default settings except for cropping and white balance, then I processed them in Imatest, measuring the  MTF50 for horizontal edges. I converted the Imatest cycles/pixel data to cycles/picture height, and made plots.

s vs efcs a7s 80-400 mtf50

s vs efcs a7s 80-400 mtf50b

Noisy results. It looks like the all electronic shutter helps a little, but not much. If I could have performed the testing with a sharper lens, it might have helped more.

 

Sony a7S silent shutter scanning speed for APS-C images

I measured the scanning speed of the Sony a7S silent (fully electronic) shutter in this post.  The answer was about 1/30 of a second. Someone asked me how fast the electronic shutter scanned in APS-C mode.

Here’s the scope shot:

_DSC3137

And here it is with a white line superimposed so that it’s easier to figure out how many divisions it takes the shutter to go from the top of the image to the bottom:

_DSC3137-Edit

The answer is a little less than 20 milliseconds, or 1/50 of a second.

Sony a7S read noise in silent mode

I received a request to compare the read noise versus ISO setting of the Sony a7S in silent shutter mode to that of single shot electronic first curtain shutter (EFCS) mode. I already had the data (look here ), so all I had to do was put both curves on the same graph:

sfcs vs ss RN

Mostly as expected, but the lower-ISO kink in the two curves occurs in different places. That’s strange. I went back and checked the raw files. No error there.

It looks like there’s a difference in the way the sensor operates in the two modes: going from low conversion gain to high conversion gain on the 1600 to 2000 ISO change in EFCS and on the 1250 to 1600 ISO change in silent shutter mode .