D7000 sensible sensor data

Started Dec 12, 2010 | Discussions thread
ForumParentFirstPrevious
Flat view
Marianne Oelund Veteran Member • Posts: 6,632
D7000 sensible sensor data
1

Some of the sensor data which has been circulated recently for the D7000 is, to put it politely, inconsistent. Reports of DR being 14 stops at ISO 100, but then only 12.6 stops at ISO 200, simply do not make sense. As the old adage goes, if you want it done right . . .

Moon Maid's Sensible Sensor Data for the D7000

1. Shot Noise at ISO 100, and Full Well Count

Saturation for the Green channel occurs at (14-bit) DN 15800. Shot noise must be measured with an exposure comfortably below this level, so a DN of 13200 was used. Standard deviation (measured over a 200 x 200 pixel area in the center of the sensor) is 70 DN in the green channel at this level, for a s/n of 189.

Particle (electron) count is (s/n)^2, or 35600 at this exposure. Scaling up to the saturation level, we have 43000 e- for maximum usable well capacity. Red and blue channels are a little lower, coming in just under 40000.

2. Fixed-pattern Noise

Another potentially important contributor to the visible noise content in the mid- to light-tone range is fixed-pattern noise, which results from variations in photosite efficiency and gain. Measured green-channel pattern noise is 0.75% of signal (s/n ratio is 135); it is slightly lower for the red channel, and the blue channel is a bit higher at about 0.85% of signal.

3. Read Noise and Dynamic Range

Read noise is a small contributor and becomes significant at low ISO settings, only if shadows are pushed rather hard. At very high ISO settings though, it can be visible even in mid-tones. Read noise, relative to the maximum signal, establishes the dynamic range of the sensor. It is commonly given in input-referred units (equivalent electrons), but it is also useful to give its output value in DN.

To avoid clipping the read noise, these measurements were made using bias frames, with exposure adjusted to keep the rate of black-clipped pixel values down to about 1%.

Here is the result summary:
@ ISO 100, read noise = 1.4DN = 3.7e-; DR is 13.5 stops
@ ISO 200, read noise = 2.1DN = 2.7e-; DR is 12.9 stops
@ ISO 400, read noise = 3.9DN = 2.5e-; DR is 11.9 stops
@ ISO 800, read noise = 7.4DN = 2.4e-; DR is 11.0 stops
@ ISO 1600, read noise = 15DN = 2.4e-; DR is 10.0 stops
@ ISO 3200, read noise = 30DN = 2.4e-; DR is 9.1 stops
@ ISO 6400, read noise = 59DN = 2.4e-; DR is 8.1 stops
@ ISO H1.0, read noise = 117DN = 2.3e-; DR is 7.1 stops

4. Black Clipping

No black clipping, other than the average 0.5DN which results from A/D converter quantization, was found.

Later posts on this thread will discuss some histograms, and related topics such as converter linearity and implementation of the ISO settings.

ForumParentFirstPrevious
Flat view
Post (hide subjects) Posted by
ForumParentFirstPrevious
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow