D300 (and D3) Read Noise
Prepared 2008-03-04 (149/117423) by Bill Claff
With the right test images and procedure (see Read Noise on
my site) it is not too difficult to measure Read Noise (RN).
Read noise is an important, sometimes dominant, component of noise in the dark
areas of an image.
Here is a table of measured values (in ADUs):
ISO |
D300_12 |
D300_14 |
D3_12 |
D3_14 |
100 |
1.74 |
6.66 |
|
|
200 |
2.12 |
8.13 |
1.27 |
4.73 |
400 |
3.97 |
15.22 |
1.90 |
6.86 |
800 |
7.59 |
28.66 |
3.01 |
11.85 |
1600 |
14.39 |
54.12 |
5.85 |
20.25 |
3200 |
16.10 |
60.32 |
11.35 |
38.84 |
6400 |
|
59.96 |
21.23 |
78.07 |
12800 |
|
|
22.03 |
|
25600 |
|
|
21.68 |
|
Note: A few non‑essential values are missing.
The following chart shows these values as a function of ISO:
Shown as a log-log chart for clarity.
The open symbols represent the numbered ISO values while the closed symbols
represent to "Lo" and "Hi" values.
The dotted lines are curve fits.
The equations and R2 are shown to illustrate how close the fit is.
Note that the "Lo" and "Hi" values do not lie on the dotted
lines.
The "Lo" values are slightly below the lowest numbered ISO value, and
the "Hi" values are about the same as the highest numbered ISO value.
The behavior of the "Lo" values is due to the gain values discussed
in D300 (and D3) Gain.
The behavior of the "Hi" values is due to the fact that these ISO
equivalents result from post-ADC multiplication rather than pre-ADC
amplification.
It's also interesting to note that for every camera for which I have data the
dotted curve fit is linear except for the D3.
I believe this implies some sort of two-stage amplification; presumably to
reduce noise.