You may be aware that Optical Black photosites are captured in the raw data for most Nikon DSLRs.
Unlike values in the active area these are not zeroed and so they are a good source of read noise data.
Until recently I had assumed, erroneously, that all Optical Black photosites are created equal.
I came to question my assumption when I was contacted by someone gathering noise data with my program regarding inconsistent results.
This person, M.R. Blume, collaborated with me in resolving the issue and in collecting excellent D700 read noise data.
D700 raw data has four columns of Optical Back data, two to the left of the active area, and two to the right.
I first modified my program to collect Optical Black data by column.
We then discovered that Optical Black columns adjacent to the active area had values that increased when the image area was exposed to more light.
So I conclude that Optical Black photosites are not perfectly shielded from light.
To collect the best read noise data from Optical Black photosites we switched to "lens cap" shots at 1/8000s.
Effectively eliminating signal and minimizing thermal noise.
Even with this column data, there was an effect that is not surprising; the data was bimodal.
Because the even and the odd rows are going through different ADCs with slightly different offsets.
The program was modified again, to collect even and odd rows separately for each Optical Black column.
Now the data in each of the eight buckets is very clean.
The final twist, is that the numbers indicate a two-stage amplification that I have not previously seen in a Nikon DSLR.
Note that the open circles indicate analog gain and the close circles indicate "Lo" and "Hi" ISO equivalents.
Here's the raw data as well:
| ISO | noise |
| 100 | 3.959 |
| 125 | 3.988 |
| 158 | 3.956 |
| 200 | 3.985 |
| 251 | 4.956 |
| 317 | 6.292 |
| 400 | 4.422 |
| 503 | 5.615 |
| 634 | 7.062 |
| 800 | 5.896 |
| 1007 | 7.548 |
| 1269 | 9.210 |
| 1600 | 12.296 |
| 2015 | 15.075 |
| 2539 | 18.643 |
| 3200 | 23.765 |
| 4031 | 30.348 |
| 5079 | 37.813 |
| 6400 | 48.506 |
| 8063 | 30.205 |
| 10159 | 38.100 |
| 12800 | 48.646 |
| 25600 | 48.948 |
:-)
Bill
