sonyalpharumors

Don’t confuse two things: The dynamic range which comes out of the camera and the “resolution” of the processing space.

The former one is driven by solid state physics: It is difficult to make a sensor more sensitive without increasing at the same time the sensitivity for temperature (aka Noise). What counts is the Signal-to-Noise-Ratio. And this is limited. Thus, as you cannot divide electrons, a higher resolution does not make sense at all, expect for larger numbers.

In the caase of image processing, 16 bit does make a lot of sense in order to avoid rounding/trucation errors (even though, you don’t get a better SNR). An example: Imagine, you want to make the image 10% brighter. A dark value, say 5 would become 5.5 but will remain at 5 due to the discret numbers instead of 5.5, as it would be correct. A value of 128 would becoume 140 instead of 140.8, etc. With each and every operation, the losses due to rouning/truncation becomes larger and larger.. In the end, you are left alone with only a view luminosity values and don’t have a smooth sky anymore.

Now, when you map the 8-bit space to 16 bit, make the calculations there, before you scale down everything again before output, you get: 5 maps to 1280 and 128 maps to 32768. Adding now 10% will give us 1408 and 36044, exact. And even if we would be one count off due to truncation, this would be an error smaller than 0.1%, even at “5″…

Why the sensor of the a580 would be worse than the one from the K-5 and D7000, I don’t know. First question is: How was this determined? Did someone open the camera and put the sensor on a standard testbench? Or was the whole camera (with optics?) measured? In the latter case, you measure the output of the whole system and it’s response to the environment, i.e. incl. readout electronics, environmental influences like temperature, etc. The housing (plastic/metal) can make a difference, too…