On Wed, Sep 06, 2006 at 01:22:43PM +0200, Toralf Lund wrote:
> 
> According to sources on the Net, the current 6MP sensors have a range 
> corresponding to slightly more than 60000 "steps", or somewhat less that 
> the full range of 16 bits. However, the full range is not really usable 
> due to noise; noise essentially means that the output for the same 
> exposure level may vary between several steps next to each other. 
> Apparently, the noise in the sensors used up to now will typically 
> correspond to 10 levels or so, equivalent to 3 or 4 bits. In other 
> words, if you were to use 16 bits, the lower 4 would probably contain 
> little more than random data caused by the noise. This means you are 
> left with a usable range corresponding to 12 bits.
> 
> With a 10MP sensor of the same size, the number of levels should be 
> reduced to something like 35000, meaning that the range even before you 
> consider the noise is closer to 15 bits than 16.

It gets a bit worse than that.  The sensitivity (or range) is proportional
to the area of the sensor site, all other things (ISO, sensor materials,
etc.) being kept the same.  But the noise doesn't increase linearly with
area; because the noise is effectively the sum of several independent
contributions the total noise level increases with the square root of
the area.  Or, to put it another way, when you decrease the size of the
sensor not only does the total sensor range decrease - the drop in the
noise is smaller than the drop in the signal, so the signal-to-noise
ratio also decreases.

That means that if you had, say, a 16-bit sensor with four bits of noise
(giving you 12 bits of signal) then the same technology with the smaller
sensor sites would give a 15-bit sensor with 3.5 bits of noise, or only
11.5 bits of signal.

-- 
PDML Pentax-Discuss Mail List
[email protected]
http://pdml.net/mailman/listinfo/pdml_pdml.net

Reply via email to