# Pixel density - can the playing field be leveled???

Started Jun 6, 2009 | Discussions thread

Steen Bay wrote:

John Sheehy wrote:

Steen Bay wrote:

Is it correct that a hypothetical 20mp G11 would have 1 stop higher per pixel (and 'image level') noise and 1 stop less DR than a 10mp G11, assuming a 'perfect scaling' of the read noise? And if the read noise isn't scaled, if the 20mp camera has the same per pixel read noise as the 10mp camera, won't the difference in DR and in the shadow/high-ISO noise then be larger than 1 stop?

Define "read noise"; IOW, what are the units?

If read noise is a value relative to saturation or metered gray, in ADUs, then "the same" pixel level read noise will decrease image read noise to 71%. If we are measuring in electrons, then maintaining read noise at the pixel level will increase read noise by 41%.

By a 'perfect scaling' I mean scaling the read noise with the pixel size like Emil described above. If for example the 10mp camera's read noise was 7,1 electrons, then the 20mp camera's per pixel read noise should be 5 electrons, in order to maintain its relative value, compared to the shot/photon noise. And by 'the same read noise' I mean the same absolute level of of read noise, like for example 7,1 electrons for both cameras. (And although both the scaled per pixel read noise and the shot noise would be lower in absolute values, then its level, measured in percent of the signal would be 1 stop (41%) higher, resulting in 1 stop lower DR for the 20mp camera)

That's not the same 'absolute level' of read noise. Read noise does not occur in electrons, it is a voltage noise, superimposed on the voltage signal that is produced from the charge in the pixel, so the same 'absolute level' of read noise is the same voltage, not the same number of electrons. You can see why this is true if we concentrate on the late stage readnoises, the ADC noise. If you keep the same ADC but change the sensor to one with different size pixels, obviously the absolute ADC noise will stay the same, however, when measured in 'electrons' this read noise, compared to the input, will change.

The mechanism that translates the charge to electrons is actually the capacitance of the source follower transistor (such that V=Q/C). As you can see, as the capacitance gets smaller 'one electron' will be represented by a progressively larger voltage. Conversely, as the capacitance gets smaller, a fixed 'absolute' read noise represents fewer electrons. Since the capacitance scales with the area of the transistor, if we assume that when we scale a pixel, we scale everything, then the capacitance, and thus the per-pixel read noise scales with the square of the pixel spacing. Given that the noise density for equal electron read noise scales with the inverse of the pixel spacing, the noise density contribution of a fixed read noise scales as the pixel spacing.

A read noise fixed in electrons would be quite artificial, it would mean that designers would have to increase the electronic noise in order to compensate for the increased charge/voltage conversion factor of the smaller pixel. The only case it would occur would be in two sensels of different sizes which shared the same source follower and read elecronics design with the bigger pixel.

-- hide signature --

Mal

Complain
Post ()
Keyboard shortcuts: