Dynamic range, resolution & ISO noise - are smaller pixels best?

[DaveC]

Sean,

I notice in your discussion on Sensor Sizes, you say "the noise gets
enlarged less for the higher pixel count sensor (for a given print size)", hence less apparent noise.

Yet on the other hand, for a lower pixel count on the same size sensor
there is a better signal-to-noise ratio.

So 2 questions (for the same size print, and everything else equal):
1. Is the first effect sufficient to counter the second?
2. If you downrez/average the higher rez images to the smaller rez how
does that compare?

In the real world, I'm trying to decide whether to wait for the d3x or
get the d3, when I like the high ISO performance of the latter. I'm hoping I can have the best of both worlds with the d3x (at 24Mpix)

Dave.

PS So did you get a 5D?

[McQ]

Hi Dave,

See below for some thoughts on this. I'm going to lump these two questions together because they are so interrelated.

I notice in your discussion on Sensor Sizes, you say "the noise gets
enlarged less for the higher pixel count sensor (for a given print size)", hence less apparent noise. Yet on the other hand, for a lower pixel count on the same size sensor there is a better signal-to-noise ratio.

So 2 questions (for the same size print, and everything else equal):
1. Is the first effect sufficient to counter the second?
2. If you downrez/average the higher rez images to the smaller rez how does that compare?

Everything else being equal, the effect of higher noise with smaller pixels is perfectly counterbalanced by the fact that you can always downsize this image (which is equivalent to what happens using the image averaging technique to reduce noise). This can be shown statistically if you assume the noise is entirely random, and that there is no space in between the photosites/pixels.

On the other hand, real world noise is not random and camera sensors also need read-out and other circuitry in addition to their photosites. If, for example, one could not shrink the non-photosite circuitry any smaller than a certain size, then a sensor with more (smaller) pixels wouldn't be able to allocate as high a fraction of the total sensor to photosites as a sensor with fewer (larger) photosites. The higher pixel density sensor would therefore have more noise, even when accounting for downsizing. This is just an example though to illustrate the importance of the manufacturing process, microlenses and other circuitry.

Something I think we are overlooking here though is that even if noise remains more or less the same with smaller (but more) pixels, the camera's dynamic range will certainly suffer. The dynamic range can in this case also be thought of as the signal to noise ratio, which you do point out as being better with the lower pixel density. To me this is often a more important factor than absolute resolution or noise. The only way to get around this would be to always use the HDR technique or generous use of graduated neutral density filters (GNDs). On the other hand, one can never create more resolution from a low pixel count sensor (unless they stitch, but that's another topic entirely).

OVERALL: Smaller pixels are a near draw for noise, a clear loss for dynamic range, but potentially more flexibility for situations where more resolution is useful (when one is willing to use HDR/GNDs)

[McQ]

Oh, and I almost forgot: I ended up getting the 5D, but not based solely on the large vs small sensor size debate. For me the 5D's viewfinder prism (vs pentamirror) with its larger and brighter viewfinder, in addition to its large back LCD screen, were other important factors.