ejmartin
Offline
• •
Upload & Sell: Off
|
Tentacle wrote:
Sorry, wrong. Per pixel (or rather, per photosite) you collect a certain amount of photons, which create an electric charge. This charge is, in the case of CMOS, converted into a voltage. This voltage is directly proportional to the charge, and thus the collected photons. The bigger the photosite, the bigger the resulting voltage that a given photon density creates.
Higher voltage output per photosite means you need less gain/amplification before the ADC can work on it. Smaller photosites need stronger amplification and, unfortunately, the noise gets amplified too.
Noise ties in with DR at the dark end of the range. A higher noise floor (as result of needing more amplification) means you loose DR. At the bright end of the dynamic range you are limited by the capacity of the photosite. A bigger photosite can collect more electrons before it saturates.
DR "at the image level" is nonsense, there is no such thing. It depends on what each photosite can register, that's not influenced by the rest of the image. (Unlike noise, which appearance is influenced by other pixels.)...Show more →
Well, let's look at the facts. The 40D measures at 1.55 electrons/raw level at ISO 200; the 1D3 has 2.5 electrons/raw level at ISO 200 (this is the ISO of maximum DR for both cameras). To measure the number of photoelectrons absorbed per unit area, one should divide by the pixel areas 7.2^2 for the 1D3 and 5.7^2 for the 40D; then the 1D3 collects .048 electrons per raw level per square micron, and the 40D collects .048 electrons per raw level per square micron -- they are equally efficient at collecting light *per unit area*. So if Canon made an APS-H size sensor with the pixel density of the 40D it would have 16MP and be just as sensitive to light as the 10MP 1D3. It's just that the light collection is divided into smaller parcels, whose only effect is more resolution; the last time I checked that was a good thing, apart from the issues of overhead in terms of processor speed and storage needs.
If we look at the noise floor, read noise at ISO 200 on the 1D3 is 12.3 electrons per pixel, for the 40D 9.6 electrons per pixel. Noises scale as the pixel pitch rather than the pixel area, so referred to a per area basis the 40D read noise scales to 9.6 * 7.2/5.7 = 12.1 when referred to the size of the 1D3 pixels.
So you see, at low ISO a sensor with the 40D pixel pitch but APS-H format would have the same dynamic range on a per area basis (which is what I meant when I wrote "at the image level") as the 1D3, and 25% more resolution.
It is the per area basis which is the fair comparison of sensors of different pixel pitch. If you print at a fixed size two images from two sensors with the same format but different pixel densities, a given feature on the image comes from a fixed *area* on the sensor, and is rendered from whatever photons are collected from that fixed area on the sensor, regardless of how many pixels that area contains.
So to compare properly one should consider sensor performance on a per area basis. For the 1D3 and a same size sensor with 40D pixel density, *at the image level*, the same number of photons are collected to render each object in the scene. *At the image level*, the read noise is the same. *At the image level*, the dynamic ranges are the same.
As I see it, the main drawback of higher pixel density is high ISO performance, not low ISO dynamic range. Thus far, Canon has been unable to get read noise at high ISO below 4 electrons/pixel; it always seems to saturate there for all recent models (since the 20D). Making pixels smaller but keeping the read noise per pixel the same raises the high ISO noise floor in inverse proportion to the pixel pitch.
Edited on Mar 11, 2008 at 10:07 AM
|
