« on: April 21, 2013, 05:25:24 PM »
As far as I understood it so far, the fact that in general FF cameras have lower noise is mainly due to the fact that the sensors are the most recent types that have been developed. I have no experience with most models that are out there, but I would think that newer crop sensors will give lower noise than older FF sensors - simply because they are newer. Any attempt to quantify high-ISO performance simply on size and MP-count alone don't work because no two sensors of different size are the same otherwise.
I am open to hearing that all I have just written is nonsense... But what I would really like to hear is what a real expert has to say. Someone who builds the sensors, or who does research in that field...
Check out the DxOMark ISO 'scores' (the highest ISO than meets a specific noise criterion, so the higher the number, the lower the camera noise). For the newest Canon APS-C camera which they've tested, the T4i/650D, the score is ISO 722. Compare that to the ISO score for the 5D (the original one): ISO 1368. So the 8 year old FF camera with sensor tech that's also 8 years older has nearly a full stop better ISO performance than a very recent APS-C sensor. By comparison, a modern FF sensor (the 5DIII) scores an ISO 2293, close to 2 stops better than its contemporary APS-C sensor.
Here's what one academic expert, Emil Martinec, has to say:
For comparing pixels on a patch of sensor of a given area, the normalizations given above are the appropriate ones. However, one is often interested in comparing the overall noise of the image coming from different sizes of sensors, so noise per area is less relevant that noise normalized to the image frame. Thus another metric for comparison is to consider the entire frame, and if comparing two different sensor formats (such as the 1D3 and 40D considered above), shoot with different focal lengths (say 130mm on the 40D and 160mm on the 1D3) so that the field of view is kept fixed. In the case of fixed framing, we should refer the noise characteristics not to a fixed spatial scale in microns, but rather to a fixed percentage of the frame height. An appropriate figure of merit is to divide the per pixel noise values by the frame height in pixels, again because the noise combines linearly in the number of pixels combined. The clear rule of thumb that emerges from such an exercise (not surprisingly) is that larger size sensor formats are less noisy than smaller size sensor formats. A simple example is the 40D relative to the 1D3; both are 10MP cameras, with 2592 pixels in the frame height, and so in this particular example the frame-referred noise levels are the same as the pixel-level noises, and these are lower (as measured in ADU) at all exposure levels and ISO's for the 1D3. The read noises are similar in photo-electron units, but the 1D3's bigger photosites capture more photons, hence have higher S/N ratio. Bigger sensors have higher S/N ratios, because bigger sensors collect more photons.
As far as I understand this it is quite similar to what I wrote/asked, except that he specifically proposes a way to standardise the measurement. With the last sentence of what you quote he basically says that the noise generated in a single pixel is independent of its size (this was one of the questions I had asked), and thus the S/N gets better if the pixel collects more photons.
Comparing particular frames is one way to do it, and possibly the most reasonable one from a photography point of view. But it is not the only one, and depending on the type of photography, the settings you have available, etc. it is also not necessarily what you need to know, is it?