[font color=\'#000000\']
The common estimate is that difraction reduces the MTF to zero at 1600/f and to 50% at 800/f. ... The Nyquist frequency is thus 69 lp/mm. But dSLRs show a limiting resolution more on the order of 2/3 the Nyquist frequency, or 46 lp/mm.
This MTF50 is a more refined approach than mine. For f/16, I get 50% MTF at 50lp/mm, and the 7 micron pixels also give 50% MTF at about 50lp/mm. The combined effects involve products of MTF factors, so at f/16 and 50lp/mm, the MTF is 50% of 50%, or 25%.
Thus, at f/16 you have lost a very noticable one stop of contrast relative to what the sensor alone is capable of at low aperture ratios. That suggests that visible degradation of fine detail will start somewhat before f/16, but I cannot tell from this how noticable it will be at f/7.
Even with that degradation, the resolution is probably comparable to about 11MP with no diffraction limitation, so still quite usable!
In fact, once processors get fast enough, I envision the combination of very small pixels and diffraction for low pass filtering, preventing moire and such without the need for the expensive added AA filter. For higher ISO settings, downsampling to a lower pixel count could then be used to restore about the same S/N ratio as using that lower number of photosites to start with.
This would be similar to over-sampling in digital audio.
P. S. I just checked Norm Koren's figures, and he is more optimistic about the 50% MTF threshold, giving about 65lp/mm for 7 micron pixel pitch. The aperture matchng that is about f/12, so it seems that at f/12 (rather than f/16), you have lost half the MTF that you get at very low apertures, out at 65lp/mm. I feel better about the idea that a small visible loss of resolution starts as one goes above f/7.
Interestingly, Norm Koren rates the Canon 24-70 at about 61lp/mm 50% MTF at f/8 (its optimal aperture?), so all the resolution numbers come together to suggest that 16MP is close to the useful limits of 35mm frame format.
Indeed, theoretical arguments like this some years ago lead to the suggestion that 16MP is about the useful limit for 24x36mm frame format (except by using good primes only, or for doing my "oversampling" trick).
Will Canon now settle into improving in other directions, like dynamic range?[/font]