Wavelengths of light vs pixel density

[walter.sk]

A friend had told me that he read that the maximum number of pixels on a camera sensor would be somewhere in the 22mp range on a full frame sensor because once the distance between sensor sites became less than the shortest wavelength of visible light there would be no way to capture that light, presumably about 400nm.  When Nikon came out with the 36mp cameras this did not seem to be a problem

On a full frame sensor, what would the limit of megapixels be to still capture the whole visible range of light?

[Wayne Fox]

Some quick research it seems light, 400 to 700nm converts to about .4 to .7 micron range, considering the d800 is sensel 4.8 microns in size I think your friend may have had a decimal point in the wrong place when coming to the conclusions.

Certainly camera phones sensels are far smaller than the d800 as are some point and shoots ...

[ejmartin]

A friend had told me that he read that the maximum number of pixels on a camera sensor would be somewhere in the 22mp range on a full frame sensor because once the distance between sensor sites became less than the shortest wavelength of visible light there would be no way to capture that light, presumably about 400nm.  When Nikon came out with the 36mp cameras this did not seem to be a problem

On a full frame sensor, what would the limit of megapixels be to still capture the whole visible range of light?

The premise of the question is in error.  Consider radio waves, another part of the electromagnetic spectrum (as is visible light).  AM radio waves have wavelengths of hundreds of meters, and yet your AM radio antenna need not be that big to detect the signal.  The detector need not be larger than the wavelength being detected, otherwise individual atoms could not absorb much of the electromagnetic spectrum (yet they do).

[walter.sk]

Thanks for the replies.  I'll inform my friend, both of the math error and the conceptual error.

[BJL]

Emil,
The AM radio antenna has a slightly easier job, since it does not have to measure the location of the photon, just detect that some sufficient part of its probability amplitude distribution overlapped with the antenna. On the other hand, another example of tiny photosites is film, where light detection is done by the intera tion of a photon with a small cluster of silver halide crystals, and AFAIK, those clusters are far smaller than the wavelength of visible light.

Anyway, one clear experimental fact is that the Nokia 808 with its 1.4 micron sensels and f/2.4 lens gives fairly sharp images at 100% pixel viewing, which shows that with a large enough aperture, resolution down to about 1.4 microns is possible ... and that would translate to 432MP in 35mm format.


P. S. It is interesting how many myths and dodgy arguments keep popping up to rationalize some current popular resolution level as the natural limit, like multiple arguments that going beyond Canon's current 22MP limit only makes sense in formats larger than 35mm, due to diffraction or whatever. I remember that back in the day of 6MP in EF-S and DX format, it was already claimed by some that "they've gone about as far as far as they can go" in that format, so that any future progress required "FULL FRAME". (Such people usually used all caps for those holy words.)

[ejmartin]

Emil,
The AM radio antenna has a slightly easier job, since it does not have to measure the location of the photon, just detect that some sufficient part of its probability amplitude distribution overlapped with the antenna. On the other hand, another example of tiny photosites is film, where light detection is done by the interaction of a photon with a small cluster of silver halide crystals, and AFAIK, those clusters are far smaller than the wavelength of visible light.

Indeed.  When the photosites get smaller than the probability distribution along the focal plane, as for instance when one is in a diffraction limited regime, the photosites don't stop detecting photons, they just don't localize them very well, or rather they do localize them, but that localization doesn't convey additional information about the image (ie no resolution finer than the width of the probability distribution -- the 'spot size').

[Wayne Fox]

So while my math was possibly correct, thus perhaps debunking the theory, it seems those that know far more have debunked the premise itself to a point.

Double debunk!

So now we can all look forward to that 400+mp sensor someday in the Canon 5dxMark9 or Nikon 9x - 

[Schewe]

So now we can all look forward to that 400+mp sensor someday in the Canon 5dxMark9 or Nikon 9x - 

Or Phase One IQ 400?

[Wayne Fox]

On the other hand, another example of tiny photosites is film, where light detection is done by the intera tion of a photon with a small cluster of silver halide crystals, and AFAIK, those clusters are far smaller than the wavelength of visible light.

Just curious here, this old article, (which is interesting as it is from Kodak of old and discusses at what pixel densities digital cameras will equal various sizes of film, all of which have been far exceeded) says the crystals were between .5 to 3 micrometers which I believe is the correct technical name for microns.  This would imply they were about the same size or slightly larger than the wavelength of light?  Not that it matters, and not trying to correct or anything,  just trying to make sure my trivia knowledge is correct .. . and I certainly may be missing something here.

[Wayne Fox]

Or Phase One IQ 400?

yeah, now we're talkin

[hjulenissen]

Emil,
The AM radio antenna has a slightly easier job, since it does not have to measure the location of the photon, just detect that some sufficient part of its probability amplitude distribution overlapped with the antenna.

Antenna arrays are used to create an "image" similar to how an image sensor works. However, they tend to transmit/receive coherent signals, where both amplitude and phase matters. Image sensors tends to tell us only the signal power.

Also, elements in such arrays may be closer to a "point sampler", while an image sensel is closer to a square integration (olpf and diffraction leads to further blurring).

I believe that wavelength/2 is a good spot to "be" for such an array, although you may sample less dense/make the main-lobe sharper by introducing assumptions about your sources (such as approximately plane-waves).

-h