Registered: Jun 27, 2007
Total Posts: 19
Country: Germany
First off, I'm not a camera or lens designer, and am thus a bit on a shaky ground here. Take everything I say with a pinch of salt. 
Rob Galbraith has said:
"when the light is especially bright and the temperature is warm, the camera's autofocus performance drops like a stone."
The circumstances described by Rob (good AF performance in overcast or rainy situations, mediocre AF when sunny and warm) indicate that infrared light could be a factor.
Camera lenses are designed under the assumption that the medium filling the space between the lens' last element and the film or sensor plane has a refraction index of one (e.g. air, gases, vacuum...)
Light rays emitted by an object at an "infinite" distance, sitting on the optical axis, will come into focus at the lens' "focal" point.
Light rays emitted by an object closer than "infinite" will come into focus behind that "focal" point.
Thus, the locus of focus points will be distributed along an axis (the Z axis) that is perpendicular to the film plane, depending on the the object's distance. Focusing is basically equivalent to moving the lens back and forth, to bring into exact coincidence the film plane and the focusing plane of the specific subject we're interested in.
In an AF SLR, the lightrays used for focusing go through the semi-transparent main mirror, are reflected by a submirror, converged by lenses, and then split into left and right halves (for horizontal AF sensors) by a splitter prism, and then finally projected onto high-resolution line sensors.
Conceptually, the AF splitter prism takes the locus of focus points distributed along the Z axis and rotates it slightly around a vertical Y axis (which goes e.g. through the camera's flash mount at the top and the tripod socket at the bottom of the camera.)
This rotation, in a sense, rotates and projects any movement along the Z axis as a movement on an horizontal X axis running parallel to the film plane.
The AF unit measures the phase difference -- i.e. distance -- between the correlated symmetrical light patterns projected on the left and right horizontal AF line detectors, and deduces the amount and direction of the focus correction that is needed, if any.
High-end lens designs using e.g. fluorite or UD elements can be made apochromatic, and such lenses are sometimes corrected so well for chromatic aberrations that even near-infrared light and visible light are focused substantially on the same focusing plane.
On the other hand, it's unlikely that the AF unit's convergence lens and splitter prism are apochromatic, as they must be designed to provide a reasonable, average optical fit with the incident light rays coming from any lens that could be mounted on the camera -- e.g. a 14mm super wide or a 1200mm+ super tele.
The amplitude of the light rays' angular rotation from the Z- to the X-axis induced by the AF splitter prism will depend on the prism's refraction index, and thus be wavelength-dependent.
Infrared, having a longer wavelength than visible light, will be refracted less, and the induced rotation from Z to X for infrared will thus be less than for visible light.
An object whose distance changes relative to the camera will cause its focusing plane to move along the Z axis. This movement, rotated and projected, will therefore cover a smaller distance on the X axis for infrared than for visible light signals.
Thus, for objects emitting large amounts of near-infrared light, there is, IMHO, a possibility that the AF module will be fooled by IR-induced spurious signals, and detect an on-sensor X-axis velocity that is too low.
In AF servo mode, this could adversely affect the AF sensor's movement prediction and tracking abilities:
- For approaching objects, an underestimated velocity will cause the focus point chosen by the camera to lag behind the object's actual position, i.e. cause back focusing.
- Conversely, for objects going away from the camera, the underestimated velocity might cause front focusing.
It's likely that the engineers at Canon are more than aware of such IR-induced issues as a non-specialist like me could imagine. Thus, it's a certainty that Canon has put serious effort into filtering any IR that could reach the AF detectors.
Nonetheless, one should note that the 1D3 has much more peripheral AF cross-sensors than the 1D2. Recall that the AF concentrator lens and splitter prism are materials with a refraction index larger than 1, which will thus affect the lightray path the camera lens designer had intended, and perhaps induce astigmatism, field curvature, coma, spherical aberrations etc. on the image projected on the AF sensors.
Compared with the 1D2, the 1D3's AF sensor must presumably have improved accuracy (to support the camera's improved 10MP resolution) and be more sensitive in low light (to respond to market demands).
This, coupled with the fact that peripheral sensors, now that they are of the "cross" type, will be more sensitive to aberrations, might have forced Canon's designers to shave as much thickness as possible from the AF sensor's >1 refraction index optical elements, including the infrared filter.
In turn, this might have led to a slightly increased IR sensitivity of the 1D3's AF module, compared with the 1D2's.
One simple way to test this hypothesis would be to put an IR rejection filter -- e.g. a B+W 486 or a Kenko DR655 -- on a lens with a shallow depth of field, with which focusing errors would be obvious, and see if this has any influence on the 1D3's AF behavior.
My 1D3 hasn't arrived yet, and I would be grateful, probably like many others, if an 1D3 owner could put this hypothesis to the test
I should probably also add that birds, unlike athletes wearing synthetic fabrics, might not be the best subjects to test the incidence of IR on an AF system's performance.
A bird feather's perceived color can be caused by a scattering or diffraction of a very narrow wavelength range. All the other light wavelengths, possibly including those close to the visible spectrum like near-infrared, can be absorbed.
See e.g. this article for a glimpse at the complexity behind some bird feathers' colors:
"The structural origin of the weak iridescence on some of the dark feathers of the black-billed magpie, Pica pic (Corvidae), is found in the structure of the ribbon-shaped barbules. The cortex of these barbules contains cylindrical holes distributed as the nodes of an hexagonal lattice in the hard layer cross section. The cortex optical properties are described starting from a photonic-crystal film theory. The yellowish-green coloration of the bird's tail can be explained by the appearance of a reflection band related to the photonic-crystal lowest-lying gap. The bluish reflections from the wings are produced by a more complicated mechanism, involving the presence of a cortex second gap."
As the bird's feather would appear near-black at the absorbed wavelengths, including near-infrared, there would be almost no spurious IR signal interfering with the feather's visible light color signal used by the camera's AF. There wouldn't be any errors in the focusing, then, and using an IR rejection filter like a BW486 would have little effect in such cases.
At any rate, Rob Galbraith's test pictures, and these ones, posted on DPreview, where I think I see a slight back focus, as the boat's wake and the water skier seem to be sharper than the boat's passengers, are consistent with the hypothesis that a large IR signal would cause the focus point to lag behind an approaching subject in AF servo mode.
Edited by H. Ludens on Jun 27, 2007 at 02:48 PM GMT
