FrankBaker wrote:
> I have difficulty understanding how focussing performance is
> different among different lenses and even at different zoom lengths
> with the same lens. My mental model of the process must be in
> error. I assume that signals from a detector, or a group of
> detectors, are evauated for some sort of disparity (say blur) and
> the lens is adjusted to minimize that parameter. If the detector
> is at the same _optical_ position as the image sensor, the image
> will then be in focus. Change the lens or a lens' optical length
> and the same criterium will apply: the lens is adjusted until the
> parameter is minimized. Essentially this is saying that the camera
> works the lens to focus the image on the focusing sensors, at which
> point it is in focus on the image sensor. Is this model, then,
> wrong? For instance that the angle of the rays falling on the
> focusing sensors influence their response, as when focusing at 17
> mm as opposed to 70mm?

The sensors are never in the same plane as they should be. Each camera is programmed with a table of offsets, telling the CPU how much to adjust the lens focus for, depending on the active sensor and distance. When you press the shutter button, the lens makes an imperceptible shift of focus before the image is taken. This normally always happens on AF-S, AF-A and AF-C but setting DMF (direct manual focus) prevents the adjustment, disengaging the lens focus the moment the target is locked. The adjustments are tiny, but with fast wide-angle lenses, tiny means a lot.

Problems with specific lenses like Sigma 17-70mm are partly to do with this (fast wide angle, shows up any errors to the maximum) and partly to do with the lens itself. The tiny 'focus tweak' given by the camera has little or no effect on a lens with a long focus throw, low geared, including many early Minolta AF lenses and most longer lenses. It is calibrated for the 50mm f1.7, which is used as the test lens to set all this up. But the Sigma 17-70mm has a very highly geared focus, with a short throw taking it from infinity to 20cm in a very rapid movement. Within the available adjustment, it has to fit from 'just in front of the lens' to infinity. This means a small tweak of the focus with this particular lens results in a big shift of focus.

The A100 is better calibrated, and has the AF sensors more accurately placed, than the 7D. The 5D was also a bit better than the 7D but the A100 has been fine-tuned more. However, it's a DSLR, and all DSLRs use this type of 'acceptable error corrected by programming', as CCD/CMOS sensor circuit boards are nearly impossible to position accurately in a focal plane, to the same standards as a film gate.

If you own a Canon or Nikon, Fixation in London will blueprint your camera (usually only done with pro bodies not D50s or Rebels - costs nearly as much as the camera...) and finely adjust the sensor plane and rotation, adjust the AF module, adjust the mirrors, and shim the focus screen. They can bring a DSLR close to film camera precision. They then reprogram the CPU to remove the original compensation and replace it - hopefully - with no compensation at all. Speeds up shooting, massively improves sharpness, and this is what the London press and professional CanNik users get done. So do the hire centres.

But the service is not offered by anyone that I know of for Minolta/Sony, or Pentax, Olympus or other makes.

David