Rapid autofocus: expectation too far?

With the birth of the new super-Nikon (the D3) and the problems reported with the competition (Canon's 1D mkIII) I was given to pausing for thought.

2 things sprang to mind. First up: will the Nikon be flogged through similar tests to see if it lives up to expectation? Or are no sports shooters going to touch it? That's really an aside.

The second point though relates to the expectations we now place on technology versus some of the practicalities of making that happen. it seems to me, in all the brouhaha that no-one stopped to think if it was even possible for the Canon's autofocus to work under the rapid fire conditions. I will use this as an example but there are more and more instances of this sort of thing. I'll explain what I mean.

So, the 1D mkIII can shoot at 10 fps (frames per second). It does this onto a 10MP sensor of 19x28mm. That's pretty small sensors allowing for some close pixel peeping. Meanwhile the autofocus is supposed to keep up. In 0.1s the camera must do the following (starting from previous shot):

• Lower the mirror
• Settle mirror in place
• Acquire target
• Focus lens
• Lock focus
• Determine exposure (assuming AE)Raise mirror
• Release shutter

The focus part is a tiny fraction of that tiny fraction of a second. It requires a control feedback loop to close (actual open loop on Canon cameras) and a mechanical device to move. This all takes time.

There is then the issue of accuracy. All autofocus systems work within a tolerance of acceptable focus. I believe for Canon this is defined as 1/3 the depth of focus range for the lens wide open. That translates to approximately 1/3 the depth of field for mid-range targets (not exactly, I know but I'm illustrating a point). With a 300mm f/2.8 lens at, say, 50m (DOF +/-1.5m) that's about 50cm tolerance. If the subject is moving slower that 5m/s (50cm/0.1s) then the AF may do nothing at all.

So we now have a situation where the entire system has to focus more accurately than before, on smaller increments of subject movement in an ever shorter time-frame. However, the manufacturing tolerances in the system haven't moved.

I've not even mentioned the issues surrounding getting the mirror back into place. That means any vibration or movement of the bumpers the mirror sits on, or expansion through heating has to be factored in.

I'm specifically trying to prove anything about the particular AF issue, just pointing out the many factors that go into determining if we might expect the results predicted/sold.

At some point we have to get realistic about our expectations. As an engineer part of my job is to try and determine if a solution is feasible or if we can work within the range of expectations in real-life conditions (not just in a controlled lab). That means thinking about some of the principles of the device and it's capabilities and figuring our if all the marketing hype approaches what might be expected. Claims for cameras are starting to quickly diverge from the bounds of feasibility, in my opinion and we need to look more critically at claims before even testing them.