Why are the autofocus sensors clustered around the middle of the frame?

[Daniel Browning]

Photographers often ask why the autofocus sensors on full frame cameras are not spread out more for better coverage. This post explains the reason why that is not possible. It essentially comes down to the fact that it would require the optical viewfinder to be removed.

This diagram by Joseph S. Wisniewski (reproduced here with explicit permission) illustrates the reason:



The main mirror directs most of the light up to the optical viewfinder. There is a small portion in the center that is partially transparent. It lets some light through to a small, secondary mirror directly behind the primary mirror. The secondary mirror reflects light to the autofocus system. As you can see from the diagram, it is physically impossible to make the secondary mirror any bigger.

On the other hand, if the optical viewfinder was replaced by an electronic viewfinder, then the entire mirror could be partially transparent and reflect light into the PD autofocus system, and the AF sensors could be anywhere on the frame. I hope that such a design will occur soon.

[Fred Doane]

Daniel,

Very interesting! I never knew there was a sub mirror there. Just when I thought I knew something about how these things work, lol. Thanks for the insight.

Fred~

[Joel]

Thanks Daniel! That is one of those things I kept wondering, but never bothered asking.

[Joel Bookhammer]

love the drawing, very good info too!


Thanks


Joel

[apersson850]

It also explains well why cropped cameras have the AF points more spread out. The lens register distance is the same (we use the same lenses), so the mechanical design constrictions are similar, but the smaller sensor sees relatively less of the image, so it appears that the AF points are closer to the edges.

[Colin500]

Is an electronic viewfinder really a step in the right direction? I'm asking because I can only imagine the quality to be worse than a good optical viewfinder due to (a) the lower spatial resolution, and (b) the compression/clamping of the dynamic range into 18 or 24 bits ... but correct me if I'm wrong, I have only ever seen cheap and/or old electronic viewfinders (that were quite horrible, in particular on the resolution side).


(But thanks already for this interesting post, I had always wondered where the AF sensors were physically located, and how they actually got any light.)

[wickerprints]

Good illustration. Here are some thoughts that occurred to me when studying it:

1. How does this schematic change when the reflex mirror is changed to a fixed pellicle mirror as found in the earlier EOS-1N RS? We know it is possible, but I am not sure how the AF system was redesigned to account for a fixed mirror. My understanding was that this camera contained 5 AF points arranged in a horizontal linear configuration across the middle of the frame. (We have come a long, long way since then, haven't we!)
   
2. Given the geometry in the diagram, there does not seem to be any reason why the AF mirror and sensor assembly cannot be made wider--that is to say, it could conceivably be made to stretch lengthwise across the frame. This would not be reflected in the cross-sectional diagram. There must be other considerations as to why this is not done.
3. I can see that one theoretical modification to this geometry which would enable greater vertical coverage would be to use multiple AF mirrors, the height of which would form a geometric progression. That is to say, the bottommost mirror would cover the bottom half of the frame, a second mirror would cover the quarter extending from the horizontal mid-line upward, and so forth. Of course, this is not practical, but in theory, multiple AF mirrors would enable greater coverage. The problem of course is that this drastically increases the complexity of the mirror assembly, and would create horizontal gaps where the mirror attaches to the main reflex mirror, where no AF points could be placed.

I agree that an electronic viewfinder would be the future, given that contrast-detect AF is not yet fast enough to yield satisfactory performance. However, this raises questions about how to deal with the behavior of the shutter, presuming that the main imaging sensor is used as the VF.

[Daniel Browning]

Thanks, everyone.

It also explains well why cropped cameras have the AF points more spread out. The lens register distance is the same (we use the same lenses), so the mechanical design constrictions are similar, but the smaller sensor sees relatively less of the image, so it appears that the AF points are closer to the edges.
<div style="clear: both;"]</div>

You're quite right. Canon decreased the mirror size and shortened the back focus distance by a small amount for EF-S lenses, but nowhere near enough to provide the same ratio between back focus and format size that exists on 35mm. One beneficial side effect is the increased autofocus coverage. One of the downsides is that it makes fast wide angle lenses far more expensive (like a theoretical 15mm f/1.4).

Is an electronic viewfinder really a
step in the right direction?

It's a personal preference. The existing EVF technology is in the stone age compared to what's coming. But even what we have now is already superior in many ways to OVF. I use the EVF on my 5D2 (aka "LCD") as much or more than the OVF, even for handheld (actually, shoulder mount) work.


Advantages of an EVF are:

• Zooming to any part of the image for critical focus
• View the actual captured image, including white balance, black and white, etc;
• Tilt/swivel.
• Movie mode.
• Live zebras, histograms, false color "raw" mode, guide lines, overlays, etc.
• 100% accuracy for framing.
• Accurate DOF (typical viewfinder screens show f/1.2 as if it were f/3.5, making critical manual focus impossible).
• Accurate rendition of the bokeh
• Usable in much darker situations, including effective f/32 for macro, etc.
• Contrast detect autofocus with movable focus points
• Allows the entirety of the primary mirror to be used for phase
  detect autofocus (instead of a small secondary mirror), so AF sensors
  can cover the entire frame instead of just the middle third portion.
• The possibility of reducing lag from 40ms to 0ms or even negative
  lag (this would not apply if a mirror is still used for phase-detect
  autofocus).

Disadvantages, of course:

• Battery usage
• Heat generation (in some circumstances can cause thermal noise to rise above the read noise floor).
• Current technology may be relatively expensive, slow frame rate, low resolution, reliability, and/or bulky.
• Dynamic range limited by the sensor instead of the eye/brain.

I'm asking because I can only imagine the
quality to be worse than a good optical viewfinder due to (a) the lower
spatial resolution, and (b) the compression/clamping of the dynamic
range into 18 or 24 bits ... but correct me if I'm wrong, I have only
ever seen cheap and/or old electronic viewfinders (that were quite
horrible, in particular on the resolution side).

The resolution and DR are definitely worse, right now. But it does have things like 10X zoom that result in far higher resolution (for that one part of the scene), which, for me, are enough to outweigh OVF advantages, even with today's technology.

1. How does this schematic change when the reflex mirror is changed to
   a fixed pellicle mirror as found in the earlier EOS-1N RS?

Great question; I have no idea.

Given the geometry in the diagram, there does not seem to be any
reason why the AF mirror and sensor assembly cannot be made wider--that
is to say, it could conceivably be made to stretch lengthwise across
the frame. This would not be reflected in the cross-sectional
diagram. There must be other considerations as to why this is not done.

Agreed. I don't know the reason either.

I can see that one theoretical modification to this geometry
which would enable greater vertical coverage would be to use multiple
AF mirrors, the height of which would form a geometric progression.
That is to say, the bottommost mirror would cover the bottom half of
the frame, a second mirror would cover the quarter extending from the
horizontal mid-line upward, and so forth. Of course, this is not
practical, but in theory, multiple AF mirrors would enable greater
coverage. The problem of course is that this drastically increases the
complexity of the mirror assembly, and would create horizontal gaps
where the mirror attaches to the main reflex mirror, where no AF points
could be placed.

Neat idea!

I agree that an electronic viewfinder would be the future, given
that contrast-detect AF is not yet fast enough to yield satisfactory
performance. However, this raises questions about how to deal with the
behavior of the shutter, presuming that the main imaging sensor is used
as the VF.

If you go to EVF but *keep* the mirror, then you can dedicate the mirror entirely to AF, and you don't have to use the slow contrast-detect AF. (Or you have the choice of either, depending on whether you feel like waiting for the mirror to slap up and down or not.)

[Colin500]

Thanks for the elaborate answer!


And seeing that my shots at f/1.8 (the largest that I currently have) are often not on focus, presumably because focus-and-recompose just doesn't work with a very shallow DOF, I will try using live-view to achieve accurate manual focus, and exactly where I want it...

Advantages of an EVF are:

• Accurate DOF (typical viewfinder screens show f/1.2 as if it were f/3.5, making critical manual focus impossible).

Aha, I had been wondering why with larger apertures the DOF in the viewfinder seemed to be much greater than in the actual picture -- is it because only a central part of the lens is used for the OVF via the mirror, thus simulating a smaller maximum aperture?


Thanks, Colin

[ShutterbugJohan]

Very good explanation, Daniel. I've heard it explained that way last year by the owner of a local camera store.

The possibility of reducing lag from 40ms to 0ms or even negative
lag

Negative shutter lag?! [] No thanks. I feel like the 40D is too fast already at 65ms shutter lag. [:P]