Doug Kerr
Well-known member
Canon's new "Dual Pixel CMOS Sensor AF" system, introduced on the Canon EOD 70D, is a stunning advance in AF technology.
Here is my take on what this is all about:
************
The "classical" phase detection AF system
In a conventional "phase-comparison" AF system (as found on many Canon digital EOS cameras), each AF detector has two little sensor arrays. Each operates through its own little aperture mask (roughly circular), the two resulting little apertures being located on opposite sides of the overall lens aperture. (Actually, on opposite sides of what the minimum usable "focusing" aperture would be.)
In a way wholly comparable to the operation of the two parts of a split prism manual focusing aid, the relative positions of the image on the two sensor arrays for the AF detector is indicative of the current discrepancy in focus of the lens. (The underlying principle is actually that of a rangefinder.)
The new Canon "AF on main sensor" system.
At each sensel location, there are actually two little sub-sensels. The arrangement of the micro lens system in front of each sensel position is such that each of the two sub-sensels operates through a separate aperture mask having a roughly rectangular shape. One of those covers about half of the entire lens aperture and the other one covers about the other half of the lens aperture (that is, they abut at a line across the center of the actual aperture).
We see that in this figure from a recent Canon patent application (US 2013/0147998 Al):
In figure 3A, outline 301 represents the current (shooting) lens aperture (here fairly large). Outlines 302a and 302b represent the aperture masks though which sub-sensels 201a and 201b (the two halves of sensel 200) operate (that is, receive their light).
Thus the actual effective aperture of sub-sensel 201a is the left-hand "D"-shaped half of the actual aperture, and the actual effective aperture of sub-sensel 201b is the right-hand "D"-shaped half of the actual aperture.
With respect to AF operation, if we take all the "a" sub-sensels and treat that as an AF sensor array, and take all the "b" sub-sensels and treat that as a second AF sensor array, then, just as for the two sensor arrays of a single conventional AF detector, the relative positions of the image on the two arrays is indicative of the current discrepancy in focus of the lens.
We can of course choose to use individual small regions of these two arrays to be separate "AF points" if desired.
Note that since the respective apertures are not small and separated by the entire diameter of the actual lens aperture, this effect is somewhat "diluted" compared to the working of the conventional AF system. Nevertheless, apparently a workable focus discrepancy indication can be obtained. (I'm sure more sophisticated image processing helps overcome this inherent handicap.) But we must have this less-than-ideal-for-AF arrangement of the two apertures for a reason I will mention next.
Because the two apertures together essentially cover the entire lens aperture, if we combine the outputs of the "a" and "a" sub-sensels at each sensel location, that will be just about like the output of a regular sensel at that location. We use the entire set of those summed outputs to develop the actual image (through demosaicing and such).
In figure 3B, we see the same thing for a situation in which the actual (shooting) aperture is smaller than in figure 3A.
Applicability
In an SLR camera, this system is usable for AF with the mirror up, as in live view still shooting or video shooting. This would also be very handsome in a mirrorless camera. I doubt if it plays any role in regular "SLR-style" AF shooting, but I may be wrong.
************
That's how it looks from here.
Best regards,
Doug
Here is my take on what this is all about:
************
The "classical" phase detection AF system
In a conventional "phase-comparison" AF system (as found on many Canon digital EOS cameras), each AF detector has two little sensor arrays. Each operates through its own little aperture mask (roughly circular), the two resulting little apertures being located on opposite sides of the overall lens aperture. (Actually, on opposite sides of what the minimum usable "focusing" aperture would be.)
In a way wholly comparable to the operation of the two parts of a split prism manual focusing aid, the relative positions of the image on the two sensor arrays for the AF detector is indicative of the current discrepancy in focus of the lens. (The underlying principle is actually that of a rangefinder.)
The new Canon "AF on main sensor" system.
At each sensel location, there are actually two little sub-sensels. The arrangement of the micro lens system in front of each sensel position is such that each of the two sub-sensels operates through a separate aperture mask having a roughly rectangular shape. One of those covers about half of the entire lens aperture and the other one covers about the other half of the lens aperture (that is, they abut at a line across the center of the actual aperture).
We see that in this figure from a recent Canon patent application (US 2013/0147998 Al):
In figure 3A, outline 301 represents the current (shooting) lens aperture (here fairly large). Outlines 302a and 302b represent the aperture masks though which sub-sensels 201a and 201b (the two halves of sensel 200) operate (that is, receive their light).
Thus the actual effective aperture of sub-sensel 201a is the left-hand "D"-shaped half of the actual aperture, and the actual effective aperture of sub-sensel 201b is the right-hand "D"-shaped half of the actual aperture.
With respect to AF operation, if we take all the "a" sub-sensels and treat that as an AF sensor array, and take all the "b" sub-sensels and treat that as a second AF sensor array, then, just as for the two sensor arrays of a single conventional AF detector, the relative positions of the image on the two arrays is indicative of the current discrepancy in focus of the lens.
We can of course choose to use individual small regions of these two arrays to be separate "AF points" if desired.
Note that since the respective apertures are not small and separated by the entire diameter of the actual lens aperture, this effect is somewhat "diluted" compared to the working of the conventional AF system. Nevertheless, apparently a workable focus discrepancy indication can be obtained. (I'm sure more sophisticated image processing helps overcome this inherent handicap.) But we must have this less-than-ideal-for-AF arrangement of the two apertures for a reason I will mention next.
Because the two apertures together essentially cover the entire lens aperture, if we combine the outputs of the "a" and "a" sub-sensels at each sensel location, that will be just about like the output of a regular sensel at that location. We use the entire set of those summed outputs to develop the actual image (through demosaicing and such).
In figure 3B, we see the same thing for a situation in which the actual (shooting) aperture is smaller than in figure 3A.
Applicability
In an SLR camera, this system is usable for AF with the mirror up, as in live view still shooting or video shooting. This would also be very handsome in a mirrorless camera. I doubt if it plays any role in regular "SLR-style" AF shooting, but I may be wrong.
************
That's how it looks from here.
Best regards,
Doug