This is the 13th post in a series of Nikon D850 tests. The series starts here.
I decided to see if I could figure out a quantitative — if statistical — test for PDAF bias. I used the same camera, the D850, and lens, the 105 mm f/1.4.
I took four of the mailers that I use to mail out copies of my book. I didn’t put books in them, but I rolled them into the shipping configuration and taped them shut. I printed out three copies of a slanted edge chart with the Horshack AF target in the center. I assembled the boxes in a sort of staircase arrangement and taped them together. Then I put a copy of the slanted edge char at each level. When I was done, I had this:
I took some gaffer tape and taped the whole thing to a door:
Now I had a target that had three different levels to it, about an inch apart. I’m calling the levels, progressing from left to right, near, mid, and far. Not very original, I admit.
I put the camera on the stand and set it about 4 meters from the target. Why not the three meters I’d used earlier? I wanted to keep the off-axis distance of the near and far slanted edges fairly low so that this didn’t turn into a lens field-flatness test.
Here’s what it looked like to the D850:
If we zoom in, we can see that this is a visual test for LoCA:
I made a set of exposures with the lens wide open using the D850 focus shift shooting feature at minimum step size, electronic shutter, ISO 64. I measured the MTF50 of each raw color plane for the three top vertical slanted edges. Here’s what I found:
The vertical axis is MTF50, measured in cycles per picture height (cy/ph). The shot taken with the lens focused to the closest position is on the left, numbered step 0. There are 10 more exposures plotted. There is no way to know the focused distance for these exposures.
You would expect the far curves to be slightly sharper than the near ones, since the target is smaller in the far ones, and that makes the edge sharper. If I were using a razor blade, the distance wouldn’t matter, but it does with an inkjet-printed target. If you remember the previous, 3-meter, curves, you note that these are slightly sharper. That’s because I was at 4 meters for these. I think it’s clear that the minimum step size the D850 can manage in Focus Shift Shooting mode is larger than you’d like for this kind of thing, even with an inkjet-printed target. It looks like some of the curves missed the real peak because the steps are so large.
Still, these curves indicate to me that the one-inch depth offsets in each direction with respect to the mid plane are enough to give useful information about misfocusing.
My next step was to make 32 exposures of the target using live view with contrast-detect autofocus (CDAF) using AF-S, single serv0, and a small spot right in the center of the image, on the Horshack cross-hatch pattern.
Here are results for each raw color plane:
The lighter blue line is the average (mean) MTF50 of all 32 images in each plane of the target. The orange and gray lines have the standard deviation of the results (sigma) subtracted from and added to the mean. About 70% of the data points would lie between those two lines if the data were Gaussian.
It is clear that the red light is being brought to a focus at a plane behind the mid plane, which is where the CDAF mechanism in the camera is focusing.
The CDAF is focusing in the correct plane for green light, which is the most important when it comes to sharpness.
Blue and green are focused at about the same distance with this lens, so the blue curves look like the green ones.
Now with PDAF, MUP, and EFCS, racking focus on the lens between shots:
The variation is far greater with PDAF than with CDAF. The camera appears to be front-focusing red light slightly. — the sharpest red plane is behind the plane the camera was asked to focus on.
However, it looks like the D850 appears to be back-focusing green light slightly. — the sharpest green plane is nearer than the plane the camera was asked to focus on.
Unsurprisingly, the blue results look like the green ones.
It’s not clear to me that the way this camera is set up is bad. Red contributes about 30% to luminance, and green most of the rest, so some compromise may be in order. There are those who would set up this PDAF system to optimize green focus, and some small adjustment would facilitate that; if that’s how you want your AF set up, you’re in luck with CDAF, because it appears that that’s how Nikon designed it.