Why does the autofocus capability depend on the aperture rather than the amount of light available?

Question

In many DSLR manuals one can read something to the effect of:

• Maximum Lens Aperture: f/3.2 - f/5.6

  With all AF points, cross-type AF sensitive to both vertical and horizontal lines is possible.

Supposedly autofocus won't work on lenses beyond f/5.6 because there wouldn't be enough light on the focus sensor, which is why cameras "helpfully" disable the autofocus on weak lenses. But why is the aperture relevant in the first place? Shouldn't the manual instead say something like "no autofocus below 50 Lux"? E.g. I'm sure that if I shoot the sun with an f/32 lens there would be more than enough light for the camera to focus and likewise shooting at f/0.95 in the middle of a cave would make it impossible to focus automatically.

Answer 1

Without going into the details of how "phase detect" autofocus works, it is not inaccurate to say that it is analogous to stereoscopic vision, which you are probably familiar with.

You can judge the distance of an object you see because your eyes are looking at it from slightly different perspectives. If you raise up a finger and look behind it, you will see two fingers. Your eyes need to turn inwards to match up the images. The closer the finger is, the greater the separation of the two images will be, and the more the eyes need to turn inwards. This is the information your brain uses to calculate the object's distance.

It is not hard to see that this effect will be stronger if the eyes are further apart. Thus a greater inter-eye distance allows for more accurate distance estimation.

An autofocus system does something very similar, but instead of using two eyes, it utilizes diagonally opposite edges of the lens. For example, it may compare the view through the leftmost part of the lens vs the view through the rightmost part. The two images will be shifted when focus is not accurate, the same way you see two fingers when you look behind it. The bigger the aperture, the bigger the separation between these views, and the more accurate the "distance" judgement (i.e. where to focus to) can be.

Roughly speaking, the angles of the two views that a focus sensor uses is fixed. Thus for every focus sensor there is an aperture limit beyond which it won't work. This has nothing to do with the amount of light. If the aperture is too small, the view of the focus sensor will be obscured. It is trying to look too far to the left and right.

---

Hopefully this illustration should clear things up:

On the left we have the lens, then the aperture blades obscuring part of it, then the AF sensor. The AF sensor is looking narrowly along two separate directions, indicated by the arrows. It is not looking through the middle of the lens. If the aperture is too small, it will block the view of the AF sensor.

Answer 2

Why does the autofocus capability depend on the aperture rather than the amount of light available?

Because the way phase detection autofocus works is dependent upon the difference in the light rays from the same area in the field of view as they strike opposite sides of the lens. The wider the effective aperture of the lens (more properly called the entrance pupil), the wider apart the light rays that are compared can be. The wider the difference between the two points, the better AF systems can perform in terms of accuracy and and speed. More light helps too because it increases contrast between lighter and darker edges, but only if that brighter light is reaching the lines on the PDAF sensor array.

Each AF point uses a pair of lines of light sensitive pixel wells similar to those found on the camera's sensor. Each pixel well is (generally) larger than the main image sensor's pixels and they are not filtered for color differentiation like a Bayer masked sensor is.

To take advantage of the wider aperture, the pairs of line sensors for a particular focus point in the AF array, or more precisely, the microlenses at the entrance to the PDAF sensor array that aim light at the lines on the surface of the PDAF sensor, must be further apart from one another. But that makes those lines not very useful when a lens with a narrower aperture is attached to the lens, because then no light is reaching those points. Some camera manufacturers hedge their bet a little. Some of the focus points are more sensitive/accurate but only function well with a large aperture lens. Other focus points are tuned to be able to use the light from lenses with narrower apertures. But those points can't take advantage of the wider light rays provided by a wide aperture lens.

This is because the two lines on the focus array for each focus point are in a fixed position. If they are close enough to each other to be able to use light that gets through each side of the lens with a narrow f/8 aperture, they are not far enough apart from each other to sense the light that gets through the edge of the lens with a wide f/2.8 or wider aperture. Even when a faster lens is on the camera they are only using light falling on parts of each side of the lens that are close enough to the center for that light to make it through the narrower aperture.

All PDAF focus points use a pair of lines. Cross type focus points use two pairs of lines: one pair for the vertical and a separate pair for the horizontal. "Diagonal cross type points" add an additional two pairs of lines oriented at 45° angles to the vertical and horizontal lines.

The distance these lines are apart on the PDAF sensor array (combined with the way the microlenses on the entrance of the PDAF array aim the light falling on various parts of the front of the camera's main lens) determine how far from the center of the lens' optical axis is being sampled by that set of lines on the PDAF sensor. If the two lines for an AF point are sampling light from two opposite points on the front of the lens that are closer to the center of the lens' optical axis they will work at narrower apertures, but they will be less sensitive. If the two lines for an AF point are sampling light from two opposite points on the front of the lens that are further from the lens' optical axis they will be more sensitive, but they will only work if the lens has an entrance pupil wide enough that there is actually light making it to those lines on the PDAF sensor array.

Camera makers, particularly Canon which is indirectly referenced in your question, hedge their bets a little. This is especially the case with PDAF systems that have a very large number of AF points. Some of the points are tuned to sample light that is passed by narrower lenses, such as f/5.6 or f/8. Other points are tuned to look further to the edge on opposite sides of the lens. These points only receive light from the lens if the entrance pupil is wide enough to allow light to reach them.

In the past, Canon tended to make each set of lines in their cross type points sensitive at different apertures. The horizontally sensitive set of lines may be sensitive to lenses with a maximum aperture of f/5.6 or wider while the vertically sensitive set of lines for the same AF point may only work with lenses having an f/4 or wider aperture. A little more recently the cutoff for each pair in the set may be f/8 and f/5.6, respectively. Some of the more recent high end models have crosspoints where both sets of lines for a single cross type AF point are capable with lenses or lens/extender/teleconverter combinations as narrow as f/8. Improvements in the technology of AF systems have allowed the smaller differences detected by sampling narrower baselines to result in performance that could previously only be obtained from AF points using a wider baseline.

Those same improvements can also be used to make f/2.8 AF points better! Canon also tended (and still do) to make the diagonal cross-type points use a wider baseline that requires a wider aperture lens, usually f/2.8 or wider. This gives those points extremely accurate sensitivity where it is needed the most: when used with wide aperture lenses that give an extremely shallow depth of field and very little margin for AF error.

Since "cross type" AF points are really two sets of lines sensitive at 90° angle to each other, there's nothing that says you can't make a cross type AF point, or even a diagonal cross type AF point, that will work with lenses having an aperture narrower than f/2.8, f/4, f/5.6, etc. It's just that camera makers, particularly the one referenced in the quote contained in the question above, have chosen to make their diagonal AF cross points their most sensitive ones.

Incidentally, the same physics that favors AF with lenses with wider apertures also favors DSLR cameras with larger sensors. Because the mirror is larger, particularly because it is wider, and allows the semi-translucent portion that allows light through to the secondary mirror to be reflected down into the PDAF array to also be wider in a full frame camera than in an APS-C camera, the baseline used for the most sensitive focus points can also be wider.

For a little deeper answer on how cross type points work and a visualization of how f/2.8 points require lines that are further apart, see this answer.

For a slightly different take and a couple of more thorough looks at how the maximum aperture can affect AF performance (as compared to How to enable Canon AF with teleconverter?), please see the two upvoted answers to: Will the Canon 5D MK II with 100-400 1:4.5-5.6 work properly with Kenko 1.4 converter?

Answer 3

Here's a diagram of phase detect autofocus

When the image is in focus, (2), the two pencils of light go to the same place. When it's out of focus, they go to different places (1, 3, or 4). The two have a difference between them encoded in the beam, e.g. making one a bit vertically elliptical and the other horizontally elliptical.

The distance between the two is an indirect measure of how out of focus the image is. The sensitivity, and thus capability to decide the image is in or out of focus, depends how quickly they separate.

You can see from the triangular geometry of things that the sensitivity is best when the angle between the two pencils is largest.

Contrast-detect autofocus works differently and is instead making a decision based on the sharpness of the image; likewise, when the image is equally crisp everywhere (small aperture, deep depth of field) it is difficult for the system to determine where best focus is.

Answer 4

The autofocus part of the camera has a couple of line sensors¹. Each pixel in the line sensor has a small prism in front of it, so it only collects light coming toward the sensor at a particular angle:

When a sensor is designed for faster lenses, that angle is larger. With a lens that's fast enough, that makes it easier to measure the phase difference.

Looking at this relative to lens speed, we get something like this:

In this case, the f/2.8 sensor is trying to respond to light that would have to have been transmitted outside the aperture. Since the aperture blocks such light, the sensor simply doesn't get light it can respond to.

A sensor designed for the aperture (the f/4 line) does receive light, so it works.

There are some borderline cases. The line sensor is exactly that--a line. The pixels in the line all receive light being transmitted at the same angle, so they don't all receive light transmitted through the exact same part of the lens. In a typical case, the camera will read data from the lens, and disable any sensors that require a large aperture than the lens provides. In the case of a tele-extender, however, it's possible for the camera to receive the data appropriate for the lens without the extender. In this case, the camera may continue to try to use a sensor even when the aperture provided is smaller than it's designed for. In this case, you could end up with the light going into the sensor looking like this:

Part of the sensor is receiving light, and the rest isn't. If a little more than half the sensor is receiving light, you still have at least some chance of the system still working, even though your aperture is smaller than it's designed for. Nonetheless, it may have a hard time finding coinciding parts of the image until it's nearly in focus, so it's a lot more likely to "hunt", and may easily just "give up", even in cases where there's enough light and contrast that it would normally focus quickly and accurately.

---

1. Yes, there are crossed sensors and such, but these are basically just more line sensors.

Answer 5

In most cases, obviously not attempting to photograph the sun or in a dark cave, the aperture selected on the lens is a rough approximation of the available light. Obviously, the smaller the aperture, the less light is admitted to the sensor. Obviously, again, the less light on the sensor means that there is less available data for phase-detect autofocus to use to determine and lock on correct focus.

Shouldn't the manual instead say something like "no autofocus below 50 Lux"

What method do you have and use to measure lux or luminance or, even better, candela per square metre? Isn’t it a term that most users of DSLRs barely understand and have little or no means to use? Do you suggest that, after purchasing a DSLR and an autofocus lens, we ought to also purchase a light meter?

I believe the statement in the manual is an attempt to briefly and quickly explain a very complicated subject. After most of us use autofocus cameras and lenses for a while, we understand from the manual explanation that, generally, as light goes down, the probability of autofocus lock also goes down.