Modern Digital Cameras and Aperture's Function ... a question...

[ajohnw]

After that one try this one

http://photographylife.com/how-phase...utofocus-works

I'm afraid it isn't the wiki that is wrong.

John
-

[Inkanyezi]

After that one try this one

http://photographylife.com/how-phase...utofocus-works

I'm afraid it isn't the wiki that is wrong.

John
-

So you go ahead and find another image that is wrong, and an explanation that is skewed.

I have explained in words how the system works, and it is exactly the same as microprisms and split-image on the screen.

Do you need it graphic to understand?

<post edit-inserted text in italics>The wiki image is wrong. The second link you provide is [del]DEAD WRONG![/del] actually correct.

[del]The AF system does not work in that way. [/del]
It is however not so easy to see how the images of the AF sensor are captured from different parts of the exit pupil of the photographic lens in front. Just as previously stated, the sensels are grouped in pairs, aiming at different parts of the exit pupil of the camera lens.
</post edit>

Half of the sensels in the AF sensor can only see a spot that is about f/11 distance from the centre of the exit pupil of the lens. The other half can see a similar point on the opposite side. The effective aperture of what these sensels receive is very small, so they always will see a relatively sharp image, due to DOF of this small aperture. Each half seeing a diminutive spot at each their own side of the exit pupil. If the diaphragm is stopped down further than a pre-defined aperture, it will cover those spots, and AF sensels will receive nothing.

You can easily simulate this by looking in the viewfinder through a lens that has a large entrance pupil and cover the centre of it vertically with for example a finger. When focusing the lens, you will see the split image, and it will coincide when focused. If you stop it down, so that the finger covers the entire entrance pupil, there will be no image in the centre of the viewfinder.

[ajohnw]

If you hold a camera lens up to the light you can actually see the exit pupil as per this. It is no where near the sensor but usually buried some where in the lens. I say usually as on one nikon lens it's right at the sensor end.

http://en.wikipedia.org/wiki/Exit_pupil

The 2nd link by the way mentions just why one production run of D7000 had focusing problems. The 1st one mentions that the sensors are F ratio limited due to dof problems.

This is the 1st time I have ever looked at any of the links i have posted by the way. I am sure there will be many more.

John
-

[Manfred M]

Sorry Urban, but John is correct and this is exactly how Nikon has implemented phase detect autofocus. I can see the parts when I pop up the mirror on my cameras.

Your comments about the iris blades blocking the sensor does not sound correct to me as the phase detect is normally done at maximum aperture, and even if it were not, both the focusing screen below the pentaprism / pentamirror and the AF module are at the same exit pupil distance as the image sensor.

The detector sits in the camera base and a secondary hinged mirror reflects the light downward onto the beamsplitter. The primary mirror is semi-silvered to allow some light through and then is reflected down to the beamsplitter that sends two separate beams of light that are then analysed and focus distance is calculated and relayed to the camera electronics to focus the lens.

The diagram that John refers to is indeed correct. If you would like a different source of information:

http://graphics.stanford.edu/courses...tofocusPD.html

http://graphics.stanford.edu/courses...150dpi-med.pdf

[Inkanyezi]

Sorry folks, if you don't want to understand, there is no way you will accept a correct explanation. You can lead a horse to water.

BTW, the first link in Manfred's post has a Flash applet that does show it correctly, just as I put it forth. The micro lenses are squinted, and therefore, each one of them is regarding a particular part of the ray bundle from the main lens. For clarity, those AF sensor sensel groups are shown side by side, while in effect, they are interlaced.

[William W]

. . . the first link in Manfred's post has a Flash applet that does show it correctly, just as I put it forth. The micro lenses are squinted, and therefore, each one of them is regarding a particular part of the ray bundle from the main lens. For clarity, those AF sensor sensel groups are shown side by side, while in effect, they are interlaced.

I concur.

And if one extrapolates the diagram to contain a smaller maximum aperture lens and also assumes the interlaced groups, it appears (to me) not too difficult to see why the AF will fail or hunt badly when a small maximum aperture lens is used (i.e. a slow lens).

This is what I was attempting to explain with the simplified diagram which was meant to show a spacial increase as the aperture becomes smaller.

WW

[jcuknz]

Without really knowing anything about the subject my immediate reaction is that using a simple meniscus lens to illustrate is misleading and one needs to show the full multi element lens and the position of the iris in it to make sense of drawings.

Regarding "Low light", again I am not sure what people mean by low light but my observations are based on a trip to the front gate and getting the camera to focus by street lights [ sodium ??? ] on one side of the road. It was the very devil of a job to STOP the lens quickly snapping into focus. The only place it didn't work was the black shadow underneath a car parked under the nearest lampost ... even after some time playing around my eyes couldn't see anything there either, and of course there was no contrasting areas to work CDAF.

The second example is when I add a TCONx1.7 onto my 14-140 Lumix which due to the mis-match between them gives me a max aperture of about f/10 and again everything worked as usual. Used on my bridge camera with I believe a 'good match' there is a light loss of about 1/3 stop rather than the 5/3 stop loss on the 14-140 which is starting from f/5.8. The saving grace is the higher ISO permissable with the MFT compared to bridge camera.

This second example was in bright sunshine so maybe not a good argument ... I must try it at night-time

[Inkanyezi]

Without really knowing anything about the subject my immediate reaction is that using a simple meniscus lens to illustrate is misleading and one needs to show the full multi element lens and the position of the iris in it to make sense of drawings.

It is not misleading, and the exact position of the iris is not crucial. The iris will be seen from the sensor, as the exit pupil. From front you can see the entrance pupil of the lens. Only those are relevant to the PDAF system, and the meniscus or single biconvex lens is a good approximation. It is shown like that for graphical clarity. Multi-element construction shown from the side will only confuse the viewer, as you will not appreciate the size of the exit pupil from such a sketch.

The problem is not the amount of light, neither has it much to do with depth of field. The PDAF system evaluates two bundles of light, one collected from one point inside the exit pupil of a sufficiently large aperture lens, the other bundle collected from the diametrically opposite side of the same exit pupil. Those small patches of the exit pupil, each produces an image on the PDAF sensor, images that are separate, as the sensels that record them are interlaced, and their respective microlenses are pointed at exactly the relevant patch of the exit pupil of the main lens.

If the exit pupil has a smaller diameter than the distance between those two patches, the microlenses cannot collect any light tor their respective sensels to produce the images from which to evaluate focus.

There has been some confusion about this as I have mentioned closing the diaphragm, which can be done with legacy lenses, but is not done with DSLR lenses, which stay fully open until the image is captured. However, adding a tele extender to a lens, will effectively shrink its exit pupil, so that the PDAF sensel groups will be obscured. The PDAF sensor does not collect all light that enters the main lens, but only the light that comes from those peripheral small patches at a particular distance from the centre of the exit pupil. If there is no part of the exit pupil extending to the patches that the PDAF sensels are regarding, they cannot make images to compare.

[jcuknz]

I'll take your word for it Urban .. you lost me on page one

The point is, for whatever reasons, PDAF bombs out from f/5.6 < f/8 onwards and CDAF doesn't ... so I'm glad that is what I have been using for the past decade.

[William W]

. . . There has been some confusion about this as I have mentioned closing the diaphragm, which can be done with legacy lenses, but is not done with DSLR lenses, which stay fully open until the image is captured. . .

Yes I agree.

Although I believe that I understood what you meant, I do think your mention of closing the diaphragm (iris) caused some confusion.

WW

[AB26]

I am not getting involved in this “argument”. I wish to thank , Urban, John and Manfred, because you made me do some research on how the AF system really works.

From what I can understand there must be “apertures” for the image splitter. I have to assume it will be these “apertures” Urban is referring to. If the Aperture is too small (Urban mentions f11) the image splitting “apertures” will not allow sufficient/any light trough to the AF sensors?

It will also be dependant on the AF module used in the camera, at what minimum aperture (the lens / combination of lens converter) the camera will be able to focus.

[ajohnw]

I suspect some confusion is arising from the term exit pupil. It often has that effect. More often than not as per the wiki the output of optical instruments will be mentioned. They contain virtual images. One way of looking at this is that the world about us is a virtual image - it needs the lens in our eye to produce a real image at the back of our eye. A real image is what a camera lens projects onto the sensor. Hold it up and look through it with an eye at the same distance and there wont be an image. Take an eyepiece or even a magnifying glass and focus that on the same image plain and it will work just like a telescope as the eyepiece will produce a virtual image that the eye can refocus. It used to be possible to buy adapters to convert camera lenses into telescopes.

It's a rather complicated subject but in simple terms an iris is best placed in a position where a virtual image exists. Only the light levels change when the diameter of the hole is changed, no visual information is lost. - other than diffraction effects because the F ratio is changing. There is no image there until another lens focuses it to produce a real image.

Light and optics are rather odd. Googling optics conjugate planes may help explain why. Again in simple terms say a camera is focused on something 2m away. There is an image on the sensor but there are other things in the view as well. In principle these are also producing focused images at different distances from the sensor right out to infinity. It may seem a bit strange that these don't fog the main image due to focus blur. An acquaintance produced a video of what is actually going on in optical systems that illustrates just how weird light is. It's here and concerns microscopes and Abbe but much the same goes on when ever real images are formed. It might interest a few people but maybe not.



A bit of explanation. The back focal plane of a microscope objective is mm away from the back of it and the real image that is being studied is around 150mm from that. Another lens is used to focus an image of an iris onto the back focal plain of the the objective and yet more optics are used to focus the light source onto the plane of the iris.

John
-

[Inkanyezi]

I suspect some confusion is arising from the term exit pupil.

It shouldn't.

We're not discussing telescopes or microscopes here, and "exit pupil" will have a different meaning regarding a photographic lens.

An image was shown here, pointing to the exit pupil of a photographic lens. The lens has an entrance pupil and an exit pupil, and those are often not of the same size. In a wide angle lens, the exit pupil is often larger than the entrance pupil, and a telephoto lens usually has an exit pupil that is smaller than the entrance pupil. These can be observed from front and back of the lens when detached.

But there is an optical element that should not be disregarded when we dicuss PDAF systems, and that is the microlenses in front of the PDAF sensor. Those are grouped in parallel groups, where they are shifted somewhat, in order to aim at those peripheral parts of the exit pupil of the main lens that are intended to produce the images that shall be compared to calculate how to position the lens to focus it. So in a horizontal group, half of the microlenses are shifted right and the other shifted left. In a vertical group, half are shifted up and half are shifted down.

To understand how it works, one must understand that the defocused image from a peripheral part of the camera lens will be shifted in position. This is what causes bokeh, unsharp rendering of the image from the entire lens surface. But the AF sensels only receive light from a diminutive part of that surface with much less bokeh. Hence, just as with a hole camera, we can disregard the fact that the two images captured by the AF sensor are not perfectly sharp.

But the condition that they should render an image at all, is that the peripheral part of the lens that the sensels are aimed at by shift of their respective microlenses, also will lie within the exit pupil. When the lens has too small diameter of the exit pupil, the microlenses will be aimed at the sides of the exit pupil, which cannot provide an image. Therefore, the system cannot work when the exit pupil of the lens is small, smaller than a particular F-number that depends on the amount of shift of the microlenses. Most DSLR AF sensors are calibrated for about f/5.6.

And it is not actually the absolute size of the exit pupil, but its angular size as seen from the sensor. Hence a more telecentric lens needs a larger exit pupil, but it all really boils down to the F-number, which is relative to the virtual focal length as seen from the sensor.

[Inkanyezi]

If the Aperture is too small (Urban mentions f11) the image splitting “apertures” will not allow sufficient/any light trough to the AF sensors?

The f/11 figure is the distance from the centre of the exit pupil for each of the patches that are looked at from the AF sensor. Each one of them is displaced by that amount from the centre. To get the minimum permissible aperture, this figure has to be doubled, hence 5.6. This means a calibration of the displacement, the shift, of the microlenses before the AF sensels. Those microlenses are shifted to receive light from an angle of the periphery of a 5.6 lens. When the lens aperture is smaller, their target will fall outside the exit pupil.

[ajohnw]

One more attempt and my last one. I will leave it to you to figure out the construct of the exit pupil when the diameter changes but the reason it is constructed this way is that the angular view does not change when the aperture is changed. Some people are inclined to think the size of the piece of glass at the sensor end is the exit pupil. It isn't. The front element on some types needn't be the entrance pupil either.

http://hyperphysics.phy-astr.gsu.edu.../exitpupil.gif

Looked at another way the rays shown in this diagram are from a minuscule point on the object. In practice all points on the object produce the same arrangement of "rays" at angles determined by their "height" on the object otherwise no image would be generated and all rays pass through the exit pupil what ever it's diameter right down to a pin hole which is so small it can behave like a lens. The problem with a pin hole is that the amount of light it allows in is rather small so huge wide angle images would be too dim. Neglecting field curvature aberrations of course.

John
-

[Inkanyezi]

One more attempt and my last one. I will leave it to you to figure out the construct of the exit pupil when the diameter changes but the reason it is constructed this way is that the angular view does not change when the aperture is changed. Some people are inclined to think the size of the piece of glass at the sensor end is the exit pupil. It isn't. The front element on some types needn't be the entrance pupil either.

Nobody has suggested that the pieces of glass at the rear or front ends should be the exit or entrance pupils. Those pupils are both virtual, and they can be observed from back and front of the lens. If you look at a lens from front and at a reasonable distance, what you see through it is the entrance pupil. The size of it can vary, according to distance of the object that is focused. The entrance pupil, when viewed from far, has a diameter that corresponds to the F-number of the lens.

The exit pupil is a similar thing, but viewed from behind, and observed from the same distance as the sensor. It is graphically explained in the link in the above post. It is the virtual image of the diaphragm as seen from the sensor. This image will have a certain angle of view, which corresponds to the F-number. The PDAF sensels are pointed at small spots near the periphery of this exit pupil. The angle of view of the exit pupil, as viewed from the sensor, shrinks when the aperture is smaller, this should be fairly evident. It bears no relation to the angle of view of the lens or the image it captures.

[Inkanyezi]

I think much of the confusion stems from the erroneous denomination of the most common AF principle in DSLR cameras.

I don't know how it started, but for some obscure reason, they call it "phase detection", misleading people to think that it has anything to do with detecting phase. The principle is a lot simpler, it is triangulation.
http://en.wikipedia.org/wiki/Triangulation

So the "PDAF" is in fact a simple triangulation system, utilizing the width of the lens as base for the triangle. If the lens is not wide enough, it cannot focus.

"Width" here, for triangulation, is the width of the entrance pupil of the lens. However the sensors will not utilise the entire width, but only the projection of the AF sensels through the microlenses via the exit pupil upon the entrance pupil, which in most cases corresponds to the width of f/5.6.

From now on, I will refuse to use the term "phase detection autofocus" for triangulating AF. There is no such thing as phase detection AF in digital cameras.