Autofocus System Design

[mosswings]

Marianne:

Thank you for publishing this wonderful series! A similar discussion is occurring on the DX forum in which several people, including myself, are a bit confused about AF performance with faster lenses. Even Roger Cicala has chimed in. Right now we're struggling with the idea of "spillover" of a faster lens providing a larger light field to the PDAF sensors, but this didn't seem right to me and based on your discussion of the optics involved is surely incorrect.

I am still uncertain about the meaning of the term "f/x circle" with regards to PDAF sensor placement. I remember seeing the derivation of it once, but can't find the reference. I believe it refers to the angle off the optical axis, but when you're talking mulitpoint PDAF systems, the simple explanations start breaking down. Can you clarify the definition of this term? I got hung up on it when visualizing your introductory posts and set it aside, but it needs to be included in your treatise.

 

[Austinian]

I like this thread very much! Impatiently awaiting the nest post. Much better than a soap opera!

So do I, and I'm very appreciative of the effort put into writing it in ways that a wider audience can understand. Well done indeed.

Even though the optics referenced are Nikon, this is extremely interesting to non-Nikon owners who are curious about PDAF systems in general. I found this thread from a notice in the PS&T forum.

Perhaps when it's completed this information could become a DPR article?

 

[Marianne Oelund]

Right now we're struggling with the idea of "spillover" of a faster lens providing a larger light field to the PDAF sensors,

Wider-aperture lenses do not provide a "larger light field" to the AF sensors - only to the imaging sensor. The only thing that a "faster" lens can do, is add flare to the AF system, which reduces contrast for the AF sensor and can actually reduce focus speed.

This was demonstrated using the 200mm f/2, masking the main lens objective to allow only a central 40mm-diameter circle to pass light. Autofocus speed actually improved a little bit. See

Understanding Focus Shift

fabgo wrote: That is interesting! Indeed, I placed a tiny piece of a post-it in front of my 28mm lens (about 6mm or so square), and the center point stops working. I was able to get the edge points to still work, but then again I'm not sure I placed it exactly in the center. And of course, I...

www.dpreview.com

but this didn't seem right to me and based on your discussion of the optics involved is surely incorrect.

You are correct. The best demonstration on this thread, is in my post "D300 AF Module in Operation" where it is clearly shown that all light passing through the separator-mask openings, corresponds to only four small patches on the main lens exit pupil. Any light passing through the main lens exit pupil, that does not go through one of those four patches, ends up hitting the separator mask, and is blocked from the AF sensors (except for weak reflections inside the AF module housing, i.e. flare).

I am still uncertain about the meaning of the term "f/x circle" with regards to PDAF sensor placement.

PDAF sensor placement is not related to the size (or portion) of the main lens exit pupil in use; these are set independently. The field lens mask selects the area of the main image that is available to its own AF points, whereas the separator lens mask determines the areas on the main lens exit pupil from which light rays can be accepted. The sizes and shapes of these masks are not related.

I remember seeing the derivation of it once, but can't find the reference. I believe it refers to the angle off the optical axis,

Yes, the angle from the optical axis can be used to specify the circle, as well as the f#. They are related: Angle = Atan(1/(2*f#)). For example, the rays within the f/5.6 circle can be up to Atan (1/(2*5.6)) = 5.1 deg. from the optical axis. So an f/5.6 lens projects rays that range from 0 deg., up to 5.1 deg. off-axis. An f/2 lens projects exactly those same rays, plus additional rays that are 5.1 deg. to 14 deg. off-axis; however all of those additional rays will hit the separator-lens mask and be blocked.

It's also important to understand that the f/2 lens passes exactly the same brightness, or light flux, within its central f/5.6 circle, as an f/5.6 lens does. It is not "brighter" across that sub-circle within its exit pupil.

My next post will discuss the Canon EOS-1Dx AF sensor and associated optics, which includes an f/2.8 cross-type AF point. Hopefully, this will clarify the concepts.

but when you're talking mulitpoint PDAF systems, the simple explanations start breaking down.

Actually, they don't break down at all. One must realize that placement of AF points is in the image plane (at the field lens mask, or on the AF sensor's surface), whereas selection of the light that will reach those sensors is in the main lens phase plane. These are independent spaces, which do not interact.

--
Qualities possessed by God in infinite proportion: Love, Grace, Power, Righteousness, Wisdom, . . .
Qualities possessed by humans in infinite proportion: Ignorance.
- Marianne

 

[Marianne Oelund]

I like this thread very much! Impatiently awaiting the nest post. Much better than a soap opera!

You interpreted the slant as an indication for oversampling, which is certainly very plausible.

All Nikon AF patents from US4561749 1984 to US20090167927 2008 (the V1 AF) talk about interpolation from the pixel values to the exact location of the miminum C(x) of the correlation function C_L = sum_i | A_i - B_i-L | where A_i and B_i are the pixel values to be compared and L indexes the C(x) values at pixel pitch.

Thanks very much for this detail, Bernard! I'll make use of it, in my discussion of focus-error calculation.

The slightly shifted pixel lines in parallel (by the skew up to 15mum) measure slightly different image locations. And it still brings in benefits from oversampling the data. Its conceivable that the parallel lines in the bunch are handled as separate measurements each with interpolated C(x) minimum and goodnes qualification calculated in parallel. In the end the asic-computer infers the best consensus result of the the minimim C(x) and the corresponding feedback to the AF servo.

I have an idea for an experiment which could "break" the sensel stagger and possibly reveal more about how the sensel array is used. More later . . .

 

[mosswings]

Right now we're struggling with the idea of "spillover" of a faster lens providing a larger light field to the PDAF sensors,

Wider-aperture lenses do not provide a "larger light field" to the AF sensors - only to the imaging sensor. The only thing that a "faster" lens can do, is add flare to the AF system, which reduces contrast for the AF sensor and can actually reduce focus speed.

This was demonstrated using the 200mm f/2, masking the main lens objective to allow only a central 40mm-diameter circle to pass light. Autofocus speed actually improved a little bit. See

http://www.dpreview.com/forums/post/50092107

but this didn't seem right to me and based on your discussion of the optics involved is surely incorrect.

You are correct. The best demonstration on this thread, is in my post "D300 AF Module in Operation" where it is clearly shown that all light passing through the separator-mask openings, corresponds to only four small patches on the main lens exit pupil. Any light passing through the main lens exit pupil, that does not go through one of those four patches, ends up hitting the separator mask, and is blocked from the AF sensors (except for weak reflections inside the AF module housing, i.e. flare).

I am still uncertain about the meaning of the term "f/x circle" with regards to PDAF sensor placement.

PDAF sensor placement is not related to the size (or portion) of the main lens exit pupil in use; these are set independently. The field lens mask selects the area of the main image that is available to its own AF points, whereas the separator lens mask determines the areas on the main lens exit pupil from which light rays can be accepted. The sizes and shapes of these masks are not related.

I remember seeing the derivation of it once, but can't find the reference. I believe it refers to the angle off the optical axis,

Yes, the angle from the optical axis can be used to specify the circle, as well as the f#. They are related: Angle = Atan(1/(2*f#)). For example, the rays within the f/5.6 circle can be up to Atan (1/(2*5.6)) = 5.1 deg. from the optical axis. So an f/5.6 lens projects rays that range from 0 deg., up to 5.1 deg. off-axis. An f/2 lens projects exactly those same rays, plus additional rays that are 5.1 deg. to 14 deg. off-axis; however all of those additional rays will hit the separator-lens mask and be blocked.

It's also important to understand that the f/2 lens passes exactly the same brightness, or light flux, within its central f/5.6 circle, as an f/5.6 lens does. It is not "brighter" across that sub-circle within its exit pupil.

My next post will discuss the Canon EOS-1Dx AF sensor and associated optics, which includes an f/2.8 cross-type AF point. Hopefully, this will clarify the concepts.

but when you're talking mulitpoint PDAF systems, the simple explanations start breaking down.

Actually, they don't break down at all. One must realize that placement of AF points is in the image plane (at the field lens mask, or on the AF sensor's surface), whereas selection of the light that will reach those sensors is in the main lens phase plane. These are independent spaces, which do not interact.

 

[Marianne Oelund]

Thanks. Your point about the independent spaces of the imaging and AF systems is to what I was referring as "the simple explanations start breaking down". It is not clear from the simplistic diagrams on the web that they are independent,

Ah, now I see what you're referring to. I have not found any PDAF explanations on the web at all, which discuss the function of the field lens as the "enforcer" of the correspondence between the separator-lens mask, and the "patches" on the main lens exit pupil. In fact, some of their diagrams do not even include a field lens!

This is why my thread development started off so carefully, discussing the different types of planes and the role of the field lens in the optical system. These are neglected concepts.

 

[Mako2011]

My next post will discuss the Canon EOS-1Dx AF sensor and associated optics, which includes an f/2.8 cross-type AF point. Hopefully, this will clarify the concepts.

Thanks so much. Been able to follow to this point and the way the f/2.8 cross-type AF point works will really add to it.

 

[mosswings]

Thanks. Your point about the independent spaces of the imaging and AF systems is to what I was referring as "the simple explanations start breaking down". It is not clear from the simplistic diagrams on the web that they are independent,

Ah, now I see what you're referring to. I have not found any PDAF explanations on the web at all, which discuss the function of the field lens as the "enforcer" of the correspondence between the separator-lens mask, and the "patches" on the main lens exit pupil. In fact, some of their diagrams do not even include a field lens!

This is why my thread development started off so carefully, discussing the different types of planes and the role of the field lens in the optical system. These are neglected concepts.

 

[Leo360]

Numbers for Focus Precision

The field lens masks are 8.4mm high, and the images projected by the separator lenses are 1.88mm high. This means the magnification at the AF sensor, relative to the main-lens image, is 1/4.5.

If we are using an f/8 lens on the camera (which matches the spread angle of the separator mask images "projected" by the field lens), the movement of the images on the detection lines will be 1/4.5 as much as the radius of the COC in the main image. For example, if the main lens is a little out of focus so that it produces a 20um-diameter (10um radius) COC in the main image, then the images on the AF sensor's detection lines will be displaced 10um/4.5 = 2.2um. We want the AF system to be able to detect a displacement of this size when an f/8 lens is in use.

Should the AF image lateral displacement also depend on the distance from separator lens to the AF sensor? Would it be possible to play with distance and get a better number than 1/4.5 ?

--Leo--

 

[Marianne Oelund]

Should the AF image lateral displacement also depend on the distance from separator lens to the AF sensor? Would it be possible to play with distance and get a better number than 1/4.5 ?

The limitation will be the overall AF module size. You can scale the AF sensor image size up, but then the module needs to be larger to accommodate it - which then means making the lower section of the camera larger.
Everything in the module scales up and down together, so you couldn't readily increase just the separator lens to AF sensor distance; the field lens to separator lens distance would also increase in proportion (unless you're happy to reduce the AF-point coverage of the camera image).

 

[Marianne Oelund]

First, a quick review of the D300 sensor and its associated separator mask for comparison. Here is the D300 sensor, which has two sets of vertical detection lines plus two sets of horizontal detection lines for the center group, and just two sets of vertical detection lines for each of the lateral groups:




D300 AF sensor



Each of the eight sets of detection lines requires its own separator lens, so we find eight openings in the D300 separator-lens mask:




D300 separator mask with eight openings for the eight separator lenses.

The Canon 1Dx design adds cross-type detection for two columns of AF points in each of the two lateral groups (with f/4 sensitivity horizontally), and f/2.8-sensitivity cross-type detection for the middle 5 AF points in the center column. This requires adding two sets of horizontal detection lines for each of the lateral groups of AF points, and four sets of detection lines to the center group. The detection lines for f/2.8 sensitivity need to be set about twice as far from center, as the f/5.6-sensitivity detection lines are. To minimize the AF sensor size, these have been set diagonally away from center, so the detection lines also need to run diagonally (photo released by Canon):




EOS-1Dx AF sensor includes a total of 16 sets of detection lines

All of the additions require quite a bit more real estate, especially since the f/2.8-sensitivity lines for the center group require moving the three groups apart some. This sensor die measures about 15mm wide by 6.8mm high (note all dimensions are inferred from photos and may not be exact). To save a little space, the line sets for the lateral groups have been crowded a little closer; this requires the separator lenses for those groups to be slightly prismatic, to re-aim their rays closer together.

If you check dimensions carefully in this photo, you can see that the center-to-center spacing of the horizontal-line sets for the outer groups, is about 1.4x the center-to-center spacing of their vertical-line sets. This is because the outer cross-type AF points have f/4 sensitivity horizontally, but f/5.6-sensitivity vertically.

Another interesting detail, is the length of the detection lines for the f/2.8-sensitivity AF points (diagonal lines). These only serve a single AF point each, yet they are about as long as the horizontal lines for the center group, which each serve 3 AF points. This extra length is very useful for f/2.8 AF points, as otherwise the out-of-focus detection range would be very narrow.

I have not found any photos of the separator mask for this sensor, but have put together my own educated guess. The locations of the openings (relative spacings) should be fairly accurate, but the sizes and shapes of the openings are just my speculation (and I've used the D300 shapes for convenience). It's reasonable to expect that the mask openings for the f/4-sensitivity and f/2.8-sensitivity separator lenses will be larger since there is space on the main lens exit pupil for larger patches, further from center:




Marianne's separator mask design for the EOS-1Dx (not patented).

On the separator mask, the spacings between pairs of mask openings must all be exactly scaled to the size of the aperture circle on the main lens which they correspond to. Thus the openings for f/4-sensitivity are Sqrt(2) times further apart than the f/5.6-sensitivity openings, etc.

Here is a look at how the camera lens exit pupils are used by the D300 and EOS-1Dx AF systems. The D300 only uses (approximately) the f/8 circle, whereas the 1Dx uses the f/5.6, f/4 and f/2.8 circles. The 1Dx field lenses are aimed so that all six of the mask openings for vertical detection lines (f/5.6-sensitivity) come from exactly the same two patches on the main lens exit pupil. Also, the four f/4-sensitivity mask openings for the lateral-group horizontal lines will share two patches:




Main-lens exit pupil patches used by D300 and 1Dx AF systems (see full-size image for clarity). Sizes and shapes of the patches shown are only approximate.

For perspective, I have shown aperture circles up to f/1.4. This underscores the baseline disadvantage for an AF system that is (nearly) restricted to using the f/8 circle.

Since the EOS-1Dx also uses 3 field lenses, the exterior appearance of the complete AF module is very similar to the D300/D3 module (photo released by Canon):




EOS-1Dx AF module (Canon photo).

The sizes of the field-lens masks correspond to the sizes of the AF-point arrays for the three groups: Center array is 7 high by 3 wide and outer arrays are 5 high by 4 wide. Note that the field lens masks are unaffected by the design choice of AF-point sensitivity (f/5.6 vs f/4 vs f/2.8).



--
Qualities possessed by God in infinite proportion: Love, Grace, Power, Righteousness, Wisdom, . . .
Qualities possessed by humans in infinite proportion: Ignorance.
- Marianne

 

[Leonard Shepherd]

It is always good to see new and original investigations which would have taken a long time to plan and undertake.

Due to pressure of work involving 2 new photographic exhibitions I intend to respond in stages to some of your other points, but over time

The separation between the pairs of mask openings is less than I had expected, and actually gives this AF module the capability of focusing with f/8 lenses.

The experience of Nikon wildlife shooters going back to Moose Peterson in the 1990's is every Nikon film and later DSLR has had some f8 AF capability, though not as good as f5.6 ability.

I chose Nikon over Canon in 1999 having checked out the f8 ability before switching from Olympus MF.

The eight separator lenses project non-overlapping images onto the AF sensor. The size and shape of those images is set by the field-lens masks, and is scaled by the ratio of the separator-lens and field-lens focal lengths.

(snipped)

As is typical for AF systems which have a large, regular array of AF points, the AF detection lines are contiguous, rather than separate for each AF point. Here we see 22 "merged" vertical detection lines, and the 10 horizontal lines which are used only for the center group of cross-type points:

[ATTACH alt="D300 AF sensor "chip" seen through its optical window"]media_3010138[/ATTACH]
D300 AF sensor "chip" seen through its optical window

This image of a 51 point AF unit is a big step forward to understanding the complexity of the Nikon AF system.

The central viewfinder AF rectangles are always described as cross type yet in this illustration the actual horizontal detection lines are positioned outside the 3 vertical detection lines.

Provided the 2 horizontal separator lenses project light onto the 2 horizontal direction areas at a slightly different angle to the 2 vertical separator lenses, the AF system can detect horizontal detection lines which are not lined up with the viewfinder rectangles. This is an ingenious arrangement very different to small individual crosses which used to be illustrated by Pentax (when Pentax was still Pentax) before AF systems became as capable as Nikon 51 point AF.

The far left and far right rows of 4 vertical detectors on the unit are much more off centre than the AF rectangles in the viewfinder. This seems to help explain why the Nikon AF patent refers to an offset focus adjustment for the outer AF detectors.

Not mentioned anywhere by Nikon that I can find is the length of the detection lines used with 51 point AF changes from a bit longer than a viewfinder rectangle in single point AF to just outside the far boundary of the next AF point in group dynamic AF.

This implies but does not confirm the number of AF cells in use on a detection line varies with the AF setting, and that with single point AF a smaller segment of the area illuminated by a mask is used than when using group dynamic. Varying the number of detection cells used on a detection line should be a relatively easy to do with the contiguous detection lines shown in the above illustration.

On a detail the left horizontal detection lines are separated from the 3 central vertical lines yet the horizontal detection lines on the right seem to touch the outer right vertical line. What effect, if any, this might have using a cross type sensor is unclear to me.



--
Leonard Shepherd
Producing good quality photographs, or being good at sport or art, involves a little more than buying appropriate equipment. Practice, some learning and perhaps natural talent often play a bigger role than the equipment in your hands.

 

[mosswings]

This image of a 51 point AF unit is a big step forward to understanding the complexity of the Nikon AF system.

On a detail the left horizontal detection lines are separated from the 3 central vertical lines yet the horizontal detection lines on the right seem to touch the outer right vertical line. What effect, if any, this might have using a cross type sensor is unclear to me.

They're actually symmetrical. The darkest black areas are the actual sensing areas. The lighter grey and silvery areas that appear to touch are lead-in and lead-out structures and wiring.

--
Leonard Shepherd
Producing good quality photographs, or being good at sport or art, involves a little more than buying appropriate equipment. Practice, some learning and perhaps natural talent often play a bigger role than the equipment in your hands.

 

[Marianne Oelund]

The central viewfinder AF rectangles are always described as cross type yet in this illustration the actual horizontal detection lines are positioned outside the 3 vertical detection lines.

There actually isn't a direct one-to-one correspondence between the viewfinder AF-point positions, and the locations on the AF sensor. It is a one-to-two or one-to-four correspondence (in the EOS-1Dx system, it is one-to-eight for five of the AF points). The action of the field and separator lenses rearrange things, so that a single AF point has two or four detector-line regions on the AF sensor that it uses. The detector-line regions used by a particular AF point do not have immediate proximity. Here is a "map" which shows how the detector-line regions correspond to the AF points:


Mapping of AF points (upper) to detector-line regions of the AF sensor that are used by them. Note this is a back-side view of the AF sensor, i.e., looking toward the separator lens cluster.

In the lower part of this image, the detector lines have been made white by use of a photographic negative, so that they will stand out; they are black in the original photo. Superimposed on the lines, is a second photo of an actual set of images projected by the separator lenses. Each detector line is divided into 3 or 5 regions, which are each used by one AF point.

This is an ingenious arrangement very different to small individual crosses which used to be illustrated by Pentax

Yes, there are fundamental differences in the way that detector lines are arranged. I had considered discussing some of the older AF systems, but could not find full details regarding their optical arrangements.

The far left and far right rows of 4 vertical detectors on the unit are much more off centre than the AF rectangles in the viewfinder.

In absolute physical terms (e.g., measuring in millimeters), the separation between left-group and right-group detectors is not greatly different than the separation between their respective areas in the camera image (in fact it is slightly less). However, the images projected by the separator lenses have undergone a 4.5:1 downscaling, which makes the distances between them appear relatively large. For example, the original image at the center field lens mask (comprising the area occupied by the 15 cross-type AF points), is even taller than the entire AF sensor chip.

Not mentioned anywhere by Nikon that I can find is the length of the detection lines used with 51 point AF changes from a bit longer than a viewfinder rectangle in single point AF to just outside the far boundary of the next AF point in group dynamic AF.

This implies but does not confirm the number of AF cells in use on a detection line varies with the AF setting, and that with single point AF a smaller segment of the area illuminated by a mask is used than when using group dynamic. Varying the number of detection cells used on a detection line should be a relatively easy to do with the contiguous detection lines shown in the above illustration.

It certainly opens up new possibilities for the designers to use the data from the detection lines, since the assignments are now in firmware. In addition to the examples you cite, it is a simple matter, for example, to extend image-shift calculations to other areas in the rows or columns which can help in finding defocus for long lenses which may have image shifts across multiple AF points.

On a detail the left horizontal detection lines are separated from the 3 central vertical lines yet the horizontal detection lines on the right seem to touch the outer right vertical line. What effect, if any, this might have using a cross type sensor is unclear to me.

The detection lines are the black regions in the sensor photo. If you refer to the above negative which makes their positions clearer, you can see that the lines are symmetrically spaced.

--
Qualities possessed by God in infinite proportion: Love, Grace, Power, Righteousness, Wisdom, . . .
Qualities possessed by humans in infinite proportion: Ignorance.
- Marianne

 

[Marianne Oelund]

Here, the D300 AF module is set up with the 200 f/2 VR, looking at a subject about 7 feet away. The images projected onto the AF sensor are made visible by substituting a piece of matte transparent tape for the sensor. These were photographed by my D800E with a macro lens and exposure set to Manual.

A second camera was used to photograph the lens aperture from the front, to provide a means of measuring the f-stop setting. As a bonus, in these photos we can also see the separator mask openings projected onto the lens pupil.

The lens aperture lever was held by a cardboard wedge - first, near wide open (almost f/2), and second, stopped down to the point where one can just see some corner vignetting start to occur in the sensor images. The f-stop for the second case turns out to be about f/7.5, which is where the lens aperture diaphragm is just starting to cover the outer edge of one of the separator mask openings.

The primary result from this demonstration, is that the brightness of the images on the AF sensor is unchanged (other than the slight vignetting mentioned above). This demonstrates that none of the extra rays from opening the lens wider than f/7.5 are arriving at the AF sensor.

It is also interesting to note that with the lens wide open, there are some areas of flare occurring. With the lens stopped down, the flare is absent. In this composite, the left side shows the 200 f/2 front view when almost wide-open and the AF sensor images for that aperture; the right side shows the lens stopped down to about f/7.5 and the corresponding AF sensor images:





If the lens is stopped down beyond f/7.5, the AF sensor images very quickly fade to black as the lens diaphragm covers the separator-mask openings.



--
Qualities possessed by God in infinite proportion: Love, Grace, Power, Righteousness, Wisdom, . . .
Qualities possessed by humans in infinite proportion: Ignorance.
- Marianne

 

[Marianne Oelund]

Today I replaced the "projection screen" on the D300 AF module, in an attempt to improve the visible image detail. Instead of frosty cellophane tape, I'm using a solid piece of clear plastic which I have filed on one side to produce a sort of "ground glass" surface. There is still quite a bit of texture visible, but I think it is possible to see more image detail now.

This demonstration will give the actual image shift in microns, for a change in lens focus. The setup is much the same as for my immediately preceding post: The 200 f/2 is aimed at a target about 7 ft. away, with the D300 AF module positioned behind it. The images projected by the separator lenses, onto the screen, are photographed by the D800E with a macro lens, at 1:1 magnification.

In the following composite, three examples of sensor images are shown (at 50% resolution). The first was taken with the lens focus ring set to the 8 ft. position. In this case, the separation between the separator-lens images in each pair, is 2.78mm.

The lens focus ring was then moved to the 7 ft. position for the second example. The separation between the images in each pair reduces to 2.70mm, a reduction of 80um, so each image has moved 40um (0.04mm) towards the other image in its pair.

For the third example, the lens focus was kept the same, but the lens was re-pointed slightly to the right. This demonstrates that all eight images move in the same direction when the subject moves, and the separation between the image pairs does not change.

The composite also includes a crop of the image taken by the camera through the 200 f/2. This crop shows the three areas which are selected by the field-lens mask; you will see that it is these three areas which are projected onto the AF sensor:





Looking carefully at these separator-lens images, you may notice that the focus has become slightly softer by changing the lens focus ring from the 8 ft. position to the 7 ft. position. This amount of defocus is considerably less than one sees in an image taken with the 200 f/2, even if it is set to f/8, when changing the focus ring between those two positions. In fact, the focus change seen here is very similar to what one sees in the camera image, if the 200 f/2 is set to f/22; this demonstrates the high depth of focus for the AF system, which is due to the small size of the patches on the main lens exit pupil that are used.

Calculated Image Shift

We can compare the 0.08mm shift in relative image positions, to the expected value that we calculate. When changing focus from 8 ft. (2438mm) to 7 ft. (2134mm), the 200 f/2 moves its image plane by 1/(1/200 - 1/2134) - 1/(1/200 - 1/2438) = 2.81mm along the optical axis. Since the AF system is using the f/8 circle, the lateral shift at the image sensor is 2.81mm/8 = 0.351mm. However, the D300 AF module scales the main image down by a factor of 4.5 for the AF sensor, so the shift seen at the AF sensor is 0.351mm/4.5 = 0.078mm, which compares well to the observed figure.

--
Qualities possessed by God in infinite proportion: Love, Grace, Power, Righteousness, Wisdom, . . .
Qualities possessed by humans in infinite proportion: Ignorance.
- Marianne

 

[luxor2]

Superb demonstration!

 

[AZ Steve]

Many, many thanks, you have put AF in focus for us.

 

[Marianne Oelund]

If you have read a few of my previous posts here, you are aware that the AF system is very selective about the light that it admits to the AF sensor, and only passes rays that come from the central f/5.6 or f/8 circle of the main lens. This means that when wider-aperture or "faster" lenses are used, there are quite a few extraneous rays shooting around inside the AF module.

We will take a look at what is happening inside the AF module housing, between the field lens array, and the separator-lens mask. To do this, I have removed the field lens array from the module housing, and set it up on a macro rail with an f/2 lens, and a screen placed exactly where the separator-lens mask would be (21mm behind the field lenses):




Field lens array (sitting on blue box) projecting the main lens exit pupil, onto the plane of the separator-lens mask (white screen).

I photographed the field-lens projections at a number of different aperture settings of the main lens, from f/8 to f/2. In order to see the screen, the camera needed to be placed at a rather steep angle off-axis, so the photos are in perspective.

Recall that the field lens projects the main lens exit-pupil plane, onto the plane of the separator-lens mask, so we will see the shape of the lens diaphragm. In the first example, for f/8, the circles are outlining the areas where the separator-mask openings are; keep this in mind as a reference (I had wanted to place an actual-size copy of the separator mask on the screen, but decided that accurate alignment would have been too difficult).

As the main lens is opened wider, the field-lens projections become wider in proportion, until they achieve a wide overlap at f/2:




Images of the main lens exit pupil, projected by the three field lenses, onto plane of separator-lens mask.

Of course, an f/1.4 lens would produce even larger circles. We see that the "fast" lenses would cause quite a problem, by mixing up light between the different field lenses.

To prevent this, and generally reduce flare from wide-aperture camera lenses, barrier walls are placed between the field-lens optical paths within the AF module housing:




AF module with field-lens array removed, showing interior barrier walls between field lenses.

Although the AF module is well-equipped with these internal black walls, they are not quite as non-reflective as some other surfaces, such as the inside of the camera's mirror box. Thus we can see noticeable flare from wide-aperture lenses (see prior post).



--
Qualities possessed by God in infinite proportion: Love, Grace, Power, Righteousness, Wisdom, . . .
Qualities possessed by humans in infinite proportion: Ignorance.
- Marianne

 

[Marianne Oelund]

That's a pretty clever invention considering ignorant humans did it.

You do not know what you believe you know.

All that you hold to, is mere shadows.