brian_caldwell

<p>Shun: Regarding lens speed, fair enough - slow lenses will work in many situations. But they won't work in all situations, and many photographers really do need fast lenses, especially for micro four thirds.</p>

<p>Regarding the 18-35 being designed to clear the reflex mirror, that's certainly true. However, adding a Speed Booster amounts to what is a pretty effective optical re-design, increasing the aperture, increasing the field of view, shortening the physical length, and increasing the sharpness and contrast. The resulting high speed lens takes full advantage of the extra room available in m43 cameras.</p>

<p>Shun: Adapting the Sigma 18-35/1.8 via a Speed Booster is the *only* way to get an f/1.2 zoom for m43. In this case its 12.8-25mm f/1.2 . And the image quality is extremely high even wide open all the way to the corner. There is nothing "native" in the m43 world that is remotely close to it.<br>

Andy L: All Canon EF lens mount Metabones m43 Speed Boosters (even the old ones) have been upgraded to have excellent AF performance. Here is an autofocus test with this a Speed Booster + EF version of the Sigma 18-35 using a a slightly older version of the firmware:

<p>"The Apo designation for large format taking lenses and photographic enlarging lenses is "that the lateral chromatic aberrations of the secondary spectrum have been correct to within a very small percentage of the focal length".<br />The Apo designation for lenses used for microscophy is the traditional Abbe definition where "the primary color rays cross at a common point". This is also the common definition."<br>

Bob:<br>

So, what would you call a photographic lens that meets the traditional Abbe definition? Such lenses really do exist: <a href="http://www.coastalopt.com/mmapomacro.html">http://www.coastalopt.com/mmapomacro.html</a> . <br>

The sad thing for S. and R. is that they've painted themselves into a corner by using a dumbed-down (and in my view completely bogus) revisionist definition of "apochromat". When Rodenstock finally sells a REAL apochromat what will they call it?<br>

BTW: I am a lens designer, optical engineer, and lens manufacturer.</p>

<p><em>

<p>P Mui: <br>

1. Strehl ratio numbers.</p>

</em><br>

For an aperture of f/4: Strehl >0.8 (~Rayleigh limit) from ~480nm to ~800nm with peaks of 0.25, 0.995, and 0.955 at 320nm, 520nm, and 770nm, respectively. The Strehl ratio is very nonlinear with respect to defocus - a defocus of only a few microns will have a noticeable impact on the Strehl ratio at f/4. At f/8 the Strehl ratio is greater than 0.8 over the entire waveband from 315nm to 1100nm.<br>

<br /><em>2. An interferometer and total fringe report.</em>

<p>I prefer projection testing for wide field broadband optics. Coastal could do on-axis interferometry at 6328nm for an extra fee. Accurate off-axis testing referenced to a true flat field would require complex fixturing and significantly greater cost. Multiple wavelength interferometry is not currently practical.<br>

<br /><em>3. An optic that will generate PERFECT, or close to accurate pinpoints for point sources. </em>

<p>Relative aperture matters here. Typical high-end apo refractors are slow at ~ f/8. If you are satisfied with these refractors then you should be satisfied with point images equal or smaller than diffraction-limited f/8 point images. Wide open at f/4, the 60 Apo satisfies this condition near the axis, and nearly satisfies this condition in the corners of your 3200 chip. One stop down at f/5.6 the condition is satisfied over the full 3200 format. Two stops down at f/8 the condition is satisfied over a full FX format.<br>

<br /><em>4. A FLAT field with very little curvature.</em>

<p>Comments above and all MTF curves shown in the online brochure pertain to a perfectly flat image field.<br>

<br /><em>5. Image circle numbers.</em>

<p>See 3 above or the published MTF data.<br>

<br /><em>6. Method of manufacture, that is, do they compensate for the melt data of each type of glass and crystalline or do they rely solely on the melt data that the manufacturer provides them?</em>

<p>No. Not practical due to extreme cost (would require custom glass measurements outside the visible band), and not needed since most of the optical elements are chemically pure substances (CaF2 or SiO2)<br>

<br /><em>7. Method of polish for the lenses. Are they high speed polyurethane or pitch polished?</em>

<p>No high speed polishing.<br>

<br /><em>8. Are the batches of lenses consistent from one to another in terms of color correction, figure, etc?</em>

<p>Yes.<br>

<br /><em>9. Are the retaining rings "trimmed" in any way for critical alignment?</em>

<p>Air space adjustment not needed due to tolerant design (unlike air-spaced APO refractors). Lateral compensators are used to adjust off-axis performance.<br>

<br /><em>The optic will be used for astronomy purposes with an SBIG (santa barbara instruments group) camera that has a KAF3200ME full frame chip. The optic that I'm looking for MUST render star points as accurately as possible for photometry and imaging. Also the optic will be adapted to larger imagers (STL11000 and larger interline and full frame sensors). </em>

<p>60 APO will work very well on KAF3200, even wide open. For STL1100 you may want to stop down to f/8 to achieve optimum corner performance.<br>

<br /><em>I have tried MANY other brands that are commercially available with great disappointment. This includes Nikon, Canon, etc. All of them are considered "top rated" but none of them have accurately (color correction and pin point wise) have even been close to satisfactory. </em>

<p>Which ones? What aperture?<br>

<br /><em>The "standard" of performance that I'm looking for would have to EQUAL or BEAT out a precision telescope from Telescope Engineering Company, Astro Physics or the high end Takahashi offerings. Not in terms of focal length, mind you, but in terms of color correction, flatness of field and rendition of point sources.</em>

<p>This is a reasonable standard, but see my comments about relative aperture in part 3 above.<br>

<br /><em>There have been several reviews of optics that were highly rated in terrestrial usage but they have been lacking in accuracy of point sources as well as color correction.<br />If anyone has any REAL numbers other than those provided by the datasheet, the info would be greatly appreciated. This also applies to the method of testing and manufacture of the lens.<br />An optic with a 1/50th rms wavefront would be a good start. Or a 1/10th PV.</em>

<p>Are you *sure* you are getting this with above-mentioned telescopes over a broad waveband? Of course, relative aperture is critically important for a wavefront error criterion. It is possible to build a 60mm lens covering 24x36mm at ~f/2 (or maybe even faster) that has smaller point images than any of the telescopes you mention. Email me if you are interested.</p>

</p>

</p>

</p>

</p>

</p>

</p>

</p>

</p>

</p>

</p>

</p>

</p>

Kelly, the older M lenses are not properly designed for a sensor that has an appreciable (>1/4mm) of flat glass in front of it. If you get rid of the IR filter near the sensor then you have to either put it inside the lens (unacceptable) or in front. Front filters have alot of issues. Like it or not, the best technical solution for a FF digital rangefinder *will* orphan the old lenses.

Sensors can be designed to work just fine with steep angles of incidence for off-axis field points. The real issue is astigmatism caused by the IR and anti-aliasing filters. The best place to locate an IR filter is near the image plane, and the physical thickness of the filter glass introduces astigmatism in inclined converging ray bundles. For short exit pupil distances the amount of astigmatism is very large, and would effectively destroy the imaging characteristics of the shorter Leica M lenses. The astigmatism is minimized by making the IR filter as thin as possible and also by eliminating an anti-aliasing filter. I believe this fully explains Leica's decision to have not anti-aliasing filter, and also their choice of a very thin (0.5mm ?) IR filter. I think this is also the reason Leica chose a sensor considerably smaller than 24x36mm. Unfortunately, the thin IR filter didn't filter enough IR, and you know the rest of the story.

 

The Rodenstock HR series of large format digital lenses work very well despite the fact that the shorter ones have very steeply inclined off-axis ray bundles. One reason they work so well is that the filter glass has been taken into account during lens optimization, so the astigmatism caused by the filter glass is exactly balanced by equal and opposite astigmatism in the lens. Short focal length Leica lenses weren't designed this way, so they make lousy digital lenses.

 

IMO the best option for a FF Rangefinder camera is to come out with an entirely new lens mount that is incompatible with existing M-mount and screw-mount lenses. Then you can design a proper digital rangefinder camera together with a series of digital-appropriate rangefinder lenses. It is not necessary to use bulky reverse-telephoto designs for the wide angle lenses in this scenario.

I agree with Ellis - it would be silly. But my reasoning is that electronic viewfinders will soon be dramatically better than any optical viewfinder relying on a groundglass. Plus no parallax problems.

Rodeo Joe has glimpsed a few truths about lens design.

 

Condsider a lens design. Lets say its an f/2.8 double Gauss 50mm lens for 35mm format. At f/2.8 this lens will have a certain maximum resolution that is determined by a combination of aberrations and diffraction.

 

Now take this lens design and scale it 2x to a focal length of 100mm. In other words, increase all thicknesses and radii of curvature by a factor of two. This 100mm lens will have less resolution (measured in lp/mm) than the corresponding 50mm lens because its geometrical aberrations are exactly twice as large as the 50mm lens. However, the 100mm lens resolution will be more than half the resolution of the 50mm lens due to the effects of diffraction.

 

Looked at another way, lets say you calculate the total number of resolved spots for each lens using some reasonable criterion. Assume that the 50mm lens can resolve 20 million spots over its entire field of view. If diffraction played no role in imaging, the 100mm scaled version would also resolve exactly 20 million spots. The reason? - even though the image size of the 100mm lens is larger, the size of the individual resolved spots is larger by the same amount.

 

However, when you consider diffraction, scaling to a larger size increases the total number of resolved spots because the larger lens is dominated more by geometrical aberrations and less by diffraction than the smaller lens. Sounds paradoxical, but its absolutely true.

Vivek:

I think those Zeiss prototypes only have an altered coating. They aren't truly made for IR in the sense that they still suffer from significant focus shift. I'm actually a bit surprised that alternate Zeiss coatings compromised the blue end as much as they do. It is possible to get a much flatter response from blue through near-IR. My guess is that they are using very few layers in order to save costs.

Bjørn is right about IR transmission - the fact that virtually all AR coatings are pretty lousy in the IR doesn't normally hurt near-IR transmission too much until you get ~1000nm. The tendency of most multilayer AR coatings is to fall off very rapidly at the blue end of their range, and to fall off fairly gently at the red end.

 

An exception would be extremely complex designs with a very large number of coated surfaces. I've seen ~40 element designs that are very good in the visible, which transmit virtually nothing in the near-IR.

 

A much bigger problem that involves coating efficiency is hot spots. Its possible to minimize hot spots by keeping ghost pupil images as defocused as possible, but this is very tedious even with fast computers and up-to-date lens design software. I think that very few photographic lenses have much "built-in" hotspot resistance. A drop in coating efficiency will dramatically increase hotspot visibility unless the underlying optical design is pretty well de-ghosted.

Shun, Lex:

Digital cameras have one or more glass plates covering the sensor. The total thickness of the glass is on the order of 2mm. Something that very few people appreciate is that this sensor glass will introduce enough astigmatism to degrade the performance of better film lenses used wider than ~f/8. A properly designed digital lens must accordingly be designed to compensate for the effects of the sensor glass.

Bob:

The central hotspot seen with many/most lenses when used in the IR is a ghost image of the aperture stop. Its intensity is determined by the stop diameter, the geometry of the lens surfaces and sensor, and the reflectivity of the AR coatings.

 

Below is a sample double Gauss design with a very sharp hotspot caused by an initial reflection off the sensor surface (surface 13) followed by a second reflection off of the second to last lens surface (surface 11). As you can see, the green ray passing through the center of the aperture stop undergoes this double bounce and then strikes the center of the image plane. This means that the ghost image is a sharply focused image of the aperture stop.

 

You can also get similar hotspots by double bounces between two lens surfaces, but those involving the sensor surface are likely to be worse, since the sensor is generally very reflective.

 

If the AR coatings were perfect, you wouldn't see this sort of thing. However, AR coatings are generally pretty terrible in the infrared, so hotspots are an ever present danger.<div></div>

Regarding lenses that cause underexposure when used wide open:

 

Even for the pixel located exactly on the optical axis, light will strike at non-normal incidence. What that central pixel "sees" as it "looks" back toward the lens is a disk of light called the exit pupil. For large apertures that disk of light is big. Light coming from the edges of the disk (a.k.a. marginal rays) strikes the disk at an oblique angle.

 

For really fast lenses the effect can be huge. For example, the marginal ray angle of an f/1 lens is 30 degrees. Many sensors are almost completely insensitive to light incoming at 30 degrees off-normal, and so an f/1 lens would be virtually wasted with such a sensor. I've long suspected that this is the real reason why Canon cancelled their 50mm f/1.

 

In many cases the effect can be beneficial. For example, at largish apertures the sensor will essentially introduce apodization which reduces the intensity of the outer parts of defocused highlights, both foreground and background. For the most part this means improved bokeh. Minolta accomplished more or less the same thing years ago when they came out with a portrait lens having an apodizing filter placed near the aperture stop.

 

Some bokeh effects can be a little weird, however. The Nikon D1x had rectangular pixels, and the angular response was noticeably different in the horizontal and vertical directions. So, on several occasions shooting wide open at f/1.2 I noticed oblong defocused highlights even on-axis.

 

In the case of f/4 and f/2.8, the marginal ray angles are 7.2 and 10.3 degrees, respectively. Whether or not this can explain the exposure differences in the example shown depends entirely on the angular falloff characteristics of the sensor.

Definitely looks like a filter ghost.

 

Such ghosts are sharply focused and have mirror symmetry with respect to the image center. I think what is going on is that the primary image at the sensor acts as a new "object". Light from a point on this new object goes back through the lens and exits as parallel light, assuming the original object was distant. The parallel light hits the flat filter and goes back through the lens to form a sharply focused ghost image on the opposite side of the optical axis.

 

The interesting thing to me is that this can't be the result of a specular reflection off the sensor. If it were, then the the ghost image wouldn't have mirror symmetry compared to the main image. Instead, I think it must be a Lambertian reflection component off of the sensor surface that causes the problem. In other words, the sensor is acting not like a mirror in this case, but rather as a weakly reflecting sheet of white paper. I would guess that this may be due to the fine patterning of the microlenses and underlying pixel circuitry. In fact, it would be interesting to compare sensors with and without microlenses, and also microlenses with and without anti-reflection coatings. Of course, the problem with AR-coating microlenses is that the coating thickness might exceed the microlens thickness!

Photoshop can only correct lateral color. The cat photo clearly shows longitudinal color. There is no possibility of correcting it.

 

Brian

What you are seeing is very normal. Your 105mm f/2 has ordinary achromatic color correction, meaning that only two colors can be brought to a focus in the same plane. The residual secondary chromatic aberration is easy to spot by looking at a defocused highlight (or in your case a bright line) in and out of focus. On one side of focus you will typically see a yellow-green fringe, and on the other side of focus you will see a purple fringe. You can get rid of it by using a true apochromat (very expensive for a 100mm f/2), or by stopping down a bit.

 

Brian

Vivek:

Just saw your link to the Masterprimes after I had mentioned them. Sorry!

Its possible to design and build ultrafast ultrawide lenses with very good image quality. The Zeiss Masterprimes for cinematography are a good example. These are all f/1.2 (T/1.3), and several cover a very wide angle.

 

But they are not small.

 

The reasons are subtle and not immediately obvious. A high speed lens requires that aperture dependent aberrations such as astigmatism and coma be extremely well corrected. A slow lens can tolerate a moderate amount of, say, astigmatism and still produce fine images. However, a fast lens with the identical amount of astigmatism will be lousy.

 

So, in a fast lens it is not enough to simply correct the spherical aberration really well - you have to correct *all* aberrations really well, which is not easy to do.

 

In reversed telephoto wide angle lenses, spherical aberration and coma are mainly corrected in the rear group, which can be fairly small even for fast wide angles for exactly the reasons you've stated. However, astigmatism and field curvature are strongly influenced by the front group, and in general this group must be made quite large to correct these field aberrations to the necessary low levels.

"Edward, is the exit pupil always coincident with the rear nodal point?"

 

Dan:

The nodal points coincide with the principal points for a lens immersed in air. The nodal points and principal points are completely unrelated to the entrance and exit pupils.

 

Brian

Published data from Zeiss shows about 5x more distortion for the 21/2.8 than for the 21/4.5, and that its the "gullwing" type.

Many people will call it coma, but that is not correct. The aberration causing this is sagittal oblique spherical aberration. True coma is never a problem in photographic lenses used at or near their optimum magnification. The only exception still encountered these days is Newtonian telescopes.

 

Brian

Kelly:

Microscope objectives are typically even more sensitive to cover glass thickness. However, the culprit here is mainly spherical aberration rather than astigmatism because the objectives are very fast and are more or less telecentric.

 

Vivek:

I'm surprised to hear about the Nikon zoom example because these are slower lenses. However, its an interesting observation. I recently talked to an engineer at one of the companies that replaces the filter pack with an IR filter to make dedicated IR cameras, and apparently there is a wide range of filter thicknesses; from the 0.5mm of the M8 to about 3mm in the case of some Canon cameras. And this doesn't include the coverslip that protects the actual sensor surface.

 

Tom:

I admit to being an optics nerd, and that the total number of MTF curves I've calculated in my career is probably greater than the total number of pictures that I've taken. I also agree that digital photography works pretty darn well, even using lenses that aren't ideal for the purpose. However, I think I'm able to shed light on the reasons behind design tradeoffs in the M8. I also think we are a long way from reaching a resolution plateau in digital photography. By the time you reach 40-50 megapixels the astigmatism problem due to filter thickness looms much larger.

 

John:

I've no idea what the exact shape of a piece of film is. Its straigtforward to show what happens to MTF when you defocus, but I'm not sure if this would lead to any real understanding of what happens on film. I used a perfect lens to isolate the filter effect. The only way to show what happens with a real lens is to either measure the effect or calculate it using the optical prescription. You can't do an accurate calculation using a published MTF curve because aberrations in the lens will interact with aberrations from the filter in a manner that can't be predicted from MTF curves alone.

Vivek:

I think that cover plate aberrations were a design-driver for Leica, and they were forced to do everything they could to minimize the effect, short of re-designing all the lenses to have a longer exit pupil distance. It certainly explains the unusually thin 0.5mm IR filter, to an extent it explains the absence of an AA filter, and it also suggests one reason why they may have chosen 1.33x instead of 1.0x. After all, they already used the trick of laterally shifting the microlenses to minimize corner shading and color shifting, so why not go ahead with a full size sensor except for the added astigmatism.

 

In theory you could correct for a known amount of astigmatism by image processing, but this is way beyond the realm of photoshop, DXO, or my favorite panotools.

 

If you did a careful test, either by measuring MTF or photographing test charts, you would probably notice the degradation with a really good f/1.4 lens like the 50/1.4 used wide open. However, the effect will be a subtle one due to all the design choices made by Leica.

 

Brian

If you place a filter between the lens and the image plane you will introduce

aberrations, particularly astigmatism. The problem is worse for short exit

pupil distances common for Leica M lenses. I've seen the issue alluded to in

Leica's press announcement regarding the IR problem, but I've never seen any

meaningful discussion in any forum.

 

So, I analyzed what would happen to a perfect f/2 lens if you put various

thickness filters near the image plane. The chart below shows MTF at 40

cycles/mm for filter thicknesses of 0, 0.5mm, and 2.0mm. I think the damage is

alot bigger than most people realize.

 

Leica state that the filter they use over their sensor is 0.5mm thick. In

addition, all sensors have a cover glass that ranges from 0.5mm to 1.0mm in

thickness, so the actual total thickness is getting into the uncomfortable

range.

 

Of course, if you stop down the problem gets reduced substantially, but still,

if you buy the latest 50/1.4 ASPH do you really want its performance degraded

to such an extent?

 

There has been much talk from various manufacturers in recent years

about "designed for digital" lenses. A lens truly designed for digital must be

corrected for a nominal filter glass near the image plane, but, strangely,

nobody ever mentions this.<div></div>

John:

The 70-200 is what I would call a classical zoom lens type that traces its roots to the zoom lenses introduced by Angenieux back in the 1950's. Broadly speaking, this type of zoom consists of a variable magnification afocal front group followed by a fixed focal length rear or prime group. As long as the aperture stop stays put with the fixed rear group during zooming the f/# will remain constant during zooming. This is true even though the physical diameter of the aperture stop is also constant. Additional examples of this type of zoom lens are the 75-150/3.5 and 50-135/3.5

 

Wide angle zooms have a completely different structure, and they have an inherently variable f/#. To make these lenses constant aperture, such as the old 25-50 f/4, it is necessary to link the zooming mechanism to the iris diaphragm to change the aperture stop diameter during zooming.