2020-10-18 - SEND FEEDBACK
Related:
depth of field, digital sensor, focus skew, image stabilization, In-Body Image Stabilization, lens mount, lens mount / sensor parallelism, lens skew, quality control, Sony, Sony A7R, Sony A7R II, Sony A7R III, Sony A7R IV, Sony FE, Sony FE 12-24mm f/2.8 GM, Sony mirrorless, swing and tilt
Sony A7R IV
I’ve had a chance to shoot two samples of the Sony FE 12-24mm f/2.8 GM on the brand-new loaner Sony A7R IV from B&H. The loaner shows similar results to my camera sent in for repair for two samples of the 12-24/2.8 GM and also the Sigma 14-24mm f/2.8 DG DN Art.
If there is interest, I might publish the full f/2.8 - f/8 aperture series for both samples.
While the error in alignment is very small (at or below what is likely buildable), it is troublesome in the 12-18mm range in particular.
The root cause is surely some sensor swing, and some lens variance.
Terminology
Sample1 = Sony FE 12-24/2.8 GM used for initial work
Sample2 = Sony FE 12-24/2.8 GM received later
My A7R IV = the Sony A7R IV I purchased early late last year, whose sensor and shutter modules were replaced in late Feb 2020
Loaner A7R IV= Sony A7R IV, brand-new
Contradictory sharpness between lenses
When there is swing and/or tilt in a camera system, confounding results can be found.
My series show that Sample1 is sharper than Sample 2 both near and far*, because the system (camera + lens) is tilted slightly to favor both foreground and distance, similar to a view camera or tilt/shift lens with a mild tilt adjustment to improve foreground to background sharpness.
Similarly, the amount of swing (left/right) favors Sample1 overall.
Focus Distance vs microns at sensor (imaging plane)
At 12mm, the requirements for lens mount / sensor parallelism are absurdly tight, on the order of 10 microns (1/100 mm). Cameras just cannot be built that “tightly”.
Blue lines are lens Sample1, Yellow Lines are Sample2
Thanks to Samuel Chia for anazlying annotating this image.f2.8 @ 1/1600 sec, ISO 100; 2020-10-17 15:13:03
Sony A7R IV + Sony FE 12-24mm f/2.8 GM @ 13mm RAW: Enhance Details, LACA corrected
[low-res image for bot]
Samuel Chia’s analysis
Samuel Chia has been cross-checking my analysis, and in a more rigorously numerical way too. Joseph Holmes has also been very helpful. I made my conclusions, then emailed him the images
What Samuel is saying is that a mere 6 microns of swing accounts for a focus of 15 feet on the left side versus 12.5 feet on the right side—easily visible. Although that looks bad to me at full-res, it is about as good as it gets.
Excellent setup!
Yes, I'm seeing the same too. I looked very carefully for the center plane of focus, to help quantify the amount of swing and tilt of each, relative to the camera's position, taking the x point as 10.5 feet.
1st sample, swing: 15' on left, 11.5' on right. This is 10 microns closer on the right image edge. Looking again, I may have drawn the far limit on the right edge too near. It should be about 15' on left, 12.5' on the right. This would be 6 microns worth instead.
1st sample tilt: 21.5' on the left, 15' on the right. This is 9 microns closer on the bottom image edge (when re-orientating the camera to landscape position).
2nd sample, swing: 27' on left, 13' on right. This is 18 microns closer on the right image edge.
2nd sample, tilt: 14' feet on left, 17' on right. This is 6 microns closer on the top image edge (when re-orientating the camera to landscape position).
I've layered your files and marked out the plane of focus, which shows field curvature too, as well as the near-far limits of the DoF at f/2.8 for the vertical as well as horizontal frames. You might find them to be a useful reference (~2.5GB file size for both)
Notice that the DoF is about twice as deep on the right image edge for Sample1 vs Sample2, because of the tilt of Sample2. Strangely I don't see a corresponding phenomenon for the DoF in the vertical frames. I need to look at them again. Also, my markings are not 100% precise, don't worry too much about that since the direction of the tilt and swing are in agreement with yours, and if I go over them again I would re-draw the lines slightly differently each time. More time looking will allow a more precise determination. This is part of the reason why it takes so long to get the correct judgment for shimming work. This morning I discovered that my Sigma 35mm shimming adjustment is a little off still. And a few days ago I fixed my V65 yet again. Very irritating.
If we do not consider bayonet machining differences, the LoanerA7R IV is probably a bit worse, a little more swung (~2-6 microns worth) nearer to the right image side than the A7R IV you sent in for repair. But only just. It would be more precise to say the two cameras are virtually alike in their swing positioning, assuming you used lens sample #1 on the repaired camera.
The errors are actually pretty small, all things considered! It's only this noticeable in part because the DoF at the edges are so thin at f/2.8, there's rearward bending of field curvature at the edges and the GM 12-24/2.8 is so sharp, the transition from in-focus to out-of-focus is so obvious, coupled with a high resolution sensor, tiny differences are highly magnified.
Sony A7R IV and Wide Angle Zooms
|
|
|
|
|
I also spent this morning looking at many 40+MP full resolution samples of Leica SL 50mm and 35mm F2 APO lenses. I found serious swing and tilt issues exceeding what you are seeing here. In $5000 lenses! I also noted many people have issues with the manual focus rings of these lenses being too tight, needing factory re-adjustment, and way too often the repair doesn't fix the problem. Occasionally it does. And that the fly-by-wire feel of these SL lenses are awful (no personal experience). How did you find the by-wire focus feeling of these lenses compared to say the V50? The SL 35, when it's in focus, can be fairly sharp right to the corners, although it has big problems with keeping PSFs round wide open, affecting a wide region off-axis. I sure wish Cosina Voigtlander would make their version of this lens, an APO-Lanthar with manual focus only. It would be better than Leica's and smaller and lighter and cheaper. I hear Leica is making 28, 24 and 21mm versions too, so it seems that wide focal length lenses can also enjoy the advantages of this superior optical design in a relatively small package. Presumably, Cosina could make them too if they wanted. I dearly wish!
DIGLLOYD: the exact details don’t matter much since all estimates are subject to eyeballing how things fall out. Bottom line is that Sample1 is a solid match for the Loaner A7R IV and Sample2 is very good also.
Samuel Chia’s estimate is ~6 microns swing and 8 microns tilt. Both of these guys feel that this is impressively good for an out-of-the-box camera. Their past efforts to get within 10 microns with shimming should be context enough.
As it turns out, Joseph Holmes characterizes this situation for Sample1 as “... 7.3 microns swing. LOL—in any case this system is CLOSE to swing free. Squeezing this much total information from a sensor so tiny is just insane”.
Reader Comments
I want to let you know that I have really enjoyed reading the posts about your Sony A7R IV and swing and tilt. I particularly enjoyed reading that the are two other people out there trying to push the limits of what is possible. Twenty years ago I was doing similar things with my view cameras trying to eke out the ultimate image performance. So I remember the pain of trying to figure this sort of thing out.
I have a few comments related to these posts that I hope are helpful.
First, for context, over 30 years ago I worked as a Mechanical Engineer in test, evaluation, and failure analysis. During this time frame our internal machining facilities were Aerospace Certified and could hold tolerances to +/- 0.0005 inch (12.7 microns), but for normal “high” precision work tolerances of +/- 0.005 inch (127 microns) were used; required tolerances depend on the application. Considering that machining technology has advanced considerably and we are discussing a mass produced consumer item, tolerances for sensor position of +/- 0.001 inch (25.4 microns) is probably reasonable. These and tighter tolerances are certainly possible, but the time and cost to do so goes up considerably with each increment.
When I read the conclusions about the amount of swing and tilt, and looked at the provided image with the blue and yellow lines, things did not add up in my mind. The estimated amounts of swing and tilt, even though they are additive, don’t seem to be enough for the result shown. Because of this I duplicated your table, but things still did not add up for me.
Looking at the results of unintended tilt or swing in an image, the effect should be centered around the focus point. A swing located at the image center that is forward on one side should have an equal amount back on the opposite side; with a proportional effect as the swing point moves off center. From your image the focus points appear to be point #15 on the tape measure for blue and point #12 for yellow. This centering would occur at the focus point even if the swing or tilt were applied at the sensor edge. While your blue line is close, the yellow line is not, so something else must be happening.
My first thought was, that in addition to tilt & swing, there may be some shift involved. Or the axis of the lens may be offset from the center of the sensor. After further thought, I realized that I have seen similar images from my Linhof Super Technika IV camera. In this case, due to an extreme drop, the revolving back was bent, and the springs in the ground glass holder would bend a plastic film holder and the film inside.
Looking at the solid blue and yellow curves gives us more information. If we look at the right side of the image it appears that both curves will overlap. This is what we expect to see. In fact, the more the focus point is moved to the right edge the more sense the swing/tilt explanation makes. It is when we move to the left side of the image that things don’t add up. As both curves seem to become steeper as you move to the left. The blue curve beginning very close to the tape measure and the yellow very close to the right hand “)” in the 10.5 ft comment. The fact that the yellow curve significantly deviates from the blue curve shows that the yellow lens has its own problems.
Without being able to see A&B images, inverted from each other, such as the Mirex adapter allows, we can’t say anything about any deviations due to the alignment of the sensor and the lens mount.
Though, the above leads me to the conclusion that this camera’s sensor, or a sensor stack component, may be slightly bent (or curved) towards the left side of the image, in addition to the suspected swing and tilt.
Finally, the fact that Sony A7IV and Sony 12-24mm lens combination allows us to read a tape measure to within 1/4 inch at 10.5 ft is very impressive. The fact that this also lets us see how well the manufacturing process works is awe inspiring.
Though this brings up serious questions for photography and the the photo industry:
Does the current IBIS system position the sensor to account for variances in manufacturing tolerances? If not, can a future IBIS system do so?
If an IBIS system can’t account for tolerance variations, can the process be improved to correct this situation and keep camera prices affordable? If not, it might be possible that the precision needed for 100+ MP will make Leica prices seem reasonable.
PS. You may want to look at Novoflex bellows options. One of these may provide the flexibility to rotate a camera or lens independently of the other.
DIGLLOYD: there is definitely more to it than sensor swing and tilt. Lenses themselves are far from perfect. It’s also true that that without the proper gear that it can be difficult to do more than deduce a likely camera swing/tilt.
The imaging system of lens + camera totals its errors—errors can cancel out to varying degrees, or can be amplified (additive).
When a group of lenses all establish the same general pattern of swing and tilt, it is fair to conclude that the camera is involved, especially if no other exceptions have been seen with other lenses (indeed, this is the case with my A7R IV which I sent for repair). With this brand-new loaner camera and the one I sent in for repair, 5 different lenses all are swung to the distance on the left side. That might argue it is all the lenses, except *no* lens has ever shown swing to the other right on my camera (the one in for repair).
There is no perfect lens. All lenses have their own swing and tilt. Furthermore, for one lens to be sharper in both the foreground and background (as in this scene) is proof enough that one of them is tilted (probably both, but in different directions). Also, both lenses look to be optically decentered slightly, even Sample1, which is the better of the two (it has trouble on the left side, as seen in House Under Construction ).
On this scene, the Sigma 14-24mm f/2.8 DG DN Art shows a similar swing to the distance at left, but it also looks to be tilted more than the two Sony lenses.
Furthermore, zoom lenses are notorious for failing to place the center of the image circle at the center of the frame. This is easily seen by switching lenses; the center of the image can be displaced considerably between two lenses, showing that one (or both) of them is nowhere near centered about the frame. See for example the awful quality control of the Nikon NIKKOR Z lenses in Shootout vs Nikon Z 14-30mm f/4S: Bonsai Above Beaver Pond in which there is a displacement of 28 pixels horizontally and 54 pixels vertically! This drove me nuts when comparing lenses. The two Sony zooms here have no significant displacement, but I have seen Sony zooms with displacement issues.
2020-10-18 - SEND FEEDBACK
Related:
depth of field, digital sensor, focus skew, focusing, Joseph Holmes, lens mount, lens mount / sensor parallelism, lens skew, quality control, Samuel Chia, swing and tiltThe issue of lens mount / sensor parallelism can be separated into swing and tilt:
Camera aside, a lens can also have swing and tilt relative to its lens flange (and also have optical decentering, further muddying the waters).
Distinguishing swing and tilt of the camera from that of the lens is difficult without a lens adapter that enables rotating the camera 180° relative to the lens. Ideally one would have a reference camera known to be ±5 microns or some such, but I know of no way to be so-blessed, and it’s nowhere close to realistic given manufacturing challenges.
A tilt that causes the top of the frame to focus farther away and the bottom to focus closer is advantagous for many shooting situations (camera in horizontal orientation). But a tilt that causes the top of the frame to focus closer than the bottom is a sharpness robbing disaster in with the lens working against against the photographer.
Master photographer Joseph Holmes (www.josephholmes.com) provides his perspective:
Standard view camera operation for landscape work involves the Scheimpflug principle: nearly always involved intentionally throwing the lensboard (front standard) and the film plane (rear standard) out of parallel by a small amount (e.g. 1 to 15mm sort of thing with a 4 x 5 camera) in order to cause the plane of focus to not be perpendicular to the optical axis. This capability helps to overcome the extremely low depth of field of large format cameras. This is especially effective when the subject is flat, e.g. flat ground, and it’s at widely differing distances from the camera in different parts of the scene.
The three planes, those of the focus, the front standard and the rear standard, always converge in a line (except when there is precisely zero swing or tilt i.e. the lens mount and the film or sensor are perfectly parallel. Typically this means using back tilt or sometimes front tilt to get land in front of the view camera into focus. Small adjustments of the camera cause large effects on the orientation of the plane of focus. Instead of objects nearer or further away than the plane of focus being progressively more out of focus, objects above or below a strongly-tilted plane of focus are progressively more out of focus. So you can stand on a floor with the camera on a tripod, pointing down at say a 30 degree angle from the horizontal and focus on the floor perfectly at a big aperture with a lens with very little depth of field, such as a normal lens on a 4 x 5 camera (150mm focal length). The entire floor area shown in the picture could be in focus. But anything above or below the floor will be increasingly out of focus. To focus on a ceiling you’d adjust the camera’s tilt the opposite way. To focus on a wall that’s not perpendicular to the optical axis, you would use a swing adjustment of either the front or rear standard, where one or the other moves rather like a door being opened.
With today’s miniaturized, high-resolution cameras, incredibly tiny misalignments of the lens and the sensor can have visible effects on the image. A typical sheet of office paper is 100 microns thick (1/10th of a mm), but we can sometimes see swung or tilted focus planes in an image when the camera and lens are misaligned by as little as 10 microns. If we’re working at medium or small apertures, rather than very large ones, tolerable misalignment is apt to be up to roughly 30 microns or higher, for example. None of this matters much if getting an entire scene into focus isn’t your bag. In such cases it would be unusual to find manufacturer tolerances for swing and tilt in either a camera or a lens insufficiently fine. — Joseph Holmes
The ideal scenario is a camera with zero swing and zero tilt, a near impossibility with current manufacturing techniques (meaning in terms of quality control cost). It would require not only tight machining tolerances but extremely fine tolerances for assembling everything, lest errors add up. And then some means of ensuring the sensor was mounted for no deviation from the plane of the lens mount. All of that is surely prohibitively expensive.
The camera and lens add their errors together. If a camera has a swing of +20 microns and a lens a swing of -20 microns, they net out at zero—perfect. But +20 and +20 nets out at +40—not nice at all. Thus a particular lens and camera can work well together, or 'fight' each other.
Expected errors for swing and tilt
For production cameras, tolerances in the ±20 micron range for cameras is a reasonable guesstimate, but it could be better or worse—little data exists. Joseph Holmes and Samuel Chia state that ±10 microns is excellent (the goal of their shimming efforts) and 5 microns would be fantastic, but that is likely to require meticulous shimming with most cameras (tedious and time-consuming).
What can we expect for left/right symmetry?
Joseph Holmes (www.josephholmes.com) writes:
The three A7R II cameras that I measured were tilted and/or swung by over 45 microns in one case, and about 30 microns in the other two cases. I would say that it’s unlikely that many more than half of the Sony Alphas are accurate to better than 20 microns, but I really don’t know. One would simply have to measure a whole lot of cameras, and each one is hard to measure.
This is the most important single finding of all my work with this and I’ve yet to publish it... so as to potentially egg Sony on to figure out how to cut their typical alignment error by ⅔ or so... assuming the small sample I was able to see, three cameras, all from very different parts of the production run, plus Samuel’s A7r II, which was also off by maybe 25 to 30 micron...
For giggles, let’s stipulate an unrealistic figure of exceptionally tight build tolerances ±10 microns of swing and and ±10 microns of tilt (1/100 millimeter). The reality is probably twice that, or more.
A camera with ±10 micron lens mount / sensor parallelism should be considered outstanding, yet at 12mm even a ±10 micron deviation shows an easily-seen asymmetry. The table below makes this plain.
For example, with a 12mm lens focused at 9.8 feet, a ±10 micron swing will push focus to either ~8 feet or ~12 feet on the opposite side. At 15 feet, a ±10 micron swing will push focus to about ~11 feet or ~23 feet on the opposite side. That’s the kind of error is just what I am seeing on the Sony A7R IV loaner camera. Still, it speaks to very tight tolerances from Sony, based on the numbers—and I cannot attribute it all to the camera versus the lens.
Stopping down masks errors to some extent, but my experience says it is not a cure, especially on distance scenes with fine detail across the frame.
It is not just left/right asymmetry but also impaired depth of field on one side versus the other. Both are at work in Big Tree at Creek Bend. On top of all that are all the various optical aberrations. But by all numbers, it is as good as we might hope for.
Focusing distance (feet) versus relative lens extension from INFINITY focus, microns @ 12mm, 24mm, 50mm
