For people who are interested in going down the rabbit hole of trying to improve the performance of their technical camera lenses, here are some general comments about calibrating lenses that might be useful, and a quick sketch of my trouble shooting and correction approach. This all comes from a program I completed at the School of Hard Knocks.
So take it for what it's worth.
Expectations Management
As I explained at the start of the thread, there are a lot of links in the "performance" chain. Sometimes the lens isn't the problem. And sometimes you just can't achieve a big improvement, or even any improvement. Furthermore, sometimes people form unreasonable expectations. It's important to learn from the MTF chart what you can reasonably expect to see.
For purposes of this post, let's assume that you've ruled out the other links in the chain. In other words, the sensor is aligned properly, and the technical camera is in good alignment when everything is set to the neutral or zero position. To be confident that the problem is the lens you're working on, it's important to try your other lenses in the same test setup (see below). If I can set up in front of my test wall and get excellent results over the whole image circle with one of my lenses, and then I put on the suspect lens and image quality falls apart, I can be reasonably confident that the suspect lens is the weak link in the chain.
Something else that some people forget is that every lens has field curvature. It’s a property of lenses, and the lens designer accounted for it in creating the design. Field curvature can vary over the image circle, and the shape of the field can change depending on the distance to the subject. In this vein, some lenses are designed for specific distances, so we shouldn't expect good performance at other distances. For example, the Schneider-Kreuznach Makro-Symmar 120mm is superb at 1.5 metres and closer, from wide open. However, as a taking lens at longer distances, it was terrible until f/11. There was nothing wrong with the lens; that's its design. There are
other tests you can use to check field curvature. I have some demonstrations of this test in my
article on the Schneider-Kreuznach WA-Digitar 28/2.8.
The nature and behaviour of field curvature is important because it can be "tweaked" a bit with tiny changes in cell spacing. However, with a lot of lenses, you must decide if you want best performance at infinity, or close up. Lenses with floating element designs give you the best of both worlds, but technical camera lenses focused by the camera body use simple unit focusing designs; thus, the lens designer may have explicitly traded off close performance for far performance, based on assumptions about how most people would use the lens.
Stopping down the aperture generally masks field curvature by increasing depth of field. However, in many cases it's not reasonable to assume that you can focus in the centre of the unshifted frame, and then perform a large shift and still get acceptable focus. By shifting, you may be moving out of the part of the field that was in focus in the centre of the unshifted image. This is why people who are shooting a flat surface like a building facade straight on will often focus near the top, and will re-focus after shifting.
Finally, in terms of field curvature, the field is not necessarily a nice parabola. It can be wavy across the image circle. I have lenses like this where the image is sharp in the centre, softens on shifting towards the edge of the image circle, and then sharpens up again at the edge of the image circle.
Creating a Test Environment
If you are serious about checking and calibrating your lenses, you need to create a suitable test setup. It doesn't have to be complicated or expensive. However, it does need to create a controlled environment where you can reliably repeat the test. A brick wall outside is not a reliable test setup. It's good for a quick check, but it doesn't provide what you need. This is what I use. Pardon the mess on my workbench. I was in the middle of a project!
This is a wall in my basement "studio". I have confirmed that the wall is close enough to vertical, and flat enough across the width for this purpose. Those are Siemens Stars printed on 8.5x11 Epson matte paper. The file is a vector drawing as a PDF. I exported it as a very high resolution TIFF and printed on my Canon Pro-1000 on matte paper; you must use matte paper because reflections on glossy or semi-gloss paper spoil the test results. The papers are not perfectly flat on the wall, and they are not perfectly aligned relative to each other, but it doesn't matter at the distances I'm working. The PDF for Siemens Stars is available here:
https://commons.wikimedia.org/wiki/File:ImageTestStarJMW.pdf
This is what the central star looks like. If printed properly, the radial lines should come to a clean central point without smearing together.
I set the stars up on my wall so that I can include all but the outer column of stars at left and right in the unshifted frame. That position lets me check performance across the centre of the frame (the middle row) and the four corners in an unshifted image. I use the outer column of stars to measure lateral shift performance. The stippling on the ceiling, and whatever happens to be sitting on my table, gives me additional information about field curvature.
You need to be properly aligned to the target for this kind of test to work, and to gain repeatability. For example, when my camera is 62.75" off the ground and 76.5" from the wall to my left, I know I'm in the right spot. The floor is concrete and I'm not near a vibration source like a train or a highway. I can leave the tripod and camera set up as long as necessary without worrying about wind, people messing with it, etc. When I'm calibrating a lens that I know is going to be tricky, I set up my laptop and tether so I can see the results quickly and easily. I cannot get these conditions in an outdoor situation.
How do I Use this Setup?
As noted above, I set the camera up square to the Siemens Star centre target, in a position that puts the outermost target at the edge of the fully shifted image circle. That leaves a group of targets in the middle for the initial adjustment. The process of finding the ideal spacing goes like this:
- Focus on centre at wide open and check edges and corners unshifted with cells in their current tightened position. I use the camera's LCD for a rough assessment, but I make photographs from wide open to f/11 for careful inspection. I'm looking for maximum possible image quality, and consistency from side-to-side and in all four corners. Importantly, I'm accounting for expected performance based on the MTF chart. If the MTF chart is telling me that the edges and corners will be crap wide open, then crappy edges and corners are not surprising. However, if the MTF chart is telling me that f/8 should be good, then I want to see good f/8 performance.
- If image quality is consistent across the frame unshifted wide open, then I check image quality wide open at maximum shift. If image quality is excellent at maximum shift wide open, on both sides, then I'm done.
- Note that "excellent" is a moving target in this case. Image quality wide open at maximum shift with high magnification will never be as good as the centre. As a result, I always check if small adjustments make the edges better.
Let's assume I'm seeing problems, or I want to know if there's room for improvement. What happens next depends on whether there's adjustment room.
- If the cells are screwed down tight to the mounting surface in the shutter or housing, then I will check if increasing cell spacing a bit by loosening the front cell improves image quality. Normally it doesn't, but sometimes it does so it's worth checking. If I see an improvement, then I go to the next steps and start testing with different shim thicknesses. You can’t loosen the cells too much or you’ll get droop, which messes up the test. As a reference point, a cell that has a 0.5mm thread pitch (like Copal 0) moves in or out 0.5mm with one full rotation. Therefore, a quarter turn of the cell is like adding or removing a shim that is 0.125mm thick. This is only a rough estimate because the cell is not perfectly parallel when it is loose.
- If image quality is poor at the edges of the unshifted frame, there are no shims, and it doesn't get better increasing spacing, then I'm out of luck unless I'm willing to modify the housing. This is possible, but permanent so I try to avoid doing this. A light sanding of the rear mounting surface will allow the rear cell to go in farther. You must be extremely careful because it may only need 0.01mm. There's also a risk of putting the glass of the rear cell in the way of the aperture if you sand too much. Of course, if you have the shims you can put it right back where it was originally (so this isn't as daunting and dangerous as it sounds). Just remember that if you sand the rear mounting surface, it's no longer a "stock" housing.
- If there are shims underneath one of the cells (usually front, but could be rear), and I've confirmed that the cells should be closer, I take out the shims and check whether less space between cells improves image quality. On a GFX camera, because of the thick cover glass, I've had to move cells of all my wide symmetrical lenses in closer than the factory specification. That is only possible if there is enough room due to the use of factory shims.
- Generally shims are supposed to be under the front cell or the rear cell. Sometimes it doesn’t matter, but sometimes it matters a lot. If I’m not making progress with different shim thicknesses, I’ll try shimming under the rear instead of the front, or vice versa, if this is possible. It’s easiest to do this with Copal 0 lenses where the front and rear cell threads are the same. Other lenses have different thread diameters, which means the shims that work on one side won’t work on the other.
- Assuming I have adjustment room and a supply of shims, I do a lot more tests. Image quality can be fine at the edges unshifted wide open, but bad shifted. Once I get to good quality unshifted wide open, I shift all the way left and all the way right wide open, checking on the screen and making comparison photographs and notes as I work.
- At this stage, I make three photos per shim spacing change: centre, full left shift, full right shift. I label them carefully and make lots of notes as I compare. Usually what happens is that the unshifted image is about equally good with a range of cell spacings, but maximum shift is only optimum with one. Another common scenario is that wide open never gets particularly good at maximum shift, but it becomes acceptable or very good as you close down to apertures you'd use. If I can get a lens to very good at f/8 with maximum shift, and excellent at f/11, then I'm happy. If it requires f/16 to be decent at max shift, then I'm not happy.
You need a supply of shims of various thicknesses to reach "perfection". Misumi makes steel shims in various thicknesses that fit most Copal 0 shutters and housings; they are not exactly the same size as proper Copal 0 shims, but they work very well and are available in 0.01, 0.02, 0.05 and 0.1 mm thicknesses. A 0.01mm change in Copal 0 cell spacing is 1/50th of a turn of the cell. With high performance wide angle lenses, I have found that 0.01mm can make a noticeable difference.
Some final thoughts on this:
- It's very common for centre image quality to be good at every spacing I test that is close to correct. The problems are always at the edges. Problems take several forms. One side can be good but the other bad. The good/bad side can switch depending on cell spacing. Left and right can be good, but top or bottom bad. All sides can be equally bad, and never reach good. Because of the way field curvature works, it’s even possible that the farthest shifted edges are better than parts of the shifted image that are closer to the centre. This is a “dip” in image quality.
- With some lenses, image quality at the edges will never be particularly good, and making it hard to tell when it gets better. Some lenses are just weak at the edges. I wouldn't use a lens like this for my setup if I had a choice, so it's helpful to have MTF charts or collective experience to know which ones are worth doing.
- It's common to have to settle on a compromise where neither side is as good as the best that is possible on one side. I'd rather have three sides be 85% of the best possible performance wide open at maximum shift versus one side be 100% and the other sides be 75%. If you can reposition the lens on the board, or reposition the board on the camera, you may be able to set up with the "bad" corner in a location where it will do the least harm, e.g., up for sky if you mostly shoot landscapes with skies.
You can easily get lost in the weeds doing this. Nobody is going to use a lens like the APO-Digitar 35/5.6 at maximum shift and f/5.6. Nonetheless, it's worth getting it as good as possible at f/5.6, because that means it's going to be superb at f/8 and f/11. People I know who have this lens only shoot it at f/11 because they think it can't be good wider; it can if it's properly adjusted.
Finally, in terms of this stage, I have failed several times to successfully shim lenses. Sometimes it’s because the lens was damaged in a way that is not visible but affects optical performance. This is frustrating, but when we're dealing with lenses that have been out and about in the world, sometimes for decades, we may just be out of luck.
Testing in the Real World
If you did a good job against your indoor test setup, you still need to check real world performance at distances you typically use. Recall that the lens designer may have prioritized far over near, so if you calibrate for near using a test setup like my Siemens Star wall, you may be throwing away far performance.
It's entirely possible that you can get excellent performance at the distance to your test wall, yet reduced or poor performance at infinity. This has happened to me many times. I have a few locations I use for infinity performance. This is a favourite because there are features many km away from where I'm standing, plus lots of foreground that lets me check field curvature. With a scene like this, I can gain an excellent sense for real-world performance. I can also shift left, right and down to evaluate shift performance.
This test confirmed that my WA-Digitar 28/2.8 could be a bit better at infinity. I adjusted calibration because I wanted better long-distance performance.
This test created a panorama using 15mm of left shift and 15mm of right shift using my Fujinon-EX 75mm f/4.5 and confirmed that at f/8 it's a strong performer at infinity even with full shift on my MAB II outfit.
