January 31, 2014
Appended February 1, 2014
Sony A7R Shutter Shake, Take Two: More Lessons After Further Research
Please see my first and third articles on this subject, from December 28th, 2013 and April 19th, 2014, respectively.
This page: http://www.josephholmes.com/news-sonya7rshuttershake_followup1.html
Future URL of this page: http://www.josephholmes.com/news/2014/01/31/sony-a7r-shutter-shake-followup1
My own quest for a complete and largely perfected outfit based on the A7R for landscape work is still incomplete, after over three months of working on it, but I've learned a few more important things since my article of December 28th. I'm hoping to have it complete by early May. I'm waiting on two critical lenses from Sony and a second-generation deadweight that I'm building.
First, it is now looking much less likely that a firmware update from Sony can yield a substantial improvement in the amount of shutter shake which plagues this otherwise highly promising camera. Our earlier hopes for a similar firmware update to that provided by Leica for one of the M cameras for much the same problem have diminished as a result of both experiments with shake analysis and an outright denial by two Sony representatives in the U.S. that the problem even exists (despite it being exceedingly simple to show it). A helpful update still could happen though, and it is also our hope that our disappointment at the extent of this problem will at the very least lead to subsequent models all having the electronic first curtain (EFC) feature.
Second, my continuing tests and development of user methods to fight the shutter shake have shown great efficacy and promise and I'm expecting that I will be able to make exposures which are consistently unaffected in any noticeable way with various lenses out to at least 280 mm, on a tripod, with the camera horizontal, and about 200 mm with the camera vertical. That would give me a fine all-around outfit which does what I need it to do, which is great (assuming the lenses work out). Many people will still have trouble with longer lenses, particularly when the camera is vertical, on a tripod, and more so when mounted on a lens foot.
I have to wonder what will happen with the Sony G 70-200 f/4 OSS lens, which is expected to ship sometime this spring. It comes with a collared tripod foot. Obviously, it is intended to work at 200 mm, with the camera vertical, mounted by the foot on a tripod, but it's just as obvious from my own testing with an adapted 70-200 Canon f/4L lens that this simply won't work worth a darn. We'll see after I get the lens, which I'm hoping can fulfill a vital role in my outfit, along with the Sony/Zeiss 24-70 f/4 OSS which should be arriving in less than a week. The short end of that lens will be soft in the corners, but the rest of the range I'm expecting to be excellent. Again, we shall see. I do expect that the 70-200 f/4 G OSS lens will work just fine for verticals if the camera and lens are connected to the tripod via the camera, the lens is supported with a secondary support setup, and an optimized deadweight is clamped to the bottom (side) of the camera for vertical images.
Clearly, a sensor with electronic first curtain, though it can theoretically leave some residual shake during the exposure as a result of the closing acceleration of the second curtain, is likely to give us superb performance. EFC results with other cameras, including perhaps most importantly the equally light weight A7, have shown in other people's tests that the problem is is largely or entirely solved by not needing to move the first shutter curtain at all, to begin the exposure. We can only hope that the next high-pixel count body from Sony in this FE format comes with this vital feature or its equivalent.
But in the meantime, I am, for the moment at least, based on many experiments, feeling quite good about what I can get from the camera with the adapted lenses I have on hand.
I built one finished deadweight, which clamps to the bottom of the camera's L-plate. The beautifully crafted Really Right Stuff L-plate is the most impressive L-plate I've seen them make, though it would benefit from being more rigid for vertical mounting, on account of this camera's special problem with shutter shake. Note that when putting this snugly-fitting two-part L-plate onto the camera, if you want a proper fit, and you have a camera which was sold for what I assume is the North American market, you will need to remove the UL sticker from the bottom of the camera (any non-recessed sticker). Use a piece of 3M #810 Magic Tape or similar to remove the residual adhesive left by the sticker. Beware that removing the sticker may well cause a camera to be considered un-returnable. This little sticker protrudes from the base of the camera by 3 or 4 thousandths of a inch, spoiling the close, metal on painted metal contact of the L-plate with the camera, which is so important for helping to quell shutter-induced and other vibrations.
With the tiny motions caused by the shutter, every thousandth of an inch of sensor motion matters a great deal. These pixels are only about two ten-thousandths of an inch across (five microns in round numbers). Thanks to angular motion of the camera+lens, it may be the case that a camera motion during the exposure of just 1 ten-thousandth of an inch is significantly damaging. OK, I couldn't resist, I just calculated the answer. If you're using a 200 mm lens on a lens foot (the see-saw arrangement as Lloyd Chambers calls it, and the foot is about one third of the way from the sensor to the front element of the lens, then a motion of the sensor in the direction of the shutter movement of 0.0001" (one ten-thousandth of an inch) will likely cause the front of the lens to move about double that in the opposite direction. In that case, the lens will be re-aimed by about 0.0019 degrees, if the length is 9 inches. This will result in blur of about 1.36 pixels, more than enough to be visible in a sharp image. Wow. One ten-thousandth of an inch is 1/40th the thickness of an ordinary sheet of paper. I have seen shutter shake in my own tests of at least 8 pixels, but without using the see-saw setup. So my sensor might have moved just over 1/2000th of an inch in order to completely ruin the image in that instance, were I using the see-saw setup, but since I was not, it surely moved much more than that, perhaps more than 1/1000th, aka one mil, or one quarter the thickness of a sheet of paper.
Here is what the shutter shake looks like at 100% with a Canon 70-200 f/4L lens at 200 mm, f/8 and 1/25th of a second (one of the two worst speeds for this setup) and 800 ISO, with a highly optimized tripod connection and lens support arrangement. This is a dual-support setup, with the camera clamped directly to the tripod head and the lens supported separately, plus the L-plate stiffened greatly by connecting the camera's left strap lug to the top of the L-plate (see picture way down the page). This is a best-case result without the deadweight for a vertical with this lens at the worst shutter speed. All vertical edges are severely compromised and the fine white line at the right is turned into three lines. This is what 8 pixels of motion blur in this particular pattern looks like.
And here is the result with everything the same except that a prototype second-generation deadweight is firmly (well-fitting metal on metal) attached to the underside of the camera (right side of the vertical body) by clamping onto the RRS L-plate. It consists of an Acratech Universal L Bracket and my first-gen stainless steel deadweight (see below) together, bumping up the mass of my deadweight by about 6 ounces to a total of over 23 ounces. So that's the RRS L-plate at around 3 ounces, plus the other two parts for an additional 23 plus. This combination has proven highly effective in all of my tests at 200 mm and shorter, but it was still very much a failure at 280 mm vertical, where the shake was only reduced by perhaps a little over half. There is a very small residual side-to-side blur here, but I am happy with this result, particularly as a worst case, where speeds above 1/50th or longer than 1/10th would show not a trace of blur. Note that the image shake is not simply proportional to the total mass centered on the camera plus the length of the overall mass (like a pendulum). Rather, there are also natural system harmonics which cause non-linear responses, such that doubling the mass of the camera doesn't simply cut the vibration in half. Adding too much weight can give worse results. Happily, so far in my testing, the optimal weight I found for the 200 mm setup also is optimal for my shorter lenses and for that zoom at shorter focal length settings, where the shake is still an issue, albeit a smaller one. Without the deadweight, I would be unable to use this camera for my work. Anti-motion blur software can at least be moderately effective as a last resort, but I certainly wouldn't plan on using it routinely for an entire segment of my captures!
Note: If you are using a Retina display (or similar with double the pixels per inch of a typical LCD), these images are being scaled up to appear the normal size on your screen. If you are also using Safari for OS X, the up-scaling method being used will cause the images to be blurred noticeably, presumably because Apple has chosen to use bilinear as the scaling method a poor choice. Firefox seems to have chosen bicubic for this scaling, which makes the blurring less, but still just visible. Chrome seems to have chosen nearest neighbor, simply turning each original pixel into a block of four identical ones, thus coming the closest to replicating the appearance of the images as they should appear. So if you want to see the detail of these two blur samples correctly, either use a regular LCD or use Chrome with a Retina-type display.
About two months from now, I will have a final second-generation deadweight in hand, to verify whether or not the final configuration is as effective as this prototype has been with lenses of several focal lengths. If it is, my issues with the shaky shutter of this camera are likely to be completely solved. Until I have the full lens outfit, including at least two of the new Sony zooms built for this camera, I won't be able to know for certain.
Here is my finished first-gen deadweight. It always is used by clamping to the base of the camera's L-plate, so it is only usable when the camera is mounted vertically, or when the camera is mounted on some kind of adapter foot or lens foot, if the dimensions of that foot provide sufficient clearance for the clamp of the deadweight to get onto the bottom of the L-plate without bumping into the adapter foot. I believe that the deadweight will not be usable with the Sony LA-EA3 or 4, for example (unverified), but that it will be usable with an RJCamera EOS to NEX smart adapter collared foot (I have tested R. J.'s current design and it does work). It's a difference of less than 2 mm in the R. J. case and probably also in the LA-EA cases. The Metabones EOS to NEX Smart III adapter's foot, if extended with a mounting plate which is offset forward, apparently allows one to clamp to that foot in some cases, but only for vertical images -- not sure. If you're using a Canon TS-E lens on the Metabones Smart III, it seems to work out fine to support the entire assembly by the camera body. Between us, my friend Mike Schultz and I have tried out the 17, the 24 II, the 45 and the 90, and they all seem fine when being supported by the body (not overly heavy or too long for it). The Sony and Metabones adapters have fixed orientation, meaning you've got a problem if you want to mount via these adapters and make vertical captures. To avoid tilting the tripod head 90 degrees, one can use an L-plate for the adapter's foot, but then one is apt to again encounter clearance issues with this coming too close to the camera's L-plate, if you're using one of those too. I see that Fred Miranda used the Hejnar spacer to raise the Metabones foot, then attached a generic-style L-plate for the adapter and got enough clearance to be able to clamp onto the adapter's L-plate for both horizontals and verticals, despite having another L-plate on the camera. That's a relief (it just clears each way). There can also be issues with the deadweight bumping into your tripod head, depending on which kind of setup you've got. I have one friend with an ARCA Cube (among several with the Cube) and he had issues attaching a weight. So it can be hard to work all these details out, depending on which combinations of lenses, mounting methods, adapters, and tripod head you would like to use. And you might also need to consider how this all works with a nodal slide, which naturally makes the camera more prone to shaking.
How much easier things would be if the camera could simply hold completely still! Instead it shakes much more than other focal plane shutter cameras when they are used in mirror-up mode, I think about triple. This is not to say that I don't assume there are plenty of reasons why the engineering of this camera's shutter, given the sensor's inability to use EFC, meant that it had to shake as hard as it does. For example, the space for the shutter is constrained by the small size of the body, which may well require more abrupt acceleration and deceleration than usual (Jim Kasson pointed out this potential issue). And the shutter achieves 1/8000th of a second exposure capability, which wouldn't necessarily affect the shake, but perhaps does for some indirect reason. And as many have mentioned, the fact that this is a live-view camera also requires a double movement of the shutter instead of a single one. Plus the camera is fantastically light. I would bet the latter is easily the largest factor in the higher effective shake of the camera and the sensor along with it. We can see that Sony cut the flash sync speed from 1/250th for the A7 shutter to 1/160th for the A7R, presumably so as to lower the shake of the A7R shutter, and they may have also included a brief delay between the first motion of the first curtain and its second motion, again to minimize the shaking. The great experiment that Mike Collette did to measure the camera's motions with great precision suggests a possible delay there of several milliseconds. There may be little or nothing more Sony or Copal could have done, save to get EFC into the sensor.
Please understand that the extent to which shutter-induced image blur appears in an A7R capture depends on several variables. To understand the basic situation, realize that the great majority of the problem reveals itself as directional blur in line with the direction of shutter motion. So if the camera shakes while horizontal, the blur is up and down. If vertical, it's side to side. Also, the shake is coming from inside the camera, so the primary remedy is to very firmly brace the camera so as to prevent the chassis of the camera from being able to move at all in response to that shake. Some shutter speeds are essentially unaffected. The range of affected shutter speeds is like a bell curve, highest shake at the peak of the curve. The curve is typically a range of up to about a dozen or more speeds, depending on the overall severity of the effect with a given lens and setup, and the peak speed is affected by the length of the effective pendulum of the system, so an A7R with a 300 mm Nikkor tele attached will have a peak problematic shutter speed which is probably close to 1/10th of a second (I saw only partial data from my friend Nick Wheeler's testing with that setup, where a result at 1/4 was awful and at 1/100th was beautiful, and this strongly suggests that his worst speed is well below 1/20th), wherease I see a peak of 1/20th or 1/25th with my Canon 70-200 f/4L, and I saw a peak at 1/100th with an adapted 150 mm lens when no deadweight was attached. With my 70-200 f/4L plus the Canon 1.4X tele-converter, I saw affected speeds between 1/5th and 1/80th, worst in the middle of that range, fading out on both ends. Attaching a deadweight may also shift the peak downward some, but I haven't run test series specifically to look for that, and I think it's at least a distinctly lesser effect than the lengthening, as the physics of pendulums would indicate.
Also, anomalous bad speeds have been observed by two trusted colleagues who both run first-class experiments, Lloyd Chambers and I think also by Jim Kasson, where a speed faster than 1/160th showed blur and speeds even immediately adjacent to it did not. In Lloyd's case it was 1/500th with a Leica 280 mm tele, if memory serves. One 400th and 1/600th were fine and 1/500th was lousy. For some focal lengths (shorter ones) and some setups (those where the camera is clamped horizontally to the tripod head, directly), the problem is minimal to functionally non-existent in my testing. I have not generally tested at speeds faster than 1/200th because my standard test setup doesn't have bright enough lighting to do so, and I am also unlikely to ever use those speeds in my own work, but I would still prefer to check them, so after I get my second-gen deadweight, I will probably run some series outside in full sunlight.
If the camera is hanging at the end of a long lens setup, and the lens+camera combination is supported by a lens foot, it takes very little force to move the camera either up and down or left and right, and similarly little force to pivot the entire system, making such setups highly vulnerable to the shutter-induced shaking, as well as wind-induced shaking. Combine this with a long focal length and one of the worst shutter speeds, and the images will be quite badly ruined. The wonderful zoom feature of the live view makes it extremely easy to appreciate this, as the smallest forces results in wild motions of the image in a soft setup such as a lens foot mount. If you simply do things which make the camera and lens move around less readily, and in particular be less readily re-aimed (pointed in a different direction), it will result in less shutter-induced image blur.
If a combination of effective measures are brought to bear on the situation, tremendous reduction in the harm can be arranged. This includes several measures:
(Assuming a sturdy tripod and tripod head)
1) Making sure that all connections between the camera and the tripod are very solid (metal on metal! Paint is OK but anodized aluminum is harder. Cork and any kind of foam or rubber-like material are a problem.)
2) Stiffening the RRS L-plate for vertical images by connecting the camera's left strap lug to the L-plate, making the plate far more rigid for vertical images. Simply clamping the triangular piece to which a strap can be connected, between the movable RRS L-plate side piece and the camera's body, with pre-loaded force, works surprisingly well. Another friend tried a nylon wire tie and is happy with the result. I saw shake reduction of about one third from doing this alone, owing to obvious stiffening of the L-plate when the camera is in vertical position, but now I'm missing the neck strap option. I may resort to using a wrist strap of some sort, or figure out a way to have the use of the left strap lug for a neck strap and still manage to connect it to the L-plate for more stiffness, but there's probably not enough space in the lug's hole to allow that. Absent this extra stiffening, it's quite obvious when you press against the camera that it can tilt side-to-side when in the vertical position and this weakness is much of what the shutter's vibrations are able to exploit. RRS could potentially make a more rigid L-piece, or a special connector for L-piece to lug. Here is an illustration:
3) For longer lenses, ones where the lens is too physically long or heavy to hang off the front of the camera, with the camera clamped to the tripod, use a dual support setup which employs a method to also support the lens, such as the Y support piece of the Really Right Stuff Long Lens Support setup. I am using their CRD rail, Y support piece, and dual thumb screw clamp setup. Here is a snapshot. Notice how it's positioned so as to be front-heavy, in order to keep the shutter directly over the tripod clamp for greater shake resistance. It's not the easiest thing to get the Y support perfectly centered under the lens, or to adjust the upward force just right. A native-mount lens would work better and so would a better-quality (tighter) EOS to NEX adapter. Novoflex has decided to build one of those for us, about a month after I wrote them to ask for one, explaining why it's become important for some setups with this new camera (not for EOS lenses, which need a smart adapter). For this test setup, I used a Canon body to pre-set the aperture of the lens to f/8, using the trick where you hold down the DOF preview button while removing the lens. Typical use of EOS lenses require a smart adapter to get any control over the aperture, let alone autofocus, image stabilization, and portions of the EXIF data recording. Nikkor G lenses use a mechanical linkage and so an adapter such as the Metabones Nikkor G to NEX non-electronic adapter can at least provide good manual control over aperture. At least with the Nikkor 70-200 f/4 VR G lens, the aperture markings on the Metabones correspond quite closely with actual f-stops, thankfully!
4) Finally, use a deadweight, a block of dense metal of the correct weight, clamped securely to the base of the camera. The easiest way to make a 23 to 24 ounce deadweight is to use brass, bronze or copper, but the bulk is more than I wanted, so I am having one made out of something rather more exotic and more difficult to obtain: a sintered alloy of tungsten, which must be custom cast and then machined. It has a specific gravity of 17.0 so it can be quite compact. It will be so heavy and dense that it could probably cause a considerable foot injury if it were to be dropped in an unlucky manner, but the mass seems to work out OK on the tripod and it won't put excessive force on the lens mount if a long lens were to be mounted by a lens foot, as the mass is very near to the plane of the lens mount. No other metal can compare with the sintered tungsten alloys for use as high density ballast, without getting into metals with other problems, such as prohibitive cost, excessive hardness, etc. Still, brass, bronze and copper work OK and cost less to build than a tungsten alloy part. My first deadweight was type 316 stainless, but it's specific gravity is only 7.85 and so it's not very well-suited for use at the greater mass. Lead would be gross for its awful toxicity. Silver would wind up costing close to a thousand dollars in all (but would be quite pretty until tarnished, and still can't compare to the density of the tungsten alloy). I have to say, it's been fun to get a refresher course in the metals portion of the periodic table.
Only by doing all four of these things, have I been able to get consistently excellent results from my lenses from this tantalizing camera in my testing so far (excluding the 70-200 Canon f4L, used with the 1.4X tele-converter at 280 mm). If electronic first curtain had been included in the sensor, little of this would have been necessary though the methods other than the deadweight are good practice with less shaky cameras too. Although I have done very little testing along these lines, it is almost certain that a deadweight will improve hand-held results significantly, but the results with the new OSS lenses and Canon IS lenses hand-held will need to be determined by many series of experimental exposures. This would show up as an issue with any 70-200 lens, I assume, but hardly ever with a 24-70. Of course, any hand-held camera could benefit from simply being connected to a substantial mass, but the shutter shake is a high-G, tiny shake, unlike body motions, so that portion of hand-held image blur would benefit most from use of a deadweight.