D800 Autofocus Problem

INT South-Africa D900 0001 bythom.jpg

I've mentioned seeing D800 bodies with a focus alignment problem and first reported on this extensively in mid-2012. I was relatively quiet for a long time about anything other than reporting that I'd seen the problem. Part of that had to do with my trying to finish my review and book on the D800, part was that I'd been doing more investigation.

After studying both the real data (things I can look at and test myself, including two bodies with the problem), second-hand information from people I've come to know and trust, plus some random data (e.g. postings on the Internet, assertions in emails, etc.), I came to the conclusion that there was a real issue affecting a group of early D800 purchasers. How big that group is, I don't know. It's really, really difficult to judge, especially given that some people who've shared samples with me don't appear to have the problem in question, but rather other issues. I'll have more to say about that in a bit.

I came up with four questions that needed to be answered:

  • What's the nature of the problem?
  • How do you test your camera to see if it has the problem?
  • What's Nikon procedure for fixing the problem?
  • At what point did new D800's coming out of the factory begin to be tested for that problem and fixed before sale?

I try not to rush to judgment. Indeed, upon verifying that I could see a real issue that was different than I had encountered before, I reached out to contacts within Nikon about the problem. Eventually, one was willing to give me Nikon's official on-the-record position, which I'll quote verbatim here:

"Any customer who is having an issue with their Nikon product should contact us [Nikon] directly so we can troubleshoot and advise on a resolution. There are many "me too" reports of various issues on-line and it's impossible to comment on a specific issue until we've had the opportunity to review sample images, provide troubleshooting steps or actually examine the product in question."

Yep. It appears that some politician's campaign staff crafted that answer. It also completely avoids mentioning the fact that Nikon deployed new tests to the subsidiaries that specifically target and correct this particular problem. Doesn't seem like they'd do that unless there was a real problem that needed fixing. So let's step through my questions and see what we come up with.

http://bythom.com/Images/aftest.jpg
  1. The nature of the problem is simple: some AF sensors are providing improper focus information to the camera. This appears to be almost always be the left side line sensors in the CAM3500 (the three center columns are cross sensors). In the two cameras I inspected, there was clearly something different in the left line sensors than what I saw from the center cross or right line sensors. Depending upon how you test (see below), you can sometimes see slight differences between the right line sensors and center, but that's not the same thing as I saw on the two bodies in question: clear out-of-tolerance results on the left, while the center and right were basically as I'd expect them.

    So what's the problem, really? The D800 has internal tables that help it understand the actual position of each of the focus sensing modules. It uses that position data in the math that the camera performs to determine focus. It appears that in some cameras, this data needs to be redone. How the information got to be wrong in the first place, we don't know. I speculate that Nikon had at least one camera calibration station in the manufacturing facility that put incorrect information into the cameras going out. But that's pure speculation on my part. I can't think of another likely reason, though. Perhaps my imagination is off and there's a different reason I'm just not seeing. But whatever the cause, Nikon isn't sharing that information.

  2. So how do you test? You need a fairly large flat surface like a large wall. On that wall you need to put test targets. Star targets are not preferred. The left autofocus sensor that you really want to test is oriented to better detect information vertically, and most star targets don't give you much of that. You want targets something like the one shown at the top of this numbered section

    Note: in Live View you may have trouble with this test pattern, especially if you are shooting in low light. If necessary, place the pattern somewhat diagonally for the Live View test if you have troubles focusing on it with the bars in the horizontal orientation. An alternative chart that works okay would be the USAF focus charts.

    The problem is more easily noticed in fast wide to normal lenses, so use a 28mm, 35mm, or 50mm prime. Which you use will dictate how big that flat surface should be and the size of your test targets (no, I'm not going to detail every last little bit, as there are too many combinations and permutations here). I'd argue that you should be placing your camera at a "normal" shooting distance. Some people are testing far too close. 6-10 feet (2-3m) sounds about right to me with a 35mm lens. Your camera needs to be level and parallel to the wall. That step alone can trip a lot of people up.

    Assuming your setup and test charts are appropriate, the testing itself is simple enough. 
    1. Shoot at base ISO and maximum aperture (preferably with a fast lens; slow lenses will mask the problem with depth of field). 
    2. Turn off lens corrections if your camera has them (the D800 has Auto Distortion Control, for example). I prefer to shoot and evaluate raw files, but I don't see any real harm in using JPEG as long as you don't crank up sharpening, contrast, or other controls that may hurt your ability to evaluate the results.
    3. Set the camera to Live View. Rack the focus to infinity manually, then use Live View autofocus to focus at the leftmost sensor position and take a picture, rack the focus to infinity and autofocus on the center sensor area and take a picture, and then rack infinity again and use the rightmost sensor area for the last picture (you're going to have to do some judging there, as Live View AF doesn't give you the 51 normal positions and is slightly different in size for focus). At the end of this step, you should have three reference shots (left, center, right). They should all seem to be in relatively good focus. If they're not, do not pass go, do not collect your $200. I should point out that in talking to a number of people about their "D800 focus problems" and their testing of it, many of those conversations never get this far in the testing sequence. They've done something wrong prior to this point, so we can't actually believe their results until we correct that. In this respect, Nikon's "contact us...so we can troubleshoot and advise on a resolution" comment is dead on. Assuming, of course, that Nikon gives the right troubleshooting information and advice, of course ;~).
    4. Now do the same left, center, right test using regular shooting with the main phase detect autofocus system. Remember to rack the lens to infinity between each test. Compare these images against your Live View images. In the case of the first two problem cameras I saw, center and right were close enough to the same between Live View and phase detect autofocus that I wouldn't quibble. But the left sensor had a very visible, large, and obvious difference. If you see that in testing your camera, contact Nikon and tell them what you found and wait for their response. Note: Some people report that an AF Fine Tune corrects the problem. No, it doesn't. This is a tricky area, and one reason why I suggested avoiding AF Fine Tune in the first place. It is not uncommon for a "good" camera and lens to have somewhat different AF Fine Tune values for left, center, and right sensors. A -10, 0, -5 wouldn't be surprising to me, and doesn't indicate the problem we're looking for. Generally, a misprogrammed left sensor response can't be brought back by AF Fine Tune. In other words, we're not looking for a modest difference in focus performance; we're looking for a gross and obvious difference in focus performance.

  3. Should you Test? Yes, especially if you got a camera shipped to you prior to July 2012 or buy a used D800 that might have been made in that initial period. One thing that you don't generally see in other discussions of the problem is that you can't really ignore this problem if your camera has it. Why? Because the only safe way to use a camera that has the problem is Single Point autofocus using a known good sensor (e.g. center). Auto Area and 3D Tracking use all the focus points, so if one or more of them are out of whack, you're going to get less than reliable focus results. Likewise, Dynamic Area is risky, too, because if the camera starts moving the sensor left to follow a subject, well, we all know what's out there.

  4. What's Nikon's procedure? It appears that Nikon's response is this: they'll ask you to send in your camera if they're convinced that you may have a real issue (which usually involves sending them test images, by the way). They'll put it through the new calibration test that's been deployed to the subsidiaries and this will update the internal camera tables for the focus sensors. If that makes everything look good, they'll send it back to you and your camera should be fine. If there are still problems, they then have to examine a lot of other potential alignment issues (mount, mirror, sub-mirror, focus sensor module, etc.). NikonUSA has been marking this as a Level B1 repair. For some reason, all the Canadian repairs I've seen are marked Level B2.

    If you buy a new camera, take it home and test it, and find yours has the left-sensor problem, immediately contact your dealer. Ask them to consult their Nikon rep for advice on what to do. If the camera is in new condition, Nikon may authorize an instant return and swap out. Other dealers will handle the shipping back to Nikon for you. It can't hurt to ask.

    Update: Cameras “fixed” by Nikon tend to be coming back with a +10 to +12 bias to AF Fine Tune. I reported that to Nikon and it appears that the test lenses/cameras Nikon sent to the subsidiaries all have that bias in them, so Nikon is essentially re-programming to a slightly out-of-spec camera. They claim it isn’t out of spec, as +10 is within tolerances. But the problem is that lens tolerances should form a bell curve around 0, not +10. Thus, you may find it difficult to AF Fine Tune some lenses after Nikon fixes the camera. The only solution then is to send both camera and lens to Nikon for repair.

  5. Are new cameras okay? The new test equipment and programmer appears to have been deployed to subsidiaries in July 2012. One would hope that this also means that no new D800's coming out of the factory would need this procedure done, but there's no way to tell when your camera was made, so that information isn't overly helpful. At present, I believe the break over point for US D800 bodies was serial number 3055000 (which implies there were 55,000 bodies made before the change sold in the US). I don't have numbers for other countries or for the D800E.

But here's the problem: the Internet Amplification Effect has a tendency to make us all camera hypochondriacs. If two people report a problem, we all think we have it and look for it in our gear. The problem is the same as with medical hypochondriacs: you get a lot of false positive reports.

Autofocus isn't quite what most people think it is. Phase detect autofocus has always traded speed for focus tolerance. The system looks at a split set of data received at the autofocus sensors, measures the difference, then tells the lens to move the direction and distance it calculates. Lenses are not 100% repeatable (I'm actually amazed at how repeatable they are). If you perform the same infinity to near focus operation 100 times, you will get a bell curve of focus around a central position, not one position repeated 100 times. This was first reported to the public by the late Herbert Keppler in the 80's, I think, and then substantiated in a series of tests by him and his staff in the 90's. The tolerances have dropped since then (and I'm surprised at how small the tolerances now are, as we may be in the sub-micron level with some lenses), but there are still tolerances. Two lenses made from the same factory on the same day from the same parts by the same person may perform slightly differently, which is one reason why we have AF Fine Tune.

So all those camera hypochondriacs hear the Internet Amplification Effect about "the focus problem," go out and try to focus something with the left autofocus sensor, see anything other than snap sharp focus, and decide their camera suffers from the left-autofocus sensor problem. I've seen six instances where there actually wasn't a left sensor problem, but something else was amiss, including some erroneous user assumptions in more than one case.

That's as much as I know. Unfortunately, Nikon's silence on this issue made it difficult to even figure this much out. I have had a couple of Nikon folk say something off the record, but I honor off the record comments; I don't report anything I learn from them unless I can find multiple sources of the same information from elsewhere.

My recommendation is this: test your D800. If you find a problem, call Nikon. Provide your evidence and let them respond. 

The one thing I think Nikon is getting wrong here is this: Nikon knows they messed up and shipped some number of defective cameras. Yet they are defacto forcing the customer not only to detect the problem, but to pay for shipping of the unit back to Nikon for repair. I think the least Nikon could (and should) do is this: for every camera they receive that actually gets fixed by this new recalibration test, Nikon should refund the shipping cost to the customer. If I were in charge, I'd go further than that and try to find some small reward I could include above and beyond that, along with a note of apology.

Yes, that's costly. But so is a US$3000 camera that doesn't focus right. Why is the customer the one that gets punished here?

You Know the Song
I'm sure some of you are now a little down in the dumps about your D800. Don't be. Maybe if you sang a little song while you tested your camera it might cheer you up. So here are the new lyrics to that tune that seems to be everywhere. Try singing them as you test:

    Hey, I just bought you
    And this is crazy,
    But here's my test chart
    So focus, maybe?

A Few More Answers
Response to what I wrote above ran very high and triggered lots of questions. So in this section I'll try to deal with those.

  • One common question was "what about AF fine tune?" I don't recommend using it prior to testing for a reason: it would take another long article about how to do that right. Moreover, in discussing this with some people who've done it, they've actually set their AF fine tune in a way that masks the problem. Let's say that your left sensor is out, but not tremendously out. If you set an AF fine tune value between what the left sensor needs and what the center sensor needs (nothing), you'd balance out the results you see. The left sensor would be out of focus one direction, the center the other. On a flat test chart on a wall that will look the same, so you'd probably report "acceptable-but-not-great results" when in actuality you have a clear difference between AF sensors. I'd suggest that you test with the camera the way it came from the factory for that reason, at least initially.
  • Another common question is "do different lenses show different results?" The answer to that appears to be yes, but I'm not entirely sure why that would be unless there are multiple tables in the camera that are involved here (lens with focus shift or field curvature would have problems with flat test charts). Since we don't know exactly the data in those tables and the calculations that are made with that data—we know only the general function—I'm not sure what to make of the fact that you can get different results from different lenses. I understand that Nikon normally uses a 50mm lens on their test rigs, though some second-hand information I received suggests that sometimes they use a 24-70mm.
  • "What about the fact that fast lenses aren't always sharp wide open?" Yes, that's just one of the factors complicating testing. However, we're not so much looking for the overall sharpness of an image as a clear difference in sharpness triggered by which sensor we use to focus. Even with a lens that isn't fully sharp wide open I've found easy-to-notice differences between the left sensor and the center/right.
  • "My camera focuses fine on vertical lines." Are you sure you're in Single point autofocus? It does appear that 3D tracking uses input from more than just a selected sensor (so does Auto area, but that wouldn't let you select the sensor to test ;~). And you may have more vertical contrast than you think in your test chart, and enough of it may fall on the test sensor to work. One person sent me an example and I could see small horizontal shadows from the test target not being absolutely flat; camera probably picked up on that.
  • "Does it make a difference how bright the light is?" Yes, it does. I'm not sure about Nikon's current focus sensors, but their old ones actually had two components, one used in brighter light, one used in darker light, and these were slightly offset from each other. This could also be part of the answer to the previous question, as well, that there is some small discrimination ability in the 90° orientation. If at all possible, test using outdoor daylight. If not, make sure you have plenty of non-frequency-based light on the chart.
  • "Is it possible to see contrast detect errors during testing?" Yes, absolutely. Live View autofocus is skipping lines on the D800/D800E, so detail can fall between the "cracks." Live View on a D800 is much more sensitive to horizontal detail rather than vertical detail.
  • "I really can't use a star chart?" No, you can, especially if you make sure you've got plenty of differing detail directly underneath the focus sensor being tested. However, the problem we have is alignment. Technically, you'd want the center of the star exactly positioned on the center of the autofocus sensor being tested. Unfortunately, you'll be guessing at that as the focus sensors in the viewfinder don't always align exactly to the focus sensor positions in the body (nor are they the same size and shape). The reason I suggested a bar chart is to make sure that we didn't have to know exactly where the center of the focus sensor was. The USAF focus charts are also acceptable for doing the test.
  • "What about field curvature?" Yes, field curvature (and focus shift if you're not using the lens wide open) can have impacts on what you see. Technically it's the reason why I insist on doing three live view focus tests (left, center, right). That way I have reference to any side-to-side changes that would occur because of lens characteristics; you should see the same characteristics the same way in the phase detect results. This gets a bit tricky, though, as we're all using subjective evaluation on our test results for the most part. As I noted in an earlier article, sometimes we see what we think we see. I'm sure this must be a huge issue for Nikon. Some people have a real problem and Nikon will want to fix that. Some people are seeing problems that aren't there, and if those bodies go back to Nikon is just slows everyone's repair times. And of course, there are some in between situations, where it's unclear from the results whether a body does or doesn't have a focus problem.

Unfortunately, we're in a Fog of War type situation here with an untrained army. The fact that there are so many small details that play a part in trying to get reliable test results and the fact that Nikon is completely silent on any information that would help users figure those things out is part of our problem. I've outlined what I think are the best steps to take to try to figure out if your camera has a problem. I can't say with certainty that those are the right steps, only the best I'm able to come up with based upon what little we do know and the testing I was able to do briefly with an affected body.

If I'm wrong in any of my instructions or assumptions I'd like to know how I'm wrong so that I can correct them. [Two years after this article first appeared, no one has done so, not even Nikon, who basically endorses my test approach behind the scenes.] I can say that it seems to me that a problem body really is almost always very clearly wrong. If you're getting vague results or ones that are difficult to interpret subjectively, I'd guess that your camera doesn't have the problem I've been investigating (or there's something wrong with your testing protocol). There's just been such a dramatic difference on the bodies that have the problem that I'd tend to recommend that you not over obsess unless you see a highly dramatic and unexpected difference.

Finally, a very cogent comment from one reader in response to my comments about what action Nikon should take: "why would Nikon want ANY user to have a defective camera in their hands?" Very well put. This is a US$3000 product that represents the best of what Nikon can create. When it works as designed it's exceptional in performance and a great ambassador to what makes Nikon products good. But if Nikon knows they shipped some number with faulty data stored inside, why wouldn't they want all of those back to update? Do they really want a common comment from the user base to be "yeah, it's 36mp but it doesn't focus very well"?

More and more it's becoming clear to me that Nikon's immediate response should have been "oh dear, we've made a mistake, let us fix that for you." I remember saying that same thing about another Japanese company in recent years: Toyota.

A Haiku of Sorts
Okay, I've depressed you again. Time for another humor interlude.

As I parsed responses to my focus survey I discovered that the software I use has the ability to bring up Word Clouds for questions where I allowed an open-ended text response (basically an analysis trying to show the most important words and phrases it thinks it found). The first time I pulled that up I found a very interesting haiku that made me laugh. As more and more responses came in (now over 1000 tests cataloged, keep them coming), the haiku changed. In fact, it's gotten downright specific and to the point. I still laugh when I see it.

Here's what I saw this morning when I pulled up the Word Cloud for "Which of the following characterize what you found?" (This was an open-ended "other" response; the Word Cloud comes from that): Sensor Soft Worse than the Right.

Sensor Soft Worse than the Right. OMG. Maybe I should just ask all open-ended questions in future survey questions ;~).

Now, just so that you don't think that people were typing something close to that in the open ended responses, here's a typical one out of 206 text answers: "After I had fine tuned 105 Macro, 70-200, 24-70, 200-400 these were OK with left/right slightly less accurate than center, but balanced. My 14-24 f/2.8 showed back focus on left and front focus on right at both 14 and 24, more pronounced at 14mm. No problems on same test done with D700, and no problems with Live View." Other answers were cryptic: "mild to moderate."

Those of you who took the survey know that there's another open-ended question in it where I asked those that reported their findings what Nikon's response to them was. Dare I look? What will the Haiku-enator say to that question?

Pictures Problem Response Returned, Sample Images Send, Shop Waiting.

Okay, so the commas are mine. Still, seems like a very valid response. I'll just note that the biggest word in the cloud was "Send." Who says computers don't know what we're thinking?

Looking for gear-specific information? Check out our other Web sites:
mirrorless: sansmirror.com | general: bythom.com| Z System: zsystemuser.com | film SLR: filmbodies.com


dslrbodies: all text and original images © 2024 Thom Hogan
portions Copyright 1999-2023 Thom Hogan
All Rights Reserved — the contents of this site, including but not limited to its text, illustrations, and concepts, 
may not be utilized, directly or indirectly, to inform, train, or improve any artificial intelligence program or system. 

Advertisement: