Trying to get consistent and predictable flash exposure using ETTL has been the source of considerable frustration to a large number of photographers. In many cases, this arises through lack of familiarity with how ETTL functions under a wide variety of circumstances. Digital SLR's (and their film counterparts) are complex instruments, with a large variety of settings under the photographer's control. This creates an almost endless number of possible combinations of camera settings, and ETTL will function differently, depending on those settings. There is a fair amount of information available online documenting how ETTL works (see for example the PhotoNotes article by NK Guy), but applying this to real world situations, and deciphering Canon's often cryptic language, is no simple task.
As I have worked through these various issues over the last 2 years, and read other's experiences reported on various online forums, I have continued to think and experiment, so that I could get the proper conceptual framework which I could then apply to real world shooting situations. I have recently discovered what appears to be a completely undocumented, but very real, aspect of ETTL behavior. If someone else has identified this before me, please let me know, but to the best of my knowledge, this has not been defined before. My first clue to this came a while back, as I was exploring the features of my newly acquired Canon EOS-1D. I enabled the feature that would show the Active Focus Points (I'll use AFP for this term in this article), and was mystified that despite my having the lens in AutoFocus (AF) mode, and the camera in One-Shot mode, they weren't always present. I eventually discovered the reason: like many other Canon users, I used one of the Custom Functions (CF) to shift AF from the shutter button to the AE Lock ( * ) button on the back of the camera (this is CF2 in the D30/D60 and CF4 in the 1D). I finally realized that many times I would press the * button to achieve AF Lock, and then would release my thumb, recompose, and shoot. It turns out that if the * button isn't actively being pressed at the time the Shutter Button is pressed, the AFP doesn't show up in the image review screen. Only if the * button is actively being pushed when the shot is taken do the AFP's get recorded, and then displayed, in the image.
Only recently did it occur to me to connect this behavior with ETTL. It is clearly documented by Canon that when AF is enabled, ETTL calculations are evaluative, with emphasis on the area around the AFP. They also state that when the lens is in ManualFocus (MF) mode, ETTL calculations are done in an 'averaged' fashion, across the entire sensor area (see Chuck Westfall's comments quoted from Rob Galbraith's forums, elsewhere on this site). I then speculated that when the * button is not being pushed, there are no AFP's and so ETTL wouldn't "know" what area to emphasize. This would be functionally the same situation as when the lens is in MF mode: if there are no AFP's, and the reading would have to be averaged across the whole sensor area. This would result in a different ETTL exposure than if the * button was being actively pushed when the shutter was tripped, where emphasis would in fact be given to the AFP.
I then set out to use my camera to determine whether or not reality matched the theory. I set my 1D on a tripod, and used the Canon 28-70L/f2.8 lens. I taped the black lens cap from this lens to a mostly white wall in my family room. I set the focal length and distance so that the lens cap essentially filled the center spot metering circle in the 1D viewfinder. I used Manual Exposure mode, ISO 200, f/5.6 and 1/250 shutter speed, so ambient light was not contributing any significant light to the final exposure. I had CF14 set to 1, to disable autoreduction of fill-flash so that there was nothing else being done by the camera to effect flash calculations. The center focus point was manually selected. AutoWhiteBalance was used, and I shot Raw+Small Jpeg. Flash EV was set to 0, and a 550EX in normal ETTL mode was used. The images here are downsampled from the Jpegs, with no other alterations done. The following images and descriptions tell the story.
|
|
|
|
|
|
|
It's pretty clear here that ETTL is pretty much the same in the first two situations: it sees and emphasizes the central AFP; this area is black, causing the overall exposure to be brighter. In the third situation, there is no AFP, so the entire scene is averaged, with no emphasis on the black in the center, resulting in a darker exposure. The following objectively demonstrates these comparisons. I put each pair of images in layers in Photoshop, and selected "Difference" to compare them. If the images are identical, the resulting Difference Image would be essentially black; any differences are thus easily visualized.
|
|
|
 |
 |
This shows that the ETTL calculation is essentially the same when the * button is pushed as in the default situation (CF=0), which is consistent with the theory I'm proposing. And there is a definite difference in ETTL exposure depending on whether or not the * button is pushed when the shutter is triggered, which is also consistent with this theory.
The differences aren't huge, but they're definitely present. How significant they will be depends a lot on what the tonality distribution is in the particular scene being photographed. If the area under the AFP happens to be of similarity tonality to the average of the rest of the image, there will be little to no difference. But if it is significantly different in tonality, then fairly different exposure will be seen, depending on what your thumb is doing. I think it's likely that issues like this, whose effect can vary quite a bit from one scene to another, contribute to the many reports of ETTL's lack of consistency. I don't think there is any actual inconsistency, but rather a lack of understanding how ETTL is operating under the specific conditions and camera settings in any given photograph.
Being aware of the ETTL behavior documented here gives the photographer another option to use when shooting - being able to get a more "evaluative" ETTL reading without having to put the lens in MF and sacrifice AutoFocus function.
Addendum:
Out of curiosity, I also took some exposures with the lens put in MF mode. According to Canon, ETTL would then be averaged across the entire sensor area. I took 2 pairs of images, one pair using a manually selected FP, and the other with the camera in AutoSelection mode, where there is no AF selection indicated at all. I then compared the CF4=3 Button Not Pushed but with the lens in AF mode with shots taken with the same setting, but the lens in MF mode. The results here were odd:
|
(All Shots) |
|
|
|
|
|
|
|
|
|
|
It's a little difficult to see here, but when the lens is in MF mode, ETTL exposure is very slightly brigher when the Center FP was manually selected vs when the camera was in AutoSelection mode. Further, when in AF mode, with the * button not pushed, the AF image was very slightly brighter than the MF when the center FP had been manually selected, while the AF image was very slightly darker than the MF image when AutoSelection was present. These differences, however, were quite subtle compared to the differences demonstrated in the other sections above.
The conclusion I draw here is that with CF4=3, and with the * button not being pushed, ETTL calculation is very close to the calculation made when the lens is in MF mode, but not precisely identical. This is the case regardless of the status of FP selection. This once again is consistent with the operating theory being presented here.
Testing with the Canon D30 and Canon D60
I borrowed my old D30 from the friend I sold it to, and tested it using exactly the same conditions as I did with the Canon 1D. I found that when AutoFocus was relocated to the * button via Custom Function 2, that ETTL metering was the same regardless of whether or not the * button was actively being pressed when the shutter was tripped.
Charles Beasley did some excellent testing, using a very similar protocol, using the Canon D60. He found that the D60's ETTL metering was the same as the D30's (and different from the 1D's). He presented his findings in this thread at Rob Galbraith's D30/D60 forum.
Thus, in the D30/D60, ETTL metering always gives emphasis to the center focus point area, as if the center focus point was active (even when technically it isn't active, since AutoFocus had not been initiated). The only way to obtain a purely averaged ETTL reading in the D30/D60, with no emphasis on any focus point area, is to put the lens in Manual Focus mode.
I was hoping this 1D "trick" would work for the D30/D60 as well, giving those photographers another option in their ETTL shooting, but this turns out not to be the case. Disappointing, but in a sense not that surprising, given how different their autofocus and metering systems are overall.