ADC response of a Canon 40D

This past Thursday I upgraded my camera. I had been shooting with a Canon Digital Rebel XT for the past two years, and while it’s served me well, I had an itch and decided to get the 40D.

While I was cruising around the web, drooling at different cameras, I ran into a web page (I can’t remember how I got there) that suggested a method to determine (in a somewhat crude fashion) the dynamic range of the sensor in a digital SLR.

The technique suggested is really simple. Using your fastest lens, take a picture of a neutral subject (white, grey, etc). Keep the aperture at wide open and set the shutter speed such that you are exposed as bright as possible without an overexposure warning. Then, keeping the same shutter speed, stop down by full stops. The histograms will show where the camera is digitizing the luminosity info from the sensor.

I did this for my new camera and found the results rather interesting. Note that the “highlight tone priority” was turned off, and ISO 100 was used for all exposures.


Composite histogram for exposures with an aperture ranging from f/32 to f/2.8

The sensor / ADC is not linear.

If it were linear, then you would see (within whatever standard deviation) a histogram value step down by half for every stop of down of the aperture.

The histogram counts for the given aperture values:
(F-stop :: Histogram Median)

f/32  :: 7
f/22  :: 15
f/16  :: 31
f/11  :: 55
f/8   :: 109
f/5.6 :: 153
f/4.0 :: 206
f/2.8 :: 241

While the first few lower entries appear to be linear, the higher values are decidedly not.

Using Excel, I plotted the median values and then did a trend line to approximate the type of curve that would best fit these results.

Picture 1.png

Curve fit of the observed histogram median results

While the curve fit shows a power based curve seems to fit, it sort of appears to be two different linear lines to me. But, it’s hard to say just looking at it.

Certainly from a technical standpoint, there’s no reason why the ADC has to be linear. While most are linear, having programmable reference levels wouldn’t be that hard to do.

My conclusion: I’m not sure. I believe that the composite histogram itself is telling. The “Expose To The Right” idea does not seem to be correct. The original idea was that the linear ADC (sensor + ADC, of course) would mean that after the highlights, you would immediately lose half the information. The histogram seems to say otherwise.

Does this mean that the end results from the ETTR are bunk? Probably not. But, it does seem to indicate that there isn’t nearly as much to be gained from the approach of controlled over-exposure.

Though I have not collated results for the “highlight tone priority,” I can say that there seems to be a shifting of the “sample points” from these results.

Technorati Tags:
, , , , , , , ,

    • Stephan
    • January 6th, 2009

    Are you sure you tested in linear colorspace and without any tone-curve/contrast applied?
    If not: you only measured RGB-Gamma and the settings used for postprocessing…

      • Nick
      • January 6th, 2009

      Truthfully, I don’t remember what I measured, but if I had to guess, I’d say that I didn’t convert to linear. Have you tried this experiment with the converter set to linear?

    • Stephan
    • January 7th, 2009

    Hello Nick, haven’t tried it. Anyway I only have a 40D.

    But check out this (first picture):
    Your values seem to match this tone curve quite well.

    The only way I know to do a test like this would be to use raw and convert with dcraw to linear 16-bit mode. Most other raw converters will apply some tone curve to increase midtone contrast and preserve highlight details.

  1. No trackbacks yet.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: