ETTR — just crank up the ISO? Part 6

I’ve been wrestling with some lack of repeatability in the measurements I’ve been reporting on in the last few posts. I haven’t completely solved the problem yet, but I thought I’d report on a series of measurements that I made yesterday.

I worried that, because it looked like some of the SNR results might be dependent on the order in which I made the test images, that I was seeing SNR effects due to sensor heating. To test that, used a D4 to make a series of exposures at a constant f-stop and a constant shutter speed (1/30 second) at ISO 100. I made a continuous string of ninety-odd images with the camera set for seven frames per second (that’s all I could get before the buffer filled up).  I first looked at the variations in exposure, as measured by the RawDigger counts in the same central area I’ve been using for the SNR work:

First, note how tight the exposures are. With a couple of exceptions, they’re all within plus or minus 2 hundredths of a stop of the mean exposure! The standard deviation is 0.017 f-stops. The trend line (a least-squares fit to the data) indicates that there is essentially no systematic change over time. It looks like there’s a periodic variation that could be vibration-related.

Looking at the SNR, corrected for the Poisson noise changes induced by the tiny exposure variations, we see this:

The standard deviation is 0.093 dB. There is a very slight decrease in SNR over the 12 or so seconds of the sequence.

I thought that there might be more heating with longer exposures, so I made a similar series with the shutter speed set to one second. Here’s the exposure variation:

From the beginning to the end, the least-squares trend line indicates the count-based exposure gets shorter by 0.01 f-stops. The standard deviation of the exposures is also 0.01.

The exposure-corrected SNR gets worse by about 0.01 dB (essentially nothing) over the sequence, and there’s 0.06 dB (standard deviation) worth of noise in the SNR:

It looks like there might be a heating effect, but it’s very small. This sequence of measurements indicates to me is that that the D4 and the 70-200 mm lens that I used are capable of amazing precision in exposure. Think of everything that has to be substantially identical to get the narrow spreads we see in these graphs: the aperture has to stop down to the same place every time, the shutter timing can’t change, and the chip sensitivity can’t vary much (that last is probably the easiest).