Fuji X-T1 Sensor Performance - problem measuring dark noise

[The Admiral]

Recently I've been trying to establish the performance of my Fuji X-T1 sensor, following the approach given by Craig Stark in his 2009 article "Signal to Noise, Part 3 - measuring your camera". The measurement of gain and read noise seemed to work out OK (though whether they are correct is another matter ). However, measuring the dark current seems to give some strange results. As I understand it, the standing dark current gives rise to a signal that is proportional to exposure time. Measuring the average noise level in a dark frame will have two components, that from the dark current, dependent on exposure time, and a fixed read noise for each exposure. Plotting the average signal level against time should therefore give rise to a straight line increasing with time.

What did I get? Something completely different:

The noise increases, though far from linearly (and that's not due to the choice of a log-log scale!), up to 30s, which is the maximum the camera will expose for in Timed (T) mode, but for longer exposures in Bulb (B) mode, the noise levels appear to fall with exposure! Also, the exposures used for the creation of bias frames, at 1/4000th second, and shown on the extreme left of the chart, are higher than for 1/15th second! What on earth is going on? Is Fuji pulling some magic out of the hat? Or have I messed up?

Ian

[The Admiral]

I'm much indebted to Steve Nickolls for sending me this fascinating link to a Craig Stark article relating to odd issues he found with the dark current in Canon DSLRs. In short, his measurements showed that although the stability of the dark current did not behave in the expected manner (i.e. increasing with time, or in other words, as the sensor warmed), the dark noise did. The reality is that camera manufacturers optimise their sensor performance for conventional photography, and this might not be ideal for astrophotography.

This chart shows the average ADU measured on a set of fifty 30s dark frames from the X-T1 taken at the beginning of an imaging session, immediately after taking the flats measurement.

Now, one would have expected that as time progressed, and the sensor warmed, the dark current would have increased  until the sensor's temperature reached equilibrium. What we see is that the average noise levels decreased! Using instead the variance of a selection of dark frames, as per the Craig Stark article, one would nevertheless have expected to see the variance increase with frame number. However, this is not what is observed.

 

All very strange.

Ian