[argyllcms] Re: ColorMunki measurement drift

  • From: Graeme Gill <graeme@xxxxxxxxxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Mon, 27 Sep 2010 12:38:09 +1000

Juergen Lilien wrote:
In a first try I attached the ColorMunki to the screen and let it take
more than 1200 white point measurements (hires adaptive mode) in a row,
this took roughly an hour. At this time the monitor was several hours
in on state, so it should be pretty stable.

It's not unexpected that such an instrument will drift due to temperature
effects over a time period like an hour, if it at a different temperature
when the measurements first started. It takes a more expensive instrument
(ie. with a heater/cooler on the sensor, or a temperature sensor and
 compensation tables, or some sort of internal calibration tile
 that it can automatically calibrate to) to be able to counteract such
effects. This is a basic characteristic of using silicon for the sensor.

The color (X/Y) measurements are even more interesting, they are
also decreasing but the variance is not constant, there are knots
in the point graphs with less variance and the period between the
knots is increasing over time - the pattern is more stretched.
At the end the measurements seem to have reached a semi constant
state with nearly no further decreasing tendency.

The "knots" could be quantization artefacts or something similar.
I'm not sure that I'd say that the Y isn't stabilizing, as it does have
a convex shape. Without an independent instrument to confirm that
the display is perfectly stable, there could also be display
drift effects, in spite of it having been on for several hours
(Consider: if the display is affected by it's temperature, then
 anything that changes the rate at which heat is dissipated from
 the display and/or instrument may affect the readings. So a change
 in air temperature due to ambient temperature change, room heating/cooling
 cycles, a door being opened or closed etc., could have subtle effects.)

My first thought was to wait an hour with the ColorMunki on the
screen, until the color measurements reach the stable bottom,
before starting calibration and profiling. The problem is, that
the first (highest) readings seem a lot more reasonable compared
to my pre-calibrated monitor presets, and even after a self-
calibration of the ColorMunki they never get back to the start
level. Yes, this effect (temperature drift?) seems to be at
least partly immune against self-calibration of the instrument!

I don't see why you are assuming that the pre-set is the reference.
The point of using an instrument is that it is assumed that it
will be more accurate that the state of the display.

To prove this I started a second series of measurements with a
self-calibration after each single measurement and I still see
decreasing luminance and X/Y color values.

That's interesting, but without some sort of independent reference,
it's hard to know exactly what is changing.

Should I reinstall the X-Rite software and check if I see the
same decreasing measurement values?

I'd certainly be interested in any evidence that there is different
behaviour with the X-Rite drivers.

And what is the practical relevance? The DE of the first
measurement was 2,6 and it dropped to 0,7 at the end,
a 1,9 DE white point difference should be visible(?).

You can't judge the visibility of a difference in degrees Kelvin
or xy coordinate. You need to convert to DE to make this assessment.

As I understand the new -I option will only help to counteract
drifting in the brightness level measurements, but does not
help with the color/white point drift?

It can help with instrument black point drift and display white
point change. It won't make the instrument more accurate in
an absolute sense, but minimizes the effect of discrepancies
between the instrument and display on calibration and profiling.
[Note you can use these corrections independently.]

Graeme Gill.

Other related posts: