[argyllcms] Re: [Possible bug in Spyd2.c] Inconsistant readings from Spyder3

  • From: Gerhard Fuernkranz <nospam456@xxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Sat, 07 Mar 2009 23:35:48 +0100

howdy555@xxxxxxxxx wrote:
>> It may be interesting to obtain the temperature from this sensor along with 
>> each reading, and to graph the temperature as well in order to identify a 
>> potential correlation.
> If only there was a way of obtaining this info. (If it would

Graeme needs to answer this - I'm not sure whether he is aware how to
access the temperature sensor in the gadget he had mentioned.

> be of any help: my room was approximately 23-24 degrees Celsius). Then we 
> could have a corelation and a possibility of
> improving the results! I however think that it would be much easier to just 
> let it measure the same white spot (as the measurements are the fastest) and 
> wait until it stabilises. The only question is: how do we know that it has 
> actually stabilised.

Hmm, is it really an option to wait until it has stabilized, if this
takes about 1..2 hours? I think I would not be so patient :-) [And is it
really worth to spend so much time if the drift in these 2 hours is only
about 3%?]

If one does not care about the accuracy of the absolute luminance, but
more about the accuracy of the readings relative to monitor white, then
I'd rather suggest to take a white reading after every N samples (or
alternatively after every M seconds), compute a "current white" value as
an exponential moving average of these interim white readings, and
correct each reading by scaling it wrt. the "current white" level at the
time when the reading is acquired. Such a procedure would also
compensate a potential drift of the monitor's white luminance while it
is being measured.

> TO HELP??? I tried but I did not get any sensible results due to serious lack 
> of knowledge on this :)

I was not thinking of some horribly complicated stuff, but the idea was
simply to capture a 4-tuple [X,Y,Z,sensor_temperature] for each reading
instead of only [X,Y,Z]. Then re-run you test again for a couple of
hours and plot both, Y versus time, and temperature versus time, and
compare the two curves. If there is a reasonable non-random relationship
between the curves, I'm sure you'll see it intuitively. Prerequisite is
certainly that the monitor is reasonably stable during the test, but
since it was obviously pretty stable during the latter 7 of the 9 hours,
I don't see any reason why it should not have been stable in the first
two hours as well (particularly since it was already turned on for
several hours before you started the test).

[Finding a mathematical function which models the relation reasonably is
a subsequent issue then - but first one would need to see the curves]


Other related posts: