Thanks a lot for your answers.
DrAzzy:
It looks like your chip is doing exactly what the graphs say it should.
Not quite, there is still a discrepancy of ~5 kHz but as you said, that might be just due to the fact that the WDT oscillator is not mean for any high precision.
In case it is useful to others, the approximate formula for the Watchdog oscillator frequency I found by fitting a 2nd degree polynomial to the data of the figure (that I manually digitize) is:
F(V,T) = p00 + p10V + p01T + p20V^2 + p11VT + p02T^2
With:
p00 = 122.6
p10 = -2.742
p01 = -0.0347
p20 = 0.2633
p11 = -0.002518
p02 = -0.0002513
A figure showing a view of the fitted surface is attached.
I plan to implement this in my application and allow it to learn and adjust the constant p00 to improve the prediction of the actual sleep times. It will be an outdoors application, powered straight from a Li-Ion battery, so it will be able to test this over a range of ambien temperatures and voltages.
If I would like to adjust it in a way that the surface match my single lab experiment I would have to set p00=127.3
It will be interesting to know how accurate the sleep times could get by just adjusting p00.
