I meant: Is there a mathematical link between the timestamps. Because then you could use something similar to a greatest common divisor to interpolate to the time points in between.
Do you have any prior knowledge over how your system should react?
Fitting the whole set of sample points (which I assume you've done) to a model is never a robust technique. Using a higher order polynomial will then always make your fitting (seem) better.
But you can make the fitting approach a bit more robust by creating two sets of data.
If you split your original set of sample points up into a "fitting" set and a validation set, then you can create your xth order polynomial based on the "fitting" set.
When your polynomial is known, then you can test your "fit" with the validation set. By comparing the sum of the absolute differences between the expected answer (by the fit) and your values in the validation set you get an idea of the error made by the xth order fit.
You can repeat this then for every order x (of the polynomial) that you want to check.
While you're gradually increasing the order x you'll see that at first the error will decrease, but at some point it will start to rise. This way you can determine a sufficient order (the ones with the smallest errors) to use.
The weak points in this method are the choice of the size of the set ("fitting" & validation), the range of orders you check and which points you choose for each set (never chose the first or the last point for your validation set, because most fitting/modelling techniques cant guarantee anything outside the fitting range).
PS: Personally I prefer using modelling techniques over fitting techniques because in the latter the presence of noise in the measurements is ignored and not modelled.But for some cases fitting techniques can be adequate.
Kind Regards,
Thierry C - CLA, CTA - Senior R&D Engineer (Former Support Engineer) - National Instruments
If someone helped you, let them know. Mark as solved and/or give a kudo. 😉
