Skip to content
This repository has been archived by the owner on May 31, 2024. It is now read-only.

Reasonably treat non-equidistant signal's sampling interval length and frequency #265

Open
BjoernLudwigPTB opened this issue Dec 17, 2021 · 0 comments

Comments

@BjoernLudwigPTB
Copy link
Member

Is your feature request related to a problem? Please describe.
At the moment the signal class if not provided by the user, computes the sampling interval length as kind of a weighted sum of the different occuring interval legths in the time vector. The stored interval length is taken as the arithmetic mean of all unique interval lengths in the time vector. The sampling frequency if not provided then is taken as the reciprocal of that. This can result in kind of inconsistent data in an instances attributes, e.g. in case there is one single interval being much larger than all other possibly very short intervals. If you then only look at the frequency and the values, you might draw wrong conclusions. We do not see clearly though, what are the use cases of the two attributes.

Describe the solution you'd like
We wish for a profound reasoning behind the way of computing the interval length and sampling frequency, which is then well documented in the corresponding part of the docs.

Describe alternatives you've considered
One way would be to find the most reasonable mean (unweighted arithmetic/geometric, harmonic, median) of interval length and compute that for the provided time vector. One could as well compute the frequency and interval length only when needed (when is that?) and maybe even interpolate the values at the resulting time instances for these cases?!

Additional context
We are grateful for any input on that from any expert user just as a comment in this issue to start with.

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Projects
None yet
Development

No branches or pull requests

1 participant