Eomys/SciDataTool

[CO] Integration of periodic signals + last point + primitive

EmileDvs opened this issue · 1 comments

EmileDvs commented

Hello,

We have implemented integration over time/space using trapz, there are few improvements/corrections that could be further done:

  1. "integrate" may be confusing: do we integrate the signal (trapz function) or do we take its primitive (cumtrapz function) ? We could directly use the function name as input argument, e.g. "time=trapz" / "time=cumtrapz".

  2. The integration interval is not given as input, probably leading to confusion in case of periodicity: should we integrate on the periodic interval ? on the full interval ? => we can set "time=integrate" as integration of full signal, "time=integrate[smallestperiod]", "time=integrate[oneperiod]", "time=integrate[0, 2*pi]",

  3. The last point is systematically neglected so the result is incorrect. The last point should be replicated assuming (anti-)peridodicity so the integration on last interval is considered.

  4. The operations "rms" / "rss" should be improved for time/angle axes using integration (trapz) instead of sum so the result is more accurate

Best regards,
Emile

EmileDvs commented

The issue is modified as follow:

  1. "integration" is in fact similar to "sum", e.g. "time=sum" should in fact perform summation using trapz (integration) instead of calling "sum" but the result is the same => the axis along which summation/integration is performed is squeezed. primitive integral is calculated using "time=antiderivate", by opposition to "time=derivate" command.
  2. To improve performances operations are always performed on smallest period if it exists, then the (anti-)periodicity factor is included to get the result on the full period. For DataPattern operations are always performed on the full pattern.
  3. doesn't change
  4. "integrate" is now merged in "sum" for "time", "angle", "z" axes. A new function to perform sum and mean is implemented depending on axis type.