search

fusion-flap / flap

Fusion Library of Analysis Programs
MIT License
11 stars 5 forks source link

MDSPlus data can have non equidistant time vector #52

Closed thelampire closed 3 years ago

thelampire commented 4 years ago

There is an issue with MDSplus that the underlying code is providing an incorrect, non-equidistant time vector when the data is fast. I am trying to read a MHz Mirnov data and the time vector is incorrect due to numerical errors. This is not an issue of FLAP (i've checked), but rather the MDSPlus module or MDSPlus itself.

I would like to raise awareness, that this needs to be tackled with. Has anyone ever read signals from the MDSPlus which was non-equidistantly placed in time? Can I generate the time vector from the first two elements of the MDSPlus time vector so it would be accurate?

sandorzoletnik commented 4 years ago

This might be an issue in other databases as well. This is why FLAP has the equidistant coordinate.

thelampire commented 4 years ago

It turned out that numpy has the same problem. This might be insightful for the future. This code generates an incorrect time vector: time=np.linspace(0,1-2e-7,5e6, dtype='float32') This should be a 1s long, equidistantly spaced time vector sampled with 5MHz. The resulting sampling time looks like this: MDSplus_time_vector_error_propagation.pdf

This returns a correct one: time=np.linspace(0,1-2e-7,5e6, dtype='float64')

I don't have time now to find how it's done in FLAP, but this problem could appear.

thelampire commented 3 years ago

I believe this was an MDSplus problem at the end and have not come up again.