ENH: add alternative function for post-hoc time-shift synchronization

Hi everyone,

Non-zero time-shifts in event-based time-series is a problem I 
encountered in multiple projects the last years.
I work on the "nearest-advocate" - function that performs a post-hoc 
time-shift synchronization of event-based time-series data. I.e., the 
data is given by two arrays that consists of the respective event's time 
stamps.
Currently, the cross-corrlation is also used in here, with the 
workaround of creating "inter-event intervals" and interpolation to 
transform the data into constantly sampled signals. However, this trick 
looses some information of the event timestamps.
It can be shown that nearest-advocate is more precise than 
cross-correlation, it also works with short arrays and the runtime is 
similar for longer measurements (as the search space is usually constant 
for this type of application). Therefore, we were even able to correct 
non-linear clock-drift in a rolling window approach.
The usage of kernel convolutions before the cross-correlation helps, but 
it is still less robust than the nearest-advocate approach.

The basic part of the algorithm and initial results are publicized here 
(in german): https://www.researchgate.net/publication/364165101
And a more comprehensive paper is in preparation and will be submitted 
in February.

The code is already put into a SciPy-like form on this fork (see the 
imported cython-module): 
https://github.com/iot-salzburg/scipy/blob/nearest_advocate_sync/scipy/signal/_nearest_advocate.py

What do you think about an integration into SciPy?

Thank you for you replies!
Christoph
_______________________________________________
SciPy-Dev mailing list -- [email protected]
To unsubscribe send an email to [email protected]
https://mail.python.org/mailman3/lists/scipy-dev.python.org/
Member address: [email protected]