Reset smoothing parameters after disabling postprocessing

sccn / liblsl

C++ lsl library for multi-modal time-synched data transmission over the local network

Other

108 stars 63 forks source link

Reset smoothing parameters after disabling postprocessing #119

Closed tstenner closed 3 years ago

tstenner commented 3 years ago

PR for #117.

Commit 1: add unit tests for postprocessing options
Commit 2: Move out smoothing parameters and smoothing code into a separate class that can be re-initialized and tested easier
Commit 3: reset postprocessing options when set_options() is called
Commit 4: adjust samples_seen when flushing samples

There's one major changes: the number of encountered samples is only tracked while proc_dejitter is active. Previously, samples were counted and enabling dejittering afterwards re-used this sample counter and therefore assumed a larger number of timestamps had been used to update the dejittering parameters.

tstenner commented 3 years ago

I think separating samples_seen and samples_since_t0 will make the estimation more robust with dropped samples, but that'll take some further tinkering.

cboulay commented 3 years ago

Edit - replacing with better plots and updated gist...

cboulay commented 3 years ago

Updated: https://gist.github.com/cboulay/006bab4dc971cbd9c7b49e7d32c9cd30

This image was generated when using a conda environment containing a lsl.dll built on the previous version:

And this one when using the new lsl.dll Don't worry about the constant offset of 0.6 msec - it's just because I'm using t[0] as my offset, even though t[0] itself is subject to postprocessing with no history so is a poor estimate.

So, in terms of fixing my problem, it looks good.

cboulay commented 3 years ago

One more time but only looking at the pre-flush stage to keep in scale for proper comparison of postprocessing functionality before and after the change.

Before change:

After change:

In addition to the above, I read through the code and unit tests, and I'm satisfied.