LSL timestamps jump in EEG hyperscanning set-up

[clementinne]

Hello,

I am conducting an EEG hyperscanning experiment, my set-up is as follows :

• 2 LiveAmps, recording 32 channels each, at 500Hz
• Robotic set-up including a master computer using C++ LSL library to send markers of events to recording computer by ethernet
• Recording computer with LSL LabRecorder which records (2 EEG streams + 1 event marker stream) into .xdf file

My offline data processing involves pyxdf to load the .xdf files, in particular .load_xdf() method with no specific options.

I compute the drift between the two EEG streams by looking at the difference EEG1['time_stamps']-EEG2['time_stamps'] across sample indices. I believe each value of EEG1['time_stamps'] or EEG2['time_stamps'] corresponds to timestamps of the LSL clock.

My problem is that for certain recordings, one of the EEG streams' timestamps have a big jump (10 to 20 seconds) at random moments. (See picture) This issue does not seam to affect the data, meaning EEG signals seem to remain continuously recorded. They do not seem to affect the event marker stream either. However, a chunk of data corresponding to the duration of the jump is missing at the end of the affected EEG stream.

I am not sure why this occurs, maybe this is due to LSL-clocks resetting during the recording. I have observed this type of timestamp jump in 5 out of 96 recordings for now, but I am planning to collect more data and would like to know if there is an explanation/solution to this issue. I perform synchrony analyses between pairs of EEG so I need a minimal delay between them.

Thank you for your help!

[cboulay]

I've seen this kind of behaviour before but never that large, at least not without one of the streams disappearing then coming back.

When loading the XDF, try keeping the synchronize_clocks option enabled, but disable all the clock-correction options. Then take a look at your timestamps in the 'problem' streams to see if you still have this issue. If yes, then I think the problem was that your source disappeared from the network for a bit then came back, and the buffering was not sufficient to compensate for the disconnection.

If the issue is gone, then great! In your case, I don't think further investigation is needed because you probably want to disable these other clock-correction options anyway. You probably need to resample (e.g. via interpolation) one of the EEG streams at the timestamps of the other EEG stream so you can do better alignment.

Resampling when aligning 2 streams is advised because 2 devices -- even identical models -- will always have slightly different clocks and therefore will have a different number of samples for the same physical (stopwatch) duration.

[clementinne]

Thank you very much for your answer ! Changing the options of the .load_xdf() method did not work. After further checks, it actually turns out timestamps are true and the data corresponding to the timestamp jump was lost. At this point I believe the issue is due to the LiveAmps more than LSL, but do you have any recommendation concerning the common use of LSL+LiveAmps to try to mitigate this recurring problem?

Thank you again