questions about timestamps in recordings

[Shawn-Intretech]

How does Pylsl get the timestamp information? Is it from device or from computer?

[cboulay]

The data are timestamped in the process that pushes the sample to the LSL Outlet. If that process provides a timestamp to the push_sample_ or push_chunk_ function, then it is very important that the provided timestamp is using LSL's local_clock (which happens to be std::chrono::steady_clock). The the process does NOT provide a timestamp, then the local_clock value at the time of the push is used. If using push_chunk_, the timestamp is assigned to the final sample in the chunk.

When creating an inlet to receive the data from an outlet, you have the option of enabling post-processing:

https://github.com/labstreaminglayer/pylsl/blob/main/src/pylsl/inlet.py#L55-L58

One possible transformation will convert the timestamps from the outlet's clock to the inlet's clock. This relies on a behind-the-scenes thread maintaining an understanding of the offsets between these 2 clocks. This thread uses the respective local_clocks for both outlet and inlet (meaning that it is impossible to do this correction if the outlet is using timestamps that do not come from local_clock).

It's up to you to decide if you need to use post-processing or not. For example, LabRecorder does not use any post processing because it separately records clock-offsets and then upon loading the complete data file it will do the various post-processing steps. The idea here is that the post-processing is more accurate when you have the benefit of the entire data file, vs trying to do post-processing live where you only have the (brief) history.

Conversely, I have some streaming applications that need live clock correction so I enable post-processing in those applications by default.