Check if pfind works with absolute time difference larger than half second

[franciumxzf]

No description provided.

[franciumxzf]

Test datasets were generated on dataset27 (bob) using generate_freqdetuneg2.py. We added time delay from 10ms to 1s and pfind works for these datasets.
For additional time delay shorter than 100ms, the hyperparameters working for original dataset27 works well.
For additional time delay longer than 100ms, we need larger acquisition time interval (epochs) and bin number (buffer order) to get it work. We can obtain results that generally make sense. However, in the time convergence test, some results vary by ~8ns which still need to improve.

[pyuxiang]

I suppose an 8ns difference is still reasonable - costream should be able to handle this downstream of pfind.

Wasn't expecting required time interval to increase though, since we're already working with acquisition times of around 3s (so the overlap is closer to 2s).

[franciumxzf]

For original dataset27, we use the acquisation time interval Ta = 2^29 ns (~500ms, so clearly that when the time delay is longer, we need to increase this). When adding time delay longer than 100ms, we increase the acquisation time interval to Ta = 2^31 ns (~2s).

[pyuxiang]

Oh I see, nice! The idea of getting pfind to work beyond an epoch size is to extend operation time as the timestamps become increasingly separated, as in #2, before we attempt another frequency recalibration.

Wonder what mechanism we can use to avoid increasing number of epochs / buffer order?

[franciumxzf]

We need to run pfind again before the time difference is larger than 0.5 seconds. Costream needs to monitor this time difference.