Hi, in the doc of psd_multitaper it says
Does that mean by default, the bandwidth is 8 for it says the half-bandwidth is 4?
Hi, in the doc of psd_multitaper it says
Hi, @BarryLiu97,
I agree that the documentation could be clearer with respect to the bandwith multitaper parameters. In this case the bandwidth argument and the default half-bandwidth are different things. Looking at the code it seems that the “window half-bandwidth” that the docstring refers to is the “time half-bandwidth product”, defined as:
TW = N * f / 2
where N
is window size (in seconds) and f
is frequency resolution in Hz (“bandwidth”).
So, the default bandwidth (and thus default frequency smoothing) is variable because it depends on time half-bandwidth product, which is a product of frequency bandwidth and signal length. For a 1-s window length a time half-bandwidth product of 4 means that the bandwidth is 8 (± 4 Hz frequency smoothing), but for a 0.5 s window that bandwidth would be 16 (± 8 Hz frequency smoothing).
Why is the default time half-bandwidth product 4? This seems to be a common default (among few others) in many multitaper applications and it results in 7 tapers being used (2 * TW - 1
).
I’ll try to submit a PR next week to clarify mne docs with respect to bandwidth parameters.
Yeah, it is pretty confusing sometimes. In the calculation of PSD and CSD, the parameter is bandwidth, but in tfr mne.time_frequency.tfr_multitaper — MNE 0.24.1 documentation, the Time x (Full) Bandwidth product is used.