Padding Short Epochs with mne.filter.resample Not Effective

💬 Support & Discussions
1

I have time series data that I processed in Matlab/Brainstorm and source reconstructed and projected into ROIs. I am now importing this data into a simple structure in MNE v1.5 on MacOS. Since the trials are very short, I want to pad each single trial before computing PSDs and averaging them to get 1 PSD estimate with better resolution per channel.

Below a code example:

import numpy as np
import mne

sfreq = 500 # Sampling frequency
tmax = 0.5
times = np.arange(0, 0.5, 0.002) # 
 
sin = np.sin(times * 10) # Multiplied by 10 for shorter cycles
cos = np.cos(times * 10)
sinX2 = sin * 2 + 0.01 * np.random.normal(size=times.size)
cosX2 = cos * 2 + 0.01 * np.random.normal(size=times.size)

data = np.array([sin, cos, sinX2, cosX2])

# Definition of channel types and names.
ch_types = ["misc", "misc", "misc", "misc"]
ch_names = ["sin", "cos", "sinX2", "cosX2"]

# create raw object

info = mne.create_info(ch_names=ch_names, sfreq=sfreq, ch_types=ch_types)
raw = mne.io.RawArray(data, info)
 
# create epochs
raw_trials = [raw, raw]
datas = [r.get_data() for r in raw_trials]
info = raw_trials[0].info
epochs = mne.EpochsArray(datas, info)
spectrum_before = epochs.compute_psd(picks="misc", method="welch", n_overlap=125)
spectrum_before.plot(picks="misc");

So far, this works as expected. However, when I try to create padded epochs using mne.filter.resample, it doesn’t seem to have an effect:

before_padding = epochs.get_data()
display(before_padding.shape)

after = mne.filter.resample(before_padding, npad=1 * sfreq, pad="wrap", verbose=True)
display(after.shape)

# compute psd
after_epochs = mne.EpochsArray(after, info)
spectrum = after_epochs.compute_psd(picks="misc", method="welch", n_overlap=125)

The shape of the epochs doesn’t change after padding with mne.filter.resample.

Is mne.filter.resample not suited for achieving my goal of padding short epochs before averaging? Or am I using it incorrectly? How can I properly pad short epochs before averaging to get better resolution of PSD estimates?

2

No. The padding that happens in mne.filter.resample is temporary and goes away after the resampling operation.

Use np.pad(np.array(datas), pad_width=(before, after), mode="wrap") before you create your EpochsArray. If I’m interpreting your intent correctly, you want before = after = sfreq/2

3 Likes