Query over medial wall activity

✉️ Mailing List Archive (read-only)
21

hi Donald,

if you want to experiment with this you can run the PSD estimation on
raw data before
and after filtering as in :

http://martinos.org/mne/auto_examples/time_frequency/plot_compute_raw_data_spectrum.html#example-time-frequency-plot-compute-raw-data-spectrum-py

you'll see how many dB you loose at 42.5Hz.

Alex

22

Don (et. al),

As suggested by Alex, here's some empirical data. Let's make some pure
tones in the passband (37.5 Hz) and stopband (42.5) in mne-python using the
testing dataset raw file as an example:

import mne
import numpy as np
raw = mne.fiff.Raw('test.fif', preload=True)
raw._data[0] = 1e-10 * np.sin(2 * np.pi * 37.5 *

np.arange(raw._data.shape[1]) / raw.info['sfreq'])

raw._data[1] = 1e-10 * np.sin(2 * np.pi * 42.5 *

np.arange(raw._data.shape[1]) / raw.info['sfreq'])

raw.save('test_sin_raw.fif')

at a shell, let's do the lowpass you're interested in (lowpasses by
default):

$ mne_process_raw --raw test_sin_raw.fif --save test_sin_lp_raw.fif
--projoff

Back in Python, let's see what happened to the RMS of our signals:

raw = mne.fiff.Raw('test_sin_raw.fif', preload=True)
raw_lp = mne.fiff.Raw('test_sin_lp_raw.fif', preload=True)
np.sqrt(np.mean(raw_lp._data[0] ** 2)) # expect close to 7.07e-11

7.0629870812673256e-11

np.sqrt(np.mean(raw_lp._data[1] ** 2)) # expect closer to zero

1.6152302769536156e-12

20 * np.log10(np.sqrt(np.mean(raw._data[0] ** 2)) /

np.sqrt(np.mean(raw_lp._data[0] ** 2)))
0.0095603499577439008

20 * np.log10(np.sqrt(np.mean(raw._data[1] ** 2)) /

np.sqrt(np.mean(raw_lp._data[1] ** 2)))
32.825155846953265

So, that would be ~0dB attenuation in the passbard (37.5 Hz) and 32 dB of
attenuation at the stop band (42.5 Hz). I also did 45 and 47.5 Hz offline,
and got 35 and 37 dB of attenuation, respectively.

Cheers,
Eric