Filtering multidimensional data with scipy.sosfilt-CodePudding

I am trying to filter a 2D array of dimension 19 channels x 10,000 samples using 'scipy.sosfilt' but am receiving an error regarding my initial filter shape. I read the solutions posted here and here but am looking for something more general than just the 2-channel solution as well as something that doesn't involve looping over each channel.

Here is what I have tried:

import numpy as np
import scipy
from scipy import signal

# data
# 19 channels x 10,000 samples
data = np.random.rand(19, 10000)

# sampling rate (Hz)
sampling_rate = 500 

# filters
nyq = sampling_rate / 2

# create bandpass filter
band_low = 0.01
band_lowcut = band_low / nyq
band_high = 50.0
band_highcut = band_high / nyq
band_sos = scipy.signal.butter(N = 3, Wn = [band_lowcut, band_highcut], btype = 'bandpass', fs = sampling_rate, output = 'sos')
band_z = scipy.signal.sosfilt_zi(band_sos)

# apply bandpass filter across the columns i.e. each of 19 channels bandpassed individually
bandpassed_data, band_z = scipy.signal.sosfilt(sos = band_sos, x = data, zi = band_z, axis = 1)

I receive the following error:

ValueError: Invalid zi shape. With axis=1, an input with shape (19, 10000), and an sos array with 3 sections, zi must have shape (3, 19, 2), got (3, 2).

How should I go about either reshaping my initial filter or my data to avoid this error?
The naive approach would be to loop through each of 19 channels, and filter each channel separately. But I am hoping I can instead filter the entire 19x10000 array without any loops.

Thanks!

CodePudding user response：

The reshaping to (3, 19, 2) can be done like this, assuming you want the same for all 19 channels:

band_z = np.repeat(np.expand_dims(band_z, axis=1), 19, axis=1)

I tried it in the latest scipy version and this worked.

Cheers