mne.filter.construct_iir_filter#

mne.filter.construct_iir_filter(iir_params, f_pass=None, f_stop=None, sfreq=None, btype=None, return_copy=True, *, phase='zero', verbose=None)[source]#

Use IIR parameters to get filtering coefficients.

This function works like a wrapper for iirdesign and iirfilter in scipy.signal to make filter coefficients for IIR filtering. It also estimates the number of padding samples based on the filter ringing. It creates a new iir_params dict (or updates the one passed to the function) with the filter coefficients (‘b’ and ‘a’) and an estimate of the padding necessary (‘padlen’) so IIR filtering can be performed.

Parameters:
iir_paramsdict

Dictionary of parameters to use for IIR filtering.

  • If iir_params['sos'] exists, it will be used as second-order sections to perform IIR filtering.

    New in v0.13.

  • Otherwise, if iir_params['b'] and iir_params['a'] exist, these will be used as coefficients to perform IIR filtering.

  • Otherwise, if iir_params['order'] and iir_params['ftype'] exist, these will be used with scipy.signal.iirfilter to make a filter. You should also supply iir_params['rs'] and iir_params['rp'] if using elliptic or Chebychev filters.

  • Otherwise, if iir_params['gpass'] and iir_params['gstop'] exist, these will be used with scipy.signal.iirdesign to design a filter.

  • iir_params['padlen'] defines the number of samples to pad (and an estimate will be calculated if it is not given). See Notes for more details.

  • iir_params['output'] defines the system output kind when designing filters, either “sos” or “ba”. For 0.13 the default is ‘ba’ but will change to ‘sos’ in 0.14.

f_passfloat or list of float

Frequency for the pass-band. Low-pass and high-pass filters should be a float, band-pass should be a 2-element list of float.

f_stopfloat or list of float

Stop-band frequency (same size as f_pass). Not used if ‘order’ is specified in iir_params.

sfreqfloat | None

The sample rate.

btypestr

Type of filter. Should be ‘lowpass’, ‘highpass’, or ‘bandpass’ (or analogous string representations known to scipy.signal.iirfilter()).

return_copybool

If False, the ‘sos’, ‘b’, ‘a’, and ‘padlen’ entries in iir_params will be set inplace (if they weren’t already). Otherwise, a new iir_params instance will be created and returned with these entries.

phasestr

Phase of the filter. phase='zero' (default) or equivalently 'zero-double' constructs and applies IIR filter twice, once forward, and once backward (making it non-causal) using filtfilt(); phase='forward' will apply the filter once in the forward (causal) direction using lfilter().

New in v0.13.

verbosebool | str | int | None

Control verbosity of the logging output. If None, use the default verbosity level. See the logging documentation and mne.verbose() for details. Should only be passed as a keyword argument.

Returns:
iir_paramsdict

Updated iir_params dict, with the entries (set only if they didn’t exist before) for ‘sos’ (or ‘b’, ‘a’), and ‘padlen’ for IIR filtering.

See also

mne.filter.filter_data
mne.io.Raw.filter

Notes

This function triages calls to scipy.signal.iirfilter() and scipy.signal.iirdesign() based on the input arguments (see linked functions for more details).

Changed in version 0.14: Second-order sections are used in filter design by default (replacing output='ba' by output='sos') to help ensure filter stability and reduce numerical error.

Examples

iir_params can have several forms. Consider constructing a low-pass filter at 40 Hz with 1000 Hz sampling rate.

In the most basic (2-parameter) form of iir_params, the order of the filter ‘N’ and the type of filtering ‘ftype’ are specified. To get coefficients for a 4th-order Butterworth filter, this would be:

>>> iir_params = dict(order=4, ftype='butter', output='sos')  
>>> iir_params = construct_iir_filter(iir_params, 40, None, 1000, 'low', return_copy=False)  
>>> print((2 * len(iir_params['sos']), iir_params['padlen']))  
(4, 82)

Filters can also be constructed using filter design methods. To get a 40 Hz Chebyshev type 1 lowpass with specific gain characteristics in the pass and stop bands (assuming the desired stop band is at 45 Hz), this would be a filter with much longer ringing:

>>> iir_params = dict(ftype='cheby1', gpass=3, gstop=20, output='sos')  
>>> iir_params = construct_iir_filter(iir_params, 40, 50, 1000, 'low')  
>>> print((2 * len(iir_params['sos']), iir_params['padlen']))  
(6, 439)

Padding and/or filter coefficients can also be manually specified. For a 10-sample moving window with no padding during filtering, for example, one can just do:

>>> iir_params = dict(b=np.ones((10)), a=[1, 0], padlen=0)  
>>> iir_params = construct_iir_filter(iir_params, return_copy=False)  
>>> print((iir_params['b'], iir_params['a'], iir_params['padlen']))  
(array([1., 1., 1., 1., 1., 1., 1., 1., 1., 1.]), [1, 0], 0)

For more information, see the tutorials Background information on filtering and Filtering and resampling data.

Examples using mne.filter.construct_iir_filter#

Background information on filtering

Background information on filtering