Skip to main content

FirFilter

Namespace: RTPROCESSINGLIB  ·  Library: DSP Library

Python equivalent

mne.filter.filter_data in MNE-Python.

#include <dsp/firfilter.h>

class UTILSLIB::FirFilter

Discoverable façade over the FilterKernel FIR-filter engine.

Mirrors the IirFilter API:

src/examples/ex_dsp_analysis/main.cpp
FilterKernel lowPass = FirFilter::design(256, FirFilter::LowPass, 20.0, 0.0, sFreq, 5.0); // order, type, cutoff(s), Hz, transition
const RowVectorXd smooth = FirFilter::applyZeroPhase(tones, lowPass);

designMne / filterData reproduce mne.filter.create_filter / filter_data:

src/examples/ex_dsp_analysis/main.cpp
const RowVectorXd mneKernel = FirFilter::designMne(sFreq, -1.0, 20.0);   // mne.filter.create_filter defaults: 165 taps
const MatrixXd mneLow = FirFilter::filterData(tones, sFreq, -1.0, 20.0); // mne.filter.filter_data(x, 250, None, 20)

Static Methods​

design(iOrder, type, dCutoffLow, dCutoffHigh, dSFreq, dTransition, method)​

Design a linear-phase FIR filter using FilterKernel as the backend.

Frequencies are given in Hz; the method converts them to the Nyquist-normalised form expected by FilterKernel internally.

Parameters:

  • iOrder : int Filter order (number of taps − 1). Must be even for linear phase.

  • type : FilterType Filter type (LowPass, HighPass, BandPass, BandStop).

  • dCutoffLow : double Lower cutoff frequency in Hz. For LowPass/HighPass: the single cutoff. For BandPass/BandStop: lower edge of the transition band.

  • dCutoffHigh : double Upper cutoff frequency in Hz (ignored for LowPass/HighPass).

  • dSFreq : double Sampling frequency in Hz.

  • dTransition : double Width of the transition band in Hz (Parks-McClellan steepness / Cosine roll-off width). Defaults to 5 Hz.

  • method : DesignMethod Design method (Cosine or ParksMcClellan).

Returns:

apply(vecData, kernel)​

Apply the filter to a single row vector.

Uses overlap-add FFT convolution. The group delay of a symmetric FIR is removed by the FilterKernel engine (it takes the output segment starting at order/2), so the result is already approximately zero-phase for symmetric designs.

Parameters:

  • vecData : const Eigen::RowVectorXd & Input row vector.

  • kernel : FilterKernel & FilterKernel (prepareFilter is called if needed).

Returns:

  • Eigen::RowVectorXd — Filtered row vector (same length as input).

applyZeroPhase(vecData, kernel)​

Apply the filter in a zero-phase forward-backward pass to a single row vector.

Runs the filter twice (forward then reverse) to achieve exactly zero phase shift. Effective order is doubled; transition bandwidth is halved.

Parameters:

Returns:

  • Eigen::RowVectorXd — Zero-phase filtered row vector.

applyZeroPhaseMatrix(matData, kernel, vecPicks)​

Apply the filter in a zero-phase forward-backward pass to every row of a matrix.

Parameters:

  • matData : const Eigen::MatrixXd & Input matrix (n_channels × n_samples).

  • kernel : FilterKernel & FilterKernel.

  • vecPicks : const Eigen::RowVectorXi & Optional row indices to filter; if empty, all rows are filtered.

Returns:

  • Eigen::MatrixXd — Zero-phase filtered matrix (same dimensions as input).

designMne(dSFreq, dLFreq, dHFreq)​

Designs the zero-phase FIR filter of mne.filter.create_filter with its defaults: firwin design with a Hamming window, "auto" transition bands (min(max(0.25 f, 2 Hz), f) below, min(max(0.25 f, 2 Hz), Nyquist - f) above) and "auto" length 3.3 / narrowest band.

Parameters:

  • dSFreq : double Sampling frequency in Hz.

  • dLFreq : double High-pass edge in Hz, or a negative value for a low-pass.

  • dHFreq : double Low-pass edge in Hz, or a negative value for a high-pass.

Returns:

  • Eigen::RowVectorXd — The odd-length impulse response.

filterData(matData, dSFreq, dLFreq, dHFreq)​

Filters every row like mne.filter.filter_data with its FIR defaults: the designMne kernel applied by overlap-add with zero phase, each row padded by reflecting it about its end samples ("reflect_limited").

Parameters:

  • matData : const Eigen::MatrixXd & Data, one row per channel.

  • dSFreq : double Sampling frequency in Hz.

  • dLFreq : double High-pass edge in Hz, or a negative value for a low-pass.

  • dHFreq : double Low-pass edge in Hz, or a negative value for a high-pass.

Returns:

  • Eigen::MatrixXd — The filtered data.

firwin2(nTaps, freq, gain, window)​

Frequency-sampling FIR design, as scipy.signal.firwin2(nTaps, freq, gain, window=...).

Parameters:

  • nTaps : int Number of taps; an even count needs zero gain at Nyquist.

  • freq : const std::vector< double > & Increasing frequencies from 0 to 1 (Nyquist).

  • gain : const std::vector< double > & Gain at each frequency.

  • window : FirWindow The window.

Returns:

  • Eigen::RowVectorXd — The impulse response.

filterDataZeroDouble(matData, dSFreq, dLFreq, dHFreq, window)​

Band-pass every row like mne.filter.filter_data(x, sfreq, l_freq, h_freq, None, "10s", 0.5, 0.5, phase="zero-double", fir_window=..., fir_design="firwin2"): Hann for MNE-Python's ECG and EOG event detection, Hamming for the data filter of compute_proj_ecg / compute_proj_eog.

Parameters:

  • matData : const Eigen::MatrixXd & Data, one row per channel.

  • dSFreq : double Sampling frequency in Hz.

  • dLFreq : double High-pass edge in Hz.

  • dHFreq : double Low-pass edge in Hz.

  • window : FirWindow The FIR window.

Returns:

  • Eigen::MatrixXd — The filtered data.

Example​

Used in examples​

ex_dsp_analysis​

DSP library: Welch PSD, FIR filtering, peak finding, envelope correlation, SPHARA and simulation.

src/examples/ex_dsp_analysis/main.cpp, line 115
FilterKernel lowPass = FirFilter::design(256, FirFilter::LowPass, 20.0, 0.0, sFreq, 5.0); // order, type, cutoff(s), Hz, transition
const RowVectorXd smooth = FirFilter::applyZeroPhase(tones, lowPass);
const double kept = toneAmplitude(smooth.segment(250, 500), 10.0, sFreq);
const double removed = toneAmplitude(smooth.segment(250, 500), 40.0, sFreq);
ok &= expect(smooth.size() == tones.size() && std::fabs(kept - 1.0) < 0.02 && removed < 0.01,
             QString("FirFilter 20 Hz low-pass keeps 10 Hz (%1) and removes 40 Hz (%2)").arg(kept, 0, 'f', 4).arg(removed, 0, 'f', 4));

Authors of this file​