An efficient Finite Impulse Response (FIR) filter class written in C++ with python wrapper.
Adaptive filtering is also implemented using the Least Mean Square (LMS) or Normalised Least Mean Square (NLMS) algorithm.
Install the python package with pip:
pip install fir1
You can also install from source:
git clone https://github.com/berndporr/fir1 cd fir1 python setup.py install
The filter is a realtime filter which receives samples one by one so it can process data as it arrives from an ADC converter. This is simulated here with the for loop:
import fir1 b = signal.firwin(999,0.1) f = fir1.Fir1(b) for i in range(len(noisy_signal)): clean_signal[i] = f.filter(noisy_signal[i])
The constructor Fir1()
receives the filter coefficients
(= impulse response) and then filtering is performed
with the method filter()
.
The file lms_50Hz_ecg_filter.py removes 50Hz from an ECG with the help of the lms filter. The filter learns its own frequency response from a reference 50Hz sine wave:
f = fir1.Fir1(NTAPS) f.setLearningRate(LEARNING_RATE); y= np.empty(len(ecg)) for i in range(len(ecg)): ref_noise = np.sin(2.0 * np.pi / 20.0 * i); canceller = f.filter(ref_noise) output_signal = ecg[i] - canceller f.lms_update(output_signal) y[i] = output_signal
You can query the filter coefficients using getCoeff. This is most useful to obtain the kernel of a trained adaptive filter:
>>> from fir1 import Fir1 >>> fir = Fir1([.25, -.5, 1, -.5, .25]) >>> fir.getCoeff() array([ 0.25, -0.5 , 1. , -0.5 , 0.25])
You may override the length of array to return and the result will be zero-padded. Specifying too small an array causes an exception to be raised:
>>> fir.getCoeff(8) array([ 0.25, -0.5 , 1. , -0.5 , 0.25, 0. , 0. , 0. ]) >>> fir.getCoeff(3) Traceback (most recent call last): File "<stdin>", line 1, in <module> File ".../fir1.py", line 97, in getCoeff return _fir1.Fir1_getCoeff(self, *args) RuntimeError: Fir1: target of getCoeff: too many weights to copy into target
Both the demo file and an explanation how the LMS filter works can be found on the homepage of the project: