Skip to content


Subversion checkout URL

You can clone with HTTPS or Subversion.

Download ZIP
Fetching contributors…

Cannot retrieve contributors at this time

544 lines (425 sloc) 18.572 kb
"""Functions for FIR filter design."""
from math import ceil, log
import numpy as np
from numpy.fft import irfft
from scipy.special import sinc
import sigtools
# Some notes on function parameters:
# `cutoff` and `width` are given as a numbers between 0 and 1. These
# are relative frequencies, expressed as a fraction of the Nyquist rate.
# For example, if the Nyquist rate is 2KHz, then width=0.15 is a width
# of 300 Hz.
# The `order` of a FIR filter is one less than the number of taps.
# This is a potential source of confusion, so in the following code,
# we will always use the number of taps as the parameterization of
# the 'size' of the filter. The "number of taps" means the number
# of coefficients, which is the same as the length of the impulse
# response of the filter.
def kaiser_beta(a):
"""Compute the Kaiser parameter `beta`, given the attenuation `a`.
a : float
The desired attenuation in the stopband and maximum ripple in
the passband, in dB. This should be a *positive* number.
beta : float
The `beta` parameter to be used in the formula for a Kaiser window.
Oppenheim, Schafer, "Discrete-Time Signal Processing", p.475-476.
if a > 50:
beta = 0.1102 * (a - 8.7)
elif a > 21:
beta = 0.5842 * (a - 21)**0.4 + 0.07886 * (a - 21)
beta = 0.0
return beta
def kaiser_atten(numtaps, width):
"""Compute the attenuation of a Kaiser FIR filter.
Given the number of taps `N` and the transition width `width`, compute the
attenuation `a` in dB, given by Kaiser's formula:
a = 2.285 * (N - 1) * pi * width + 7.95
N : int
The number of taps in the FIR filter.
width : float
The desired width of the transition region between passband and stopband
(or, in general, at any discontinuity) for the filter.
a : float
The attenuation of the ripple, in dB.
See Also
kaiserord, kaiser_beta
a = 2.285 * (numtaps - 1) * np.pi * width + 7.95
return a
def kaiserord(ripple, width):
"""Design a Kaiser window to limit ripple and width of transition region.
ripple : float
Positive number specifying maximum ripple in passband (dB) and minimum
ripple in stopband.
width : float
Width of transition region (normalized so that 1 corresponds to pi
radians / sample).
numtaps : int
The length of the kaiser window.
beta :
The beta parameter for the kaiser window.
There are several ways to obtain the Kaiser window:
signal.kaiser(numtaps, beta, sym=0)
signal.get_window(beta, numtaps)
signal.get_window(('kaiser', beta), numtaps)
The empirical equations discovered by Kaiser are used.
See Also
kaiser_beta, kaiser_atten
Oppenheim, Schafer, "Discrete-Time Signal Processing", p.475-476.
A = abs(ripple) # in case somebody is confused as to what's meant
if A < 8:
# Formula for N is not valid in this range.
raise ValueError("Requested maximum ripple attentuation %f is too "
"small for the Kaiser formula." % A)
beta = kaiser_beta(A)
# Kaiser's formula (as given in Oppenheim and Schafer) is for the filter
# order, so we have to add 1 to get the number of taps.
numtaps = (A - 7.95) / 2.285 / (np.pi * width) + 1
return int(ceil(numtaps)), beta
def firwin(numtaps, cutoff, width=None, window='hamming', pass_zero=True,
scale=True, nyq=1.0):
FIR filter design using the window method.
This function computes the coefficients of a finite impulse response filter.
The filter will have linear phase; it will be Type I if `numtaps` is odd and
Type II if `numtaps` is even.
Type II filters always have zero response at the Nyquist rate, so a
ValueError exception is raised if firwin is called with `numtaps` even and
having a passband whose right end is at the Nyquist rate.
numtaps : int
Length of the filter (number of coefficients, i.e. the filter
order + 1). `numtaps` must be even if a passband includes the
Nyquist frequency.
cutoff : float or 1D array_like
Cutoff frequency of filter (expressed in the same units as `nyq`)
OR an array of cutoff frequencies (that is, band edges). In the
latter case, the frequencies in `cutoff` should be positive and
monotonically increasing between 0 and `nyq`. The values 0 and
`nyq` must not be included in `cutoff`.
width : float or None
If `width` is not None, then assume it is the approximate width
of the transition region (expressed in the same units as `nyq`)
for use in Kaiser FIR filter design. In this case, the `window`
argument is ignored.
window : string or tuple of string and parameter values
Desired window to use. See `scipy.signal.get_window` for a list
of windows and required parameters.
pass_zero : bool
If True, the gain at the frequency 0 (i.e. the "DC gain") is 1.
Otherwise the DC gain is 0.
scale : bool
Set to True to scale the coefficients so that the frequency
response is exactly unity at a certain frequency.
That frequency is either:
0 (DC) if the first passband starts at 0 (i.e. pass_zero
is True);
`nyq` (the Nyquist rate) if the first passband ends at
`nyq` (i.e the filter is a single band highpass filter);
center of first passband otherwise.
nyq : float
Nyquist frequency. Each frequency in `cutoff` must be between 0
and `nyq`.
h : 1D ndarray
Coefficients of length `numtaps` FIR filter.
If any value in `cutoff` is less than or equal to 0 or greater
than or equal to `nyq`, if the values in `cutoff` are not strictly
monotonically increasing, or if `numtaps` is even but a passband
includes the Nyquist frequency.
Low-pass from 0 to f::
>>> firwin(numtaps, f)
Use a specific window function::
>>> firwin(numtaps, f, window='nuttall')
High-pass ('stop' from 0 to f)::
>>> firwin(numtaps, f, pass_zero=False)
>>> firwin(numtaps, [f1, f2], pass_zero=False)
>>> firwin(numtaps, [f1, f2])
Multi-band (passbands are [0, f1], [f2, f3] and [f4, 1])::
>>>firwin(numtaps, [f1, f2, f3, f4])
Multi-band (passbands are [f1, f2] and [f3,f4])::
>>> firwin(numtaps, [f1, f2, f3, f4], pass_zero=False)
See also
# The major enhancements to this function added in November 2010 were
# developed by Tom Krauss (see ticket #902).
cutoff = np.atleast_1d(cutoff) / float(nyq)
# Check for invalid input.
if cutoff.ndim > 1:
raise ValueError("The cutoff argument must be at most one-dimensional.")
if cutoff.size == 0:
raise ValueError("At least one cutoff frequency must be given.")
if cutoff.min() <= 0 or cutoff.max() >= 1:
raise ValueError("Invalid cutoff frequency: frequencies must be greater than 0 and less than nyq.")
if np.any(np.diff(cutoff) <= 0):
raise ValueError("Invalid cutoff frequencies: the frequencies must be strictly increasing.")
if width is not None:
# A width was given. Find the beta parameter of the Kaiser window
# and set `window`. This overrides the value of `window` passed in.
atten = kaiser_atten(numtaps, float(width)/nyq)
beta = kaiser_beta(atten)
window = ('kaiser', beta)
pass_nyquist = bool(cutoff.size & 1) ^ pass_zero
if pass_nyquist and numtaps % 2 == 0:
raise ValueError("A filter with an even number of coefficients must "
"have zero response at the Nyquist rate.")
# Insert 0 and/or 1 at the ends of cutoff so that the length of cutoff is even,
# and each pair in cutoff corresponds to passband.
cutoff = np.hstack(([0.0]*pass_zero, cutoff, [1.0]*pass_nyquist))
# `bands` is a 2D array; each row gives the left and right edges of a passband.
bands = cutoff.reshape(-1,2)
# Build up the coefficients.
alpha = 0.5 * (numtaps-1)
m = np.arange(0, numtaps) - alpha
h = 0
for left, right in bands:
h += right * sinc(right * m)
h -= left * sinc(left * m)
# Get and apply the window function.
from signaltools import get_window
win = get_window(window, numtaps, fftbins=False)
h *= win
# Now handle scaling if desired.
if scale:
# Get the first passband.
left, right = bands[0]
if left == 0:
scale_frequency = 0.0
elif right == 1:
scale_frequency = 1.0
scale_frequency = 0.5 * (left + right)
c = np.cos(np.pi * m * scale_frequency)
s = np.sum(h * c)
h /= s
return h
# Original version of firwin2 from scipy ticket #457, submitted by "tash".
# Rewritten by Warren Weckesser, 2010.
def firwin2(numtaps, freq, gain, nfreqs=None, window='hamming', nyq=1.0):
"""FIR filter design using the window method.
From the given frequencies `freq` and corresponding gains `gain`,
this function constructs an FIR filter with linear phase and
(approximately) the given frequency response.
numtaps : int
The number of taps in the FIR filter. `numtaps` must be less than
`nfreqs`. If the gain at the Nyquist rate, `gain[-1]`, is not 0,
then `numtaps` must be odd.
freq : array-like, 1D
The frequency sampling points. Typically 0.0 to 1.0 with 1.0 being
Nyquist. The Nyquist frequency can be redefined with the argument
The values in `freq` must be nondecreasing. A value can be repeated
once to implement a discontinuity. The first value in `freq` must
be 0, and the last value must be `nyq`.
gain : array-like
The filter gains at the frequency sampling points.
nfreqs : int, optional
The size of the interpolation mesh used to construct the filter.
For most efficient behavior, this should be a power of 2 plus 1
(e.g, 129, 257, etc). The default is one more than the smallest
power of 2 that is not less than `numtaps`. `nfreqs` must be greater
than `numtaps`.
window : string or (string, float) or float, or None, optional
Window function to use. Default is "hamming". See
`scipy.signal.get_window` for the complete list of possible values.
If None, no window function is applied.
nyq : float
Nyquist frequency. Each frequency in `freq` must be between 0 and
`nyq` (inclusive).
taps : numpy 1D array of length `numtaps`
The filter coefficients of the FIR filter.
A lowpass FIR filter with a response that is 1 on [0.0, 0.5], and
that decreases linearly on [0.5, 1.0] from 1 to 0:
>>> taps = firwin2(150, [0.0, 0.5, 1.0], [1.0, 1.0, 0.0])
>>> print(taps[72:78])
[-0.02286961 -0.06362756 0.57310236 0.57310236 -0.06362756 -0.02286961]
See also
From the given set of frequencies and gains, the desired response is
constructed in the frequency domain. The inverse FFT is applied to the
desired response to create the associated convolution kernel, and the
first `numtaps` coefficients of this kernel, scaled by `window`, are
The FIR filter will have linear phase. The filter is Type I if `numtaps`
is odd and Type II if `numtaps` is even. Because Type II filters always
have a zero at the Nyquist frequency, `numtaps` must be odd if `gain[-1]`
is not zero.
.. versionadded:: 0.9.0
.. [1] Oppenheim, A. V. and Schafer, R. W., "Discrete-Time Signal
Processing", Prentice-Hall, Englewood Cliffs, New Jersey (1989).
(See, for example, Section 7.4.)
.. [2] Smith, Steven W., "The Scientist and Engineer's Guide to Digital
Signal Processing", Ch. 17.
if len(freq) != len(gain):
raise ValueError('freq and gain must be of same length.')
if nfreqs is not None and numtaps >= nfreqs:
raise ValueError('ntaps must be less than nfreqs, but firwin2 was '
'called with ntaps=%d and nfreqs=%s' % (numtaps, nfreqs))
if freq[0] != 0 or freq[-1] != nyq:
raise ValueError('freq must start with 0 and end with `nyq`.')
d = np.diff(freq)
if (d < 0).any():
raise ValueError('The values in freq must be nondecreasing.')
d2 = d[:-1] + d[1:]
if (d2 == 0).any():
raise ValueError('A value in freq must not occur more than twice.')
if numtaps % 2 == 0 and gain[-1] != 0.0:
raise ValueError("A filter with an even number of coefficients must "
"have zero gain at the Nyquist rate.")
if nfreqs is None:
nfreqs = 1 + 2 ** int(ceil(log(numtaps,2)))
# Tweak any repeated values in freq so that interp works.
eps = np.finfo(float).eps
for k in range(len(freq)):
if k < len(freq)-1 and freq[k] == freq[k+1]:
freq[k] = freq[k] - eps
freq[k+1] = freq[k+1] + eps
# Linearly interpolate the desired response on a uniform mesh `x`.
x = np.linspace(0.0, nyq, nfreqs)
fx = np.interp(x, freq, gain)
# Adjust the phases of the coefficients so that the first `ntaps` of the
# inverse FFT are the desired filter coefficients.
shift = np.exp(-(numtaps-1)/2. * 1.j * np.pi * x / nyq)
fx2 = fx * shift
# Use irfft to compute the inverse FFT.
out_full = irfft(fx2)
if window is not None:
# Create the window to apply to the filter coefficients.
from signaltools import get_window
wind = get_window(window, numtaps, fftbins=False)
wind = 1
# Keep only the first `numtaps` coefficients in `out`, and multiply by
# the window.
out = out_full[:numtaps] * wind
return out
def remez(numtaps, bands, desired, weight=None, Hz=1, type='bandpass',
maxiter=25, grid_density=16):
Calculate the minimax optimal filter using the Remez exchange algorithm.
Calculate the filter-coefficients for the finite impulse response
(FIR) filter whose transfer function minimizes the maximum error
between the desired gain and the realized gain in the specified
frequency bands using the Remez exchange algorithm.
numtaps : int
The desired number of taps in the filter. The number of taps is
the number of terms in the filter, or the filter order plus one.
bands : array_like
A monotonic sequence containing the band edges in Hz.
All elements must be non-negative and less than half the sampling
frequency as given by `Hz`.
desired : array_like
A sequence half the size of bands containing the desired gain
in each of the specified bands.
weight : array_like, optional
A relative weighting to give to each band region. The length of
`weight` has to be half the length of `bands`.
Hz : scalar, optional
The sampling frequency in Hz. Default is 1.
type : {'bandpass', 'differentiator', 'hilbert'}, optional
The type of filter:
'bandpass' : flat response in bands. This is the default.
'differentiator' : frequency proportional response in bands.
'hilbert' : filter with odd symmetry, that is, type III
(for even order) or type IV (for odd order)
linear phase filters.
maxiter : int, optional
Maximum number of iterations of the algorithm. Default is 25.
grid_density : int, optional
Grid density. The dense grid used in `remez` is of size
``(numtaps + 1) * grid_density``. Default is 16.
out : ndarray
A rank-1 array containing the coefficients of the optimal
(in a minimax sense) filter.
See Also
freqz : Compute the frequency response of a digital filter.
.. [1] J. H. McClellan and T. W. Parks, "A unified approach to the
design of optimum FIR linear phase digital filters",
IEEE Trans. Circuit Theory, vol. CT-20, pp. 697-701, 1973.
.. [2] J. H. McClellan, T. W. Parks and L. R. Rabiner, "A Computer
Program for Designing Optimum FIR Linear Phase Digital
Filters", IEEE Trans. Audio Electroacoust., vol. AU-21,
pp. 506-525, 1973.
We want to construct a filter with a passband at 0.2-0.4 Hz, and
stop bands at 0-0.1 Hz and 0.45-0.5 Hz. Note that this means that the
behavior in the frequency ranges between those bands is unspecified and
may overshoot.
>>> bpass = sp.signal.remez(72, [0, 0.1, 0.2, 0.4, 0.45, 0.5], [0, 1, 0])
>>> freq, response = sp.signal.freqz(bpass)
>>> ampl = np.abs(response)
>>> import matplotlib.pyplot as plt
>>> fig = plt.figure()
>>> ax1 = fig.add_subplot(111)
>>> ax1.semilogy(freq/(2*np.pi), ampl, 'b-') # freq in Hz
[<matplotlib.lines.Line2D object at 0xf486790>]
# Convert type
tnum = {'bandpass':1, 'differentiator':2, 'hilbert':3}[type]
except KeyError:
raise ValueError("Type must be 'bandpass', 'differentiator', or 'hilbert'")
# Convert weight
if weight is None:
weight = [1] * len(desired)
bands = np.asarray(bands).copy()
return sigtools._remez(numtaps, bands, desired, weight, tnum, Hz,
maxiter, grid_density)
Jump to Line
Something went wrong with that request. Please try again.