def test_boxcar(self):
w = windows.get_window('boxcar', 12)
assert_array_equal(w, np.ones_like(w))
# window is a tuple of len 1
w = windows.get_window(('boxcar',), 16)
assert_array_equal(w, np.ones_like(w))
def test_boxcar(self):
w = windows.get_window('boxcar', 12)
assert_array_equal(w, np.ones_like(w))
# window is a tuple of len 1
w = windows.get_window(('boxcar',), 16)
assert_array_equal(w, np.ones_like(w))
def windowff(nslices,kind=None):
"""
Returns a window function to enforce quasi-preiodicity
on an aperiodic signal. Window options are:
BlackmanHarris, Hanning, Blackman, FlatTop, Welch, Tukey
"""
import scipy.signal.windows as w
if kind is None:
return np.ones(nslices)
elif kind.lower() == 'tukey':
return w.get_window((kind.lower(),0.1),nslices)
else:
return w.get_window(kind.lower(),nslices)
def dziel_na_segmenty(self, window, nperseg, input_length):
if nperseg > input_length:
nperseg = input_length
win = get_window(window, nperseg)
return win, nperseg
def window_filter(curve: 'Curve',
window_size: int,
window_type: ty.Union[str, abc.Callable] = 'hann',
mode: str = 'reflect',
cval: float = 0.0) -> np.ndarray:
"""Smoothes a curve using moving average filter with the given window [1]_
References
----------
.. [1] `The windows in scipy
<https://docs.scipy.org/doc/scipy/reference/signal.windows.html#module-scipy.signal.windows>`_
"""
if callable(window_type):
try:
window = window_type(window_size)
except Exception as err:
raise ValueError(
f'Cannot create the window using {window_type}: {err}'
) from err
else:
window = windows.get_window(window_type, window_size, fftbins=False)
window /= window.sum()
return ndimage.convolve1d(
curve.data,
weights=window,
mode=mode,
cval=cval,
axis=0,
)
def test_rollingmean_detrend(series):
x = series(np.arange(12 * 365.25), "tas")
det = RollingMeanDetrend(group="time", win=29, min_periods=1)
fx = det.fit(x)
dx = fx.detrend(x)
xt = fx.retrend(dx)
np.testing.assert_array_almost_equal(dx.isel(time=slice(30, 3500)), 0)
np.testing.assert_array_almost_equal(xt, x)
# weights + grouping
x = xr.DataArray(
np.sin(2 * np.pi * np.arange(11 * 365) / 365),
dims=("time", ),
coords={
"time":
xr.cftime_range("2010-01-01",
periods=11 * 365,
freq="D",
calendar="noleap")
},
)
w = windows.get_window("triang", 11, False)
det = RollingMeanDetrend(group=Grouper("time.dayofyear", window=3),
win=11,
weights=w)
fx = det.fit(x)
assert fx.ds.trend.notnull().sum() == 365
def test_basic(self):
M = 100
w = windows.kaiser_bessel_derived(M, beta=4.0)
w2 = windows.get_window(('kaiser bessel derived', 4.0),
M,
fftbins=False)
assert_allclose(w, w2)
# Test for Princen-Bradley condition
assert_allclose(w[:M // 2]**2 + w[-M // 2:]**2, 1.)
# Test actual values from other implementations
# M = 2: sqrt(2) / 2
# M = 4: 0.518562710536, 0.855039598640
# M = 6: 0.436168993154, 0.707106781187, 0.899864772847
# Ref:https://github.com/scipy/scipy/pull/4747#issuecomment-172849418
assert_allclose(
windows.kaiser_bessel_derived(2, beta=np.pi / 2)[:1],
np.sqrt(2) / 2)
assert_allclose(
windows.kaiser_bessel_derived(4, beta=np.pi / 2)[:2],
[0.518562710536, 0.855039598640])
assert_allclose(
windows.kaiser_bessel_derived(6, beta=np.pi / 2)[:3],
[0.436168993154, 0.707106781187, 0.899864772847])
def test_array_as_window(self):
# github issue 3603
osfactor = 128
sig = np.arange(128)
win = windows.get_window(('kaiser', 8.0), osfactor // 2)
with assert_raises(ValueError, match='must have the same length'):
resample(sig, len(sig) * osfactor, window=win)
def __init__(self, window, source=None, **kw_get_win):
if isinstance(window, str) and source:
from scipy.signal.windows import get_window
window = get_window(window, Nx=source.blocksize, **kw_get_win)
else:
from numpy import asarray
window = asarray(window, dtype=float)
self.window = window
def test_array_as_window(self):
# github issue 3603
osfactor = 128
sig = np.arange(128)
win = windows.get_window(('kaiser', 8.0), osfactor // 2)
assert_raises(ValueError, resample,
(sig, len(sig) * osfactor), {'window': win})
def test_array_as_window(self):
# github issue 3603
osfactor = 128
sig = np.arange(128)
win = windows.get_window(('kaiser', 8.0), osfactor // 2)
assert_raises(ValueError, resample,
(sig, len(sig) * osfactor), {'window': win})
def compute_iq(spectra, fields_in_list, fields_out_list, window=None):
"""
Computes the IQ data from the spectra through an inverse Fourier transform
Parameters
----------
spectra : Spectra radar object
Object containing the spectra
fields_in_list : list of str
list of input spectra fields names
fields_out_list : list of str
list with the output IQ fields names obtained from the input fields
window : string, tupple or None
Parameters of the window used to obtain the spectra. The parameters
are the ones corresponding to function
scipy.signal.windows.get_window. If it is not None the inverse will be
used to multiply the IQ data obtained by the IFFT
Returns
-------
radar : IQ radar object
radar object containing the IQ fields
"""
radar = deepcopy(spectra)
radar.fields = {}
for field_name_in, field_name_out in zip(fields_in_list, fields_out_list):
if field_name_out in ('IQ_hh_ADU', 'IQ_vv_ADU'):
iq = np.ma.masked_all(
(spectra.nrays, spectra.ngates, spectra.npulses_max),
dtype=np.complex64)
for ray, npuls in enumerate(spectra.npulses['data']):
ray_data = spectra.fields[field_name_in]['data'][
ray, :, 0:npuls].filled(0.)
iq[ray, :,
0:npuls] = np.fft.ifft(np.fft.ifftshift(ray_data, axes=-1),
axis=-1) * npuls
if window is not None:
wind = get_window(window, npuls)
wind = wind / np.sqrt(np.sum(np.power(wind, 2.)) / npuls)
wind = np.broadcast_to(np.atleast_2d(wind),
(spectra.ngates, npuls))
iq[ray, :, 0:npuls] /= wind
else:
iq = np.ma.masked_all(
(spectra.nrays, spectra.ngates, spectra.npulses_max),
dtype=np.float32)
for ray, npuls in enumerate(spectra.npulses['data']):
iq[ray, :, 0:npuls] = spectra.fields[field_name_in]['data'][
ray, :, 0:npuls] * npuls
field_dict = get_metadata(field_name_out)
field_dict['data'] = iq
radar.fields.update({field_name_out: field_dict})
return radar
def get_offset_features(features_matlab_path, songName, data_out,
target_smear_width, frame_len, context_len,
audio_offset, pitch_shift, boundaries):
"""Load feature data files of offset hop durations and append only the boundary-neighborhood frames to the dataset. To improve data class imbalance.
Parameters:
--
songName (string): name of the data file
data_out (dict): features for default hop duration
target_smear_width (int): size of boundary-neighborhood (in seconds)
frame_len (float): duration of a frame (in seconds)
context_len (int): number of +-context frames
audio_offset (float): offset value (in seconds) added to audio signal
pitch_shift (int): pitch shift (in +/-semitones)
boundaries (numpy array or list): set of boundary positions for the song (in seconds)
Returns:
--
numpy ndarray: feature data for frames in the neighborhood
"""
data_offset = sio.loadmat(
os.path.join(
features_matlab_path,
songName + '_' + str(audio_offset) + '_' + str(pitch_shift)))
data_offset = make_feature_subsets(data_offset, context_len)
for bound in boundaries:
bound = int(bound.strip()) - audio_offset
bound_frame = int(bound / frame_len)
data_out['rhythm'] = np.vstack(
(data_out['rhythm'],
data_offset['rhythm'][bound_frame -
int(target_smear_width / 2):bound_frame +
int(target_smear_width / 2), :]))
data_out['mfcc'] = np.vstack(
(data_out['mfcc'],
data_offset['mfcc'][bound_frame -
int(target_smear_width / 2):bound_frame +
int(target_smear_width / 2), :]))
data_out['timbre'] = np.vstack(
(data_out['timbre'],
data_offset['timbre'][bound_frame -
int(target_smear_width / 2):bound_frame +
int(target_smear_width / 2), :]))
data_out['all'] = np.vstack(
(data_out['all'],
data_offset['all'][bound_frame -
int(target_smear_width / 2):bound_frame +
int(target_smear_width / 2), :]))
data_out['labels'] = np.vstack(
(data_out['labels'],
np.atleast_2d(windows.get_window('boxcar',
target_smear_width)).T))
return data_out
def compute_spectra(radar, fields_in_list, fields_out_list, window=None):
"""
Computes the spectra from IQ data through a Fourier transform
Parameters
----------
radar : radar object
Object containing the IQ data
fields_in_list : list of str
list of input IQ data fields names
fields_out_list : list of str
list with the output spectra fields names obtained from the input
fields
window : string, tupple or None
Parameters of the window used to obtain the spectra. The
parameters are the ones corresponding to function
scipy.signal.windows.get_window. If None no window will be used
Returns
-------
spectra : spectra radar object
radar object containing the spectra fields
"""
spectra = deepcopy(radar)
spectra.fields = {}
for field_name_in, field_name_out in zip(fields_in_list, fields_out_list):
if field_name_in in ('IQ_hh_ADU', 'IQ_vv_ADU'):
spectrum = np.ma.masked_all(
(radar.nrays, radar.ngates, radar.npulses_max),
dtype=np.complex64)
for ray, npuls in enumerate(radar.npulses['data']):
wind_data = radar.fields[field_name_in]['data'][
ray, :, 0:npuls].filled(0.)
if window is not None:
wind = get_window(window, npuls)
wind = wind/np.sqrt(np.sum(np.power(wind, 2.))/npuls)
wind = np.broadcast_to(
np.atleast_2d(wind), (radar.ngates, npuls))
wind_data *= wind
spectrum[ray, :, 0:npuls] = np.fft.fftshift(
np.fft.fft(wind_data, axis=-1)/npuls, axes=-1)
else:
spectrum = np.ma.masked_all(
(radar.nrays, radar.ngates, radar.npulses_max),
dtype=np.float32)
for ray, npuls in enumerate(radar.npulses['data']):
spectrum[ray, :, 0:npuls] = radar.fields[field_name_in]['data'][
ray, :, 0:npuls]/npuls
field_dict = get_metadata(field_name_out)
field_dict['data'] = spectrum
spectra.fields.update({field_name_out: field_dict})
return spectra
def filter_spectrum(intensity: float, window="hanning", sigma=0.5):
"""
Apply a specified window function to a signal. The window functions are
taken from the `signal.windows` module of SciPy, so check what is available
before throwing it into this function.
The window function is convolved with the signal by taking the time-
domain product, and doing the inverse FFT to get the convolved spectrum
back.
The one exception is the gaussian window - if a user specifies "gaussian"
for the window function, the actual window function applied here is a
half gaussian, i.e. a 1D gaussian blur.
Parameters
----------
dataframe: pandas DataFrame
Pandas dataframe containing the spectral information
int_col: str, optional
Column name to reference the signal
window: str, optional
Name of the window function as implemented in SciPy.
sigma:
Returns
-------
new_y: array_like
Numpy 1D array containing the convolved signal
"""
if window not in dir(windows):
raise Exception("Specified window not available in SciPy.")
data_length = len(intensity)
if window == "gaussian":
x = np.arange(data_length)
window = lineshapes.gaussian(
x, 1., 0., sigma
)
else:
window = windows.get_window(window, data_length)
fft_y = np.fft.fft(intensity)
# Convolve the signal with the window function
new_y = np.fft.ifft(
window * fft_y
)
# Return only the real part of the FFT
new_y = np.abs(new_y)
return new_y
def forward(self,
signal,
ndct=512,
numoverlap=256,
window='boxcar',
MFCC=False):
signal = np.array(signal)
if MFCC == False:
win = get_window(window, ndct)
listdct = []
for i in range(0, signal.shape[0] - ndct, numoverlap):
listdct.append(dct(signal[i:i + ndct] * win))
else:
listdct = mfcc(signal,
samplerate=512,
winlen=1,
winstep=0.5,
nfft=ndct)
return np.array(listdct)
def fourier_transform(dipole,
dt,
omega,
max_harmonic=30,
zero_padding_factor=10,
window='blackman',
t_phase=0,
time_axis=0):
'''Handles the Fourier transform for HHG with all the extra subtleties.
'dipole' is the time-dependent dipole moment (as a numpy array)
'dt' is the time step
'omega' is the reference frequency, to measure the harmonics against
'max_harmonic' is the highest frequency harmonic to return
'zero_padding_factor' is the factor by which the signal duration is increased via zero padding
'window' is the name of the window to use (see scipy.signal for the options)
't_phase' is the time relative to which the phase of the harmonics is measured. Typically the peak of the laser pulse.
'axis' is used when 'dipole' is a multidimensional array to indicate which axis is time
This function returns both the frequencies (in units of omega) and the (complex) hhg spectrum.
'''
#Total duration of the signal, after zero padding
num_samples = dipole.shape[time_axis]
t_max = zero_padding_factor * dt * num_samples
#Spacing in frequency domain
dw = 2 * np.pi / t_max
#Frequencies in units of omega
freq = np.arange(0, max_harmonic, dw / omega)
#Calculate the dipole acceleration in time domain by taking the second derivative.
dipole = np.gradient(np.gradient(dipole, dt, axis=time_axis),
dt,
axis=time_axis)
#Multiply by the window function
dipole *= get_window(window, num_samples)
#Compute the Fourier transform with zero-padding
spec = np.fft.rfft(dipole,
n=zero_padding_factor * num_samples,
axis=time_axis)
#Shift the phase to be measured relative to 't_phase'.
spec = multiply_along_axis(spec, np.exp(1j * t_phase * freq * omega))
#Truncate the spectrum at the max harmonic, and also remove
#the first point to avoid division by zero later
return freq[1:], spec[1:len(freq)]
def save_feats_labels(songdata, audio_offsets=[], pitch_shifts=[]):
""" Generate frame-level labels from GT boundary annotations (binary: 1 for boundary, 0 for non-boundary) and save the features and labels together
Parameters:
--
_songdata (pd DataFrame): data from GT annotations csv file
"""
for songNo in range(songdata.shape[0]):
boundaries = songdata['Boundaries'][songNo].split(',')
data = sio.loadmat(
os.path.join(features_matlab_path,
songdata['Concert name'][songNo]))
data_out = utils.make_feature_subsets(data, context_len)
n_frames = data_out['all'].shape[0]
smear_win = windows.get_window('boxcar', target_smear_width)
boundLabels = np.zeros(n_frames)
for i_bound in range(1, len(boundaries)):
boundary = int(boundaries[i_bound].strip())
boundLabels[boundary // frame_len -
target_smear_width // 2:boundary // frame_len +
target_smear_width // 2] = smear_win
data_out['labels'] = np.atleast_2d(boundLabels).T
for pitch_shift in pitch_shifts:
for audio_offset in audio_offsets:
if (pitch_shift == 0) & (audio_offset == 0): continue
data_out = utils.get_offset_features(
features_matlab_path, songdata['Concert name'][songNo],
data_out, target_smear_width, frame_len, context_len,
audio_offset, pitch_shift, boundaries)
np.save(
os.path.join(features_RF_path,
songdata['Concert name'][songNo] + '.npy'), data_out)
return
def main(fname, window):
d = ""
with open(fname, 'r') as f:
d = f.read()
header, data = ReadKeyValueDataFile(d, header_char='#', delim = ':')
header = dict(header)
x, y = get_adjusted_values(header, data)
sampling_rate = x[1]-x[0]
window_obj = windows.get_window(window, len(x))
x = np.array(x*window_obj)
f = scipy.fft.fft(y)
sp = scipy.fft.fftshift(f)
freq = scipy.fft.fftshift(scipy.fft.fftfreq(len(x), sampling_rate))
plt.plot(freq/1e6, np.abs(sp))
plt.xlabel("Frequency (MHz)")
yunit = header['YUNIT'].strip("\"'")
plt.ylabel("Amplitude ({})".format(yunit))
plt.xlim([0, max(freq)/1e6])
plt.yscale("log")
plt.show()
def test_kaiser_float(self):
win1 = windows.get_window(7.2, 64)
win2 = windows.kaiser(64, 7.2, False)
assert_allclose(win1, win2)
def CreateSegments(self, Window, nperseg, inputLenght):
if nperseg > inputLenght:
nperseg = inputLenght
Window = get_window(Window, nperseg)
return Window, nperseg
Description:
Make sure the windowing effects are well-understood
Date:
3/9/2021
Author: Hunter Akins
Institution: Scripps Institution of Oceanography, UC San Diego
"""
from scipy.signal.windows import get_window
window_len = 1024
N_fft = 4 * 8096
hamm = get_window('hamming', window_len)
transfer_func = np.fft.fft(hamm)
freqs = np.fft.fftfreq(window_len, 1 / 1500)
transfer_func = np.square(abs(transfer_func))
transfer_func /= np.max(transfer_func)
trans_db = 10 * np.log10(transfer_func)
plt.figure()
plt.plot(freqs, trans_db)
hamm_pad = np.pad(hamm, (0, N_fft - window_len), 'constant')
transfer_func = np.fft.fft(hamm_pad)
freqs = np.fft.fftfreq(N_fft, 1 / 1500)
def test_cheb_even(self):
with suppress_warnings() as sup:
sup.filter(UserWarning, "This window is not suitable")
w = windows.get_window(('chebwin', 40), 54, fftbins=False)
assert_array_almost_equal(w, cheb_even_true, decimal=4)
def gen_dataset(features_out_path,
texwin_len,
context_len,
targ_smear_width,
audio_offsets=[0],
pitch_shifts=[0]):
smear_win = windows.get_window('boxcar', targ_smear_width)
dataset_ids = {}
dataset_labels = {}
print('\n----------\n')
for i_song in range(songdata.shape[0]):
print('Generating data for song no %d' % i_song)
dataset_ids[str(i_song)] = []
dataset_labels[str(i_song)] = {}
boundaries = songdata['Boundaries'][i_song].split(',')
for i_pitch in pitch_shifts:
if i_pitch == 0:
filepath = os.path.join(
audio_dir, songdata['Concert name'][i_song] + '.wav')
else:
filepath = os.path.join(
audio_dir, 'pitch_shifted',
songdata['Concert name'][i_song] + '_' + str(i_pitch) +
'.wav')
for offset in audio_offsets:
audio, sr = librosa.load(filepath, sr=16000)
audio = audio[int(sr * offset):]
features = utils.get_frame_level_melgrams(audio)
audio = []
labels = np.zeros(features.shape[0])
for boundary in boundaries:
boundary = int(boundary.strip()) - offset
labels[int(boundary / hop_len_sub) -
targ_smear_width // 2:int(boundary / hop_len_sub) +
targ_smear_width // 2] = smear_win
dataset_ids_temp = [] #IDs of this version of the audio
labels_temp = [
] #retaining labels only in boundary neighborhood
for i_frame in range(features.shape[0]):
ID = '%d_%d_%f_%d' % (i_song, i_pitch, offset, i_frame)
mel_specgram_frame = features[i_frame, ]
if (((i_pitch == 0) & (offset == 0)) |
(labels[i_frame] > 0)):
np.save(
os.path.join(features_out_path, ID),
np.reshape(mel_specgram_frame,
(mel_specgram_frame.shape[0],
mel_specgram_frame.shape[1], 1)))
dataset_ids_temp.append(ID)
labels_temp.append(labels[i_frame])
dataset_ids[str(i_song)].extend(dataset_ids_temp)
dataset_labels[str(i_song)].update(
dict(zip(dataset_ids_temp, labels_temp)))
return dataset_ids, dataset_labels
def test_get_pcen_settings():
settings = get_pcen_settings()
assert type(settings) == dict
assert 'sr' in settings
assert isinstance(settings['sr'], Real)
assert settings['sr'] > 0
assert 'fmin' in settings
assert isinstance(settings['fmin'], Real)
assert settings['fmin'] > 0
assert 'fmax' in settings
assert isinstance(settings['fmax'], Real)
assert settings['fmax'] > settings['fmin']
assert settings['sr'] / 2.0 >= settings['fmax']
assert 'hop_length' in settings
assert isinstance(settings['hop_length'], int)
assert settings['hop_length'] > 0
assert 'n_fft' in settings
assert isinstance(settings['n_fft'], int)
assert settings['n_fft'] > 0
assert 'n_mels' in settings
assert isinstance(settings['n_mels'], int)
assert settings['n_fft'] > settings['n_mels'] > 0
assert 'pcen_delta' in settings
assert isinstance(settings['pcen_delta'], float)
assert settings['pcen_delta'] > 0
assert 'pcen_time_constant' in settings
assert isinstance(settings['pcen_time_constant'], float)
assert settings['pcen_time_constant'] > 0
assert 'pcen_norm_exponent' in settings
assert isinstance(settings['pcen_norm_exponent'], float)
assert settings['pcen_norm_exponent'] > 0
assert 'pcen_power' in settings
assert isinstance(settings['pcen_power'], float)
assert settings['pcen_power'] > 0
assert 'top_freq_id' in settings
assert isinstance(settings['top_freq_id'], int)
assert settings['top_freq_id'] > 0
assert 'win_length' in settings
assert isinstance(settings['win_length'], int)
assert settings['win_length'] > 0
assert 'n_hops' in settings
assert isinstance(settings['n_hops'], int)
assert settings['n_hops'] > 0
assert 'window' in settings
assert isinstance(settings['window'], string_types)
# Make sure window is valid
get_window(settings['window'], 5)
def test_dpss(self):
win1 = windows.get_window(('dpss', 3), 64, fftbins=False)
win2 = windows.dpss(64, 3)
assert_array_almost_equal(win1, win2, decimal=4)
def test_cheb_even(self):
with suppress_warnings() as sup:
sup.filter(UserWarning, "This window is not suitable")
w = windows.get_window(('chebwin', 40), 54, fftbins=False)
assert_array_almost_equal(w, cheb_even_true, decimal=4)
def tocovdata(self, timeseries):
"""
Return auto covariance function from data.
Return
-------
acf : CovData1D object
with attributes:
data : ACF vector length L+1
args : time lags length L+1
sigma : estimated large lag standard deviation of the estimate
assuming x is a Gaussian process:
if acf[k]=0 for all lags k>q then an approximation
of the variance for large samples due to Bartlett
var(acf[k])=1/N*(acf[0]**2+2*acf[1]**2+2*acf[2]**2+ ..+2*acf[q]**2)
for k>q and where N=length(x). Special case is
white noise where it equals acf[0]**2/N for k>0
norm : bool
If false indicating that auto_cov is not normalized
Example:
--------
>>> import wafo.data
>>> import wafo.objects as wo
>>> x = wafo.data.sea()
>>> ts = wo.mat2timeseries(x)
>>> acf = ts.tocovdata(150)
h = acf.plot()
"""
lag = self.lag
window = self.window
detrend = self.detrend
try:
x = timeseries.data.flatten('F')
dt = timeseries.sampling_period()
except Exception:
x = timeseries[:, 1:].flatten('F')
dt = sampling_period(timeseries[:, 0])
if self.dt is not None:
dt = self.dt
if self.tr is not None:
x = self.tr.dat2gauss(x)
n = len(x)
indnan = np.isnan(x)
if any(indnan):
x = x - x[1 - indnan].mean()
Ncens = n - indnan.sum()
x[indnan] = 0.
else:
Ncens = n
x = x - x.mean()
if hasattr(detrend, '__call__'):
x = detrend(x)
nfft = 2 ** nextpow2(n)
raw_periodogram = abs(fft(x, nfft)) ** 2 / Ncens
# ifft = fft/nfft since raw_periodogram is real!
auto_cov = np.real(fft(raw_periodogram)) / nfft
if self.flag.startswith('unbiased'):
# unbiased result, i.e. divide by n-abs(lag)
auto_cov = auto_cov[:Ncens] * Ncens / np.arange(Ncens, 1, -1)
if self.norm:
auto_cov = auto_cov / auto_cov[0]
if lag is None:
lag = self._estimate_lag(auto_cov, Ncens)
lag = min(lag, n - 2)
if isinstance(window, str) or type(window) is tuple:
win = get_window(window, 2 * lag - 1)
else:
win = np.asarray(window)
auto_cov[:lag] = auto_cov[:lag] * win[lag - 1::]
auto_cov[lag] = 0
lags = slice(0, lag + 1)
t = np.linspace(0, lag * dt, lag + 1)
acf = CovData1D(auto_cov[lags], t)
acf.sigma = np.sqrt(np.r_[0, auto_cov[0] ** 2,
auto_cov[0] ** 2 + 2 * np.cumsum(auto_cov[1:] ** 2)] / Ncens)
acf.children = [PlotData(-2. * acf.sigma[lags], t),
PlotData(2. * acf.sigma[lags], t)]
acf.plot_args_children = ['r:']
acf.norm = self.norm
return acf
def _welch(x, y, fs=1.0, window='hanning', nperseg=256, noverlap=None,
nfft=None, detrend='constant', scaling='density', axis=-1):
"""
A helper function to estimate cross spectral density using Welch's method.
This function is a slightly modified version of `scipy.signal.welch()` with
modifications based on `matplotlib.mlab._spectral_helper()`.
Welch's method [1]_ computes an estimate of the cross spectral density
by dividing the data into overlapping segments, computing a modified
periodogram for each segment and averaging the cross-periodograms.
Parameters
----------
x, y : array_like
Time series of measurement values
fs : float, optional
Sampling frequency of the `x` and `y` time series in units of Hz.
Defaults to 1.0.
window : str or tuple or array_like, optional
Desired window to use. See `get_window` for a list of windows and
required parameters. If `window` is array_like it will be used
directly as the window and its length will be used for nperseg.
Defaults to 'hanning'.
nperseg : int, optional
Length of each segment. Defaults to 256.
noverlap: int, optional
Number of points to overlap between segments. If None,
``noverlap = nperseg / 2``. Defaults to None.
nfft : int, optional
Length of the FFT used, if a zero padded FFT is desired. If None,
the FFT length is `nperseg`. Defaults to None.
detrend : str or function, optional
Specifies how to detrend each segment. If `detrend` is a string,
it is passed as the ``type`` argument to `detrend`. If it is a
function, it takes a segment and returns a detrended segment.
Defaults to 'constant'.
scaling : { 'density', 'spectrum' }, optional
Selects between computing the power spectral density ('density')
where Pxx has units of V**2/Hz if x is measured in V and computing
the power spectrum ('spectrum') where Pxx has units of V**2 if x is
measured in V. Defaults to 'density'.
axis : int, optional
Axis along which the periodogram is computed; the default is over
the last axis (i.e. ``axis=-1``).
Returns
-------
f : ndarray
Array of sample frequencies.
Pxy : ndarray
Cross spectral density or cross spectrum of x and y.
Notes
-----
An appropriate amount of overlap will depend on the choice of window
and on your requirements. For the default 'hanning' window an
overlap of 50% is a reasonable trade off between accurately estimating
the signal power, while not over counting any of the data. Narrower
windows may require a larger overlap.
If `noverlap` is 0, this method is equivalent to Bartlett's method [2]_.
References
----------
.. [1] P. Welch, "The use of the fast Fourier transform for the
estimation of power spectra: A method based on time averaging
over short, modified periodograms", IEEE Trans. Audio
Electroacoust. vol. 15, pp. 70-73, 1967.
.. [2] M.S. Bartlett, "Periodogram Analysis and Continuous Spectra",
Biometrika, vol. 37, pp. 1-16, 1950.
"""
# TODO: This function should be replaced by `scipy.signal.csd()`, which
# will appear in SciPy 0.16.0.
# The checks for if y is x are so that we can use the same function to
# obtain both power spectrum and cross spectrum without doing extra
# calculations.
same_data = y is x
# Make sure we're dealing with a numpy array. If y and x were the same
# object to start with, keep them that way
x = np.asarray(x)
if same_data:
y = x
else:
if x.shape != y.shape:
raise ValueError("x and y must be of the same shape.")
y = np.asarray(y)
if x.size == 0:
return np.empty(x.shape), np.empty(x.shape)
if axis != -1:
x = np.rollaxis(x, axis, len(x.shape))
if not same_data:
y = np.rollaxis(y, axis, len(y.shape))
if x.shape[-1] < nperseg:
warnings.warn('nperseg = %d, is greater than x.shape[%d] = %d, using '
'nperseg = x.shape[%d]'
% (nperseg, axis, x.shape[axis], axis))
nperseg = x.shape[-1]
if isinstance(window, string_types) or type(window) is tuple:
win = get_window(window, nperseg)
else:
win = np.asarray(window)
if len(win.shape) != 1:
raise ValueError('window must be 1-D')
if win.shape[0] > x.shape[-1]:
raise ValueError('window is longer than x.')
nperseg = win.shape[0]
if scaling == 'density':
scale = 1.0 / (fs * (win * win).sum())
elif scaling == 'spectrum':
scale = 1.0 / win.sum()**2
else:
raise ValueError('Unknown scaling: %r' % scaling)
if noverlap is None:
noverlap = nperseg // 2
elif noverlap >= nperseg:
raise ValueError('noverlap must be less than nperseg.')
if nfft is None:
nfft = nperseg
elif nfft < nperseg:
raise ValueError('nfft must be greater than or equal to nperseg.')
if not hasattr(detrend, '__call__'):
detrend_func = lambda seg: signaltools.detrend(seg, type=detrend)
elif axis != -1:
# Wrap this function so that it receives a shape that it could
# reasonably expect to receive.
def detrend_func(seg):
seg = np.rollaxis(seg, -1, axis)
seg = detrend(seg)
return np.rollaxis(seg, axis, len(seg.shape))
else:
detrend_func = detrend
step = nperseg - noverlap
indices = np.arange(0, x.shape[-1] - nperseg + 1, step)
for k, ind in enumerate(indices):
x_dt = detrend_func(x[..., ind:ind + nperseg])
xft = fftpack.fft(x_dt * win, nfft)
if same_data:
yft = xft
else:
y_dt = detrend_func(y[..., ind:ind + nperseg])
yft = fftpack.fft(y_dt * win, nfft)
if k == 0:
Pxy = (xft * yft.conj())
else:
Pxy *= k / (k + 1.0)
Pxy += (xft * yft.conj()) / (k + 1.0)
Pxy *= scale
f = fftpack.fftfreq(nfft, 1.0 / fs)
if axis != -1:
Pxy = np.rollaxis(Pxy, -1, axis)
return f, Pxy
def _spectral_helper(x, y, fs=1.0, window='hann', nperseg=256,
noverlap=None, nfft=None, detrend='constant',
return_onesided=True, scaling='spectrum', axis=-1,
mode='psd'):
"""
Calculate various forms of windowed FFTs for PSD, CSD, etc.
This is a helper function that implements the commonality between the
psd, csd, and spectrogram functions. It is not designed to be called
externally. The windows are not averaged over; the result from each window
is returned.
Parameters
---------
x : array_like
Array or sequence containing the data to be analyzed.
y : array_like
Array or sequence containing the data to be analyzed. If this is
the same object in memoery as x (i.e. _spectral_helper(x, x, ...)),
the extra computations are spared.
fs : float, optional
Sampling frequency of the time series. Defaults to 1.0.
window : str or tuple or array_like, optional
Desired window to use. See `get_window` for a list of windows and
required parameters. If `window` is array_like it will be used
directly as the window and its length will be used for nperseg.
Defaults to 'hann'.
nperseg : int, optional
Length of each segment. Defaults to 256.
noverlap : int, optional
Number of points to overlap between segments. If None,
``noverlap = nperseg // 2``. Defaults to None.
nfft : int, optional
Length of the FFT used, if a zero padded FFT is desired. If None,
the FFT length is `nperseg`. Defaults to None.
detrend : str or function or False, optional
Specifies how to detrend each segment. If `detrend` is a string,
it is passed as the ``type`` argument to `detrend`. If it is a
function, it takes a segment and returns a detrended segment.
If `detrend` is False, no detrending is done. Defaults to 'constant'.
return_onesided : bool, optional
If True, return a one-sided spectrum for real data. If False return
a two-sided spectrum. Note that for complex data, a two-sided
spectrum is always returned.
scaling : { 'density', 'spectrum' }, optional
Selects between computing the cross spectral density ('density')
where `Pxy` has units of V**2/Hz and computing the cross spectrum
('spectrum') where `Pxy` has units of V**2, if `x` and `y` are
measured in V and fs is measured in Hz. Defaults to 'density'
axis : int, optional
Axis along which the periodogram is computed; the default is over
the last axis (i.e. ``axis=-1``).
mode : str, optional
Defines what kind of return values are expected. Options are ['psd',
'complex', 'magnitude', 'angle', 'phase'].
Returns
-------
freqs : ndarray
Array of sample frequencies.
t : ndarray
Array of times corresponding to each data segment
result : ndarray
Array of output data, contents dependent on *mode* kwarg.
References
----------
.. [1] Stack Overflow, "Rolling window for 1D arrays in Numpy?",
http://stackoverflow.com/a/6811241
.. [2] Stack Overflow, "Using strides for an efficient moving average
filter", http://stackoverflow.com/a/4947453
Notes
-----
Adapted from matplotlib.mlab
.. versionadded:: 0.16.0
"""
from scipy import fftpack
from scipy.signal import signaltools
from scipy.signal.windows import get_window
if mode not in ['psd', 'complex', 'magnitude', 'angle', 'phase']:
raise ValueError("Unknown value for mode %s, must be one of: "
"'default', 'psd', 'complex', "
"'magnitude', 'angle', 'phase'" % mode)
# If x and y are the same object we can save ourselves some computation.
same_data = y is x
if not same_data and mode != 'psd':
raise ValueError("x and y must be equal if mode is not 'psd'")
axis = int(axis)
# Ensure we have np.arrays, get outdtype
x = np.asarray(x)
if not same_data:
y = np.asarray(y)
outdtype = np.result_type(x,y,np.complex64)
else:
outdtype = np.result_type(x,np.complex64)
if not same_data:
# Check if we can broadcast the outer axes together
xouter = list(x.shape)
youter = list(y.shape)
xouter.pop(axis)
youter.pop(axis)
try:
outershape = np.broadcast(np.empty(xouter), np.empty(youter)).shape
except ValueError:
raise ValueError('x and y cannot be broadcast together.')
if same_data:
if x.size == 0:
return np.empty(x.shape), np.empty(x.shape), np.empty(x.shape)
else:
if x.size == 0 or y.size == 0:
outshape = outershape + (min([x.shape[axis], y.shape[axis]]),)
emptyout = np.rollaxis(np.empty(outshape), -1, axis)
return emptyout, emptyout, emptyout
if x.ndim > 1:
if axis != -1:
x = np.rollaxis(x, axis, len(x.shape))
if not same_data and y.ndim > 1:
y = np.rollaxis(y, axis, len(y.shape))
# Check if x and y are the same length, zero-pad if necessary
if not same_data:
if x.shape[-1] != y.shape[-1]:
if x.shape[-1] < y.shape[-1]:
pad_shape = list(x.shape)
pad_shape[-1] = y.shape[-1] - x.shape[-1]
x = np.concatenate((x, np.zeros(pad_shape)), -1)
else:
pad_shape = list(y.shape)
pad_shape[-1] = x.shape[-1] - y.shape[-1]
y = np.concatenate((y, np.zeros(pad_shape)), -1)
# X and Y are same length now, can test nperseg with either
if x.shape[-1] < nperseg:
warnings.warn('nperseg = {0:d}, is greater than input length = {1:d}, '
'using nperseg = {1:d}'.format(nperseg, x.shape[-1]))
nperseg = x.shape[-1]
nperseg = int(nperseg)
if nperseg < 1:
raise ValueError('nperseg must be a positive integer')
if nfft is None:
nfft = nperseg
elif nfft < nperseg:
raise ValueError('nfft must be greater than or equal to nperseg.')
else:
nfft = int(nfft)
if noverlap is None:
noverlap = nperseg//2
elif noverlap >= nperseg:
raise ValueError('noverlap must be less than nperseg.')
else:
noverlap = int(noverlap)
# Handle detrending and window functions
if not detrend:
def detrend_func(d):
return d
elif not hasattr(detrend, '__call__'):
def detrend_func(d):
return signaltools.detrend(d, type=detrend, axis=-1)
elif axis != -1:
# Wrap this function so that it receives a shape that it could
# reasonably expect to receive.
def detrend_func(d):
d = np.rollaxis(d, -1, axis)
d = detrend(d)
return np.rollaxis(d, axis, len(d.shape))
else:
detrend_func = detrend
if isinstance(window, string_types) or type(window) is tuple:
win = get_window(window, nperseg)
else:
win = np.asarray(window)
if len(win.shape) != 1:
raise ValueError('window must be 1-D')
if win.shape[0] != nperseg:
raise ValueError('window must have length of nperseg')
if np.result_type(win,np.complex64) != outdtype:
win = win.astype(outdtype)
if mode == 'psd':
if scaling == 'density':
scale = 1.0 / (fs * (win*win).sum())
elif scaling == 'spectrum':
scale = 1.0 / win.sum()**2
else:
raise ValueError('Unknown scaling: %r' % scaling)
else:
scale = 1
if return_onesided is True:
if np.iscomplexobj(x):
sides = 'twosided'
else:
sides = 'onesided'
if not same_data:
if np.iscomplexobj(y):
sides = 'twosided'
else:
sides = 'twosided'
if sides == 'twosided':
num_freqs = nfft
elif sides == 'onesided':
if nfft % 2:
num_freqs = (nfft + 1)//2
else:
num_freqs = nfft//2 + 1
# Perform the windowed FFTs
result = _fft_helper(x, win, detrend_func, nperseg, noverlap, nfft)
result = result[..., :num_freqs]
freqs = fftpack.fftfreq(nfft, 1/fs)[:num_freqs]
if not same_data:
# All the same operations on the y data
result_y = _fft_helper(y, win, detrend_func, nperseg, noverlap, nfft)
result_y = result_y[..., :num_freqs]
result = np.conjugate(result) * result_y
elif mode == 'psd':
result = np.conjugate(result) * result
elif mode == 'magnitude':
result = np.absolute(result)
elif mode == 'angle' or mode == 'phase':
result = np.angle(result)
elif mode == 'complex':
pass
result *= scale
if sides == 'onesided':
if nfft % 2:
result[...,1:] *= 2
else:
# Last point is unpaired Nyquist freq point, don't double
result[...,1:-1] *= 2
t = np.arange(nperseg/2, x.shape[-1] - nperseg/2 + 1, nperseg - noverlap)/float(fs)
if sides != 'twosided' and not nfft % 2:
# get the last value correctly, it is negative otherwise
freqs[-1] *= -1
# we unwrap the phase here to handle the onesided vs. twosided case
if mode == 'phase':
result = np.unwrap(result, axis=-1)
result = result.astype(outdtype)
# All imaginary parts are zero anyways
if same_data and mode != 'complex':
result = result.real
# Output is going to have new last axis for window index
if axis != -1:
# Specify as positive axis index
if axis < 0:
axis = len(result.shape)-1-axis
# Roll frequency axis back to axis where the data came from
result = np.rollaxis(result, -1, axis)
else:
# Make sure window/time index is last axis
result = np.rollaxis(result, -1, -2)
return freqs, t, result
def energy_correction(window, n_samples):
window = windows.get_window(window, n_samples)
return len(window) / np.sum(window**2)
def timeseries_to_spectrogram(timeseries: Timeseries,
fft_size: int = 1024,
n_samples: int = 1024,
n_records: int = None,
synchronization_offset: int = 0,
overlap: float = 0.0,
window: str = None):
"""
# TODO: Introduce overlap + windowing
Args:
n_records:
synchronization_offset:
timeseries:
fft_size:
n_samples:
overlap:
window: str
Returns:
"""
if overlap >= 1:
raise ValueError("Overlap percentage must be less than 100%")
if overlap < 0:
raise ValueError("Overlap percentage must be equal or greater than 0%")
total_samples = n_samples + synchronization_offset
overlap_samples = n_samples * (1 - overlap)
if n_records is None:
n_records = int(
((len(timeseries.data) - total_samples) // overlap_samples)) + 1
start_samples = [
int(i * total_samples * (1 - overlap)) for i in range(n_records)
]
time_data = np.zeros((fft_size, n_records), dtype=float)
# Reshape the timeseries data into bins of n_time_samples:
input_data = np.reshape(np.array(
list(
zip(*(timeseries.data['amplitude'][i:i + total_samples]
for i in start_samples)))), (total_samples, n_records),
order='F')
# input_data = np.reshape(timeseries.data['amplitude'][:(n_records * total_samples)], (total_samples, n_records),
# order='F')
time_data[:n_samples, :] = input_data[:n_samples, :]
if window is None:
window = np.ones(n_samples)
else:
window = windows.get_window(window, n_samples)
# Apply window
time_data *= np.transpose(np.tile(window, (time_data.shape[1], 1)))
spectrogram_data = fft.fft(time_data / timeseries.sample_rate,
fft_size,
axis=0)
t_res = timeseries.duration() / (int(
((len(timeseries.data) - total_samples) // overlap_samples)) + 1)
f_res = timeseries.sample_rate / fft_size
return Spectrogram(spectrogram_data, f_res, t_res)
def _welch(x,
y,
fs=1.0,
window='hanning',
nperseg=256,
noverlap=None,
nfft=None,
detrend='constant',
scaling='density',
axis=-1):
"""
A helper function to estimate cross spectral density using Welch's method.
This function is a slightly modified version of `scipy.signal.welch()` with
modifications based on `matplotlib.mlab._spectral_helper()`.
Welch's method [1]_ computes an estimate of the cross spectral density
by dividing the data into overlapping segments, computing a modified
periodogram for each segment and averaging the cross-periodograms.
Parameters
----------
x, y : array_like
Time series of measurement values
fs : float, optional
Sampling frequency of the `x` and `y` time series in units of Hz.
Defaults to 1.0.
window : str or tuple or array_like, optional
Desired window to use. See `get_window` for a list of windows and
required parameters. If `window` is array_like it will be used
directly as the window and its length will be used for nperseg.
Defaults to 'hanning'.
nperseg : int, optional
Length of each segment. Defaults to 256.
noverlap: int, optional
Number of points to overlap between segments. If None,
``noverlap = nperseg / 2``. Defaults to None.
nfft : int, optional
Length of the FFT used, if a zero padded FFT is desired. If None,
the FFT length is `nperseg`. Defaults to None.
detrend : str or function, optional
Specifies how to detrend each segment. If `detrend` is a string,
it is passed as the ``type`` argument to `detrend`. If it is a
function, it takes a segment and returns a detrended segment.
Defaults to 'constant'.
scaling : { 'density', 'spectrum' }, optional
Selects between computing the power spectral density ('density')
where Pxx has units of V**2/Hz if x is measured in V and computing
the power spectrum ('spectrum') where Pxx has units of V**2 if x is
measured in V. Defaults to 'density'.
axis : int, optional
Axis along which the periodogram is computed; the default is over
the last axis (i.e. ``axis=-1``).
Returns
-------
f : ndarray
Array of sample frequencies.
Pxy : ndarray
Cross spectral density or cross spectrum of x and y.
Notes
-----
An appropriate amount of overlap will depend on the choice of window
and on your requirements. For the default 'hanning' window an
overlap of 50% is a reasonable trade off between accurately estimating
the signal power, while not over counting any of the data. Narrower
windows may require a larger overlap.
If `noverlap` is 0, this method is equivalent to Bartlett's method [2]_.
References
----------
.. [1] P. Welch, "The use of the fast Fourier transform for the
estimation of power spectra: A method based on time averaging
over short, modified periodograms", IEEE Trans. Audio
Electroacoust. vol. 15, pp. 70-73, 1967.
.. [2] M.S. Bartlett, "Periodogram Analysis and Continuous Spectra",
Biometrika, vol. 37, pp. 1-16, 1950.
"""
# TODO: This function should be replaced by `scipy.signal.csd()`, which
# will appear in SciPy 0.16.0.
# The checks for if y is x are so that we can use the same function to
# obtain both power spectrum and cross spectrum without doing extra
# calculations.
same_data = y is x
# Make sure we're dealing with a numpy array. If y and x were the same
# object to start with, keep them that way
x = np.asarray(x)
if same_data:
y = x
else:
if x.shape != y.shape:
raise ValueError("x and y must be of the same shape.")
y = np.asarray(y)
if x.size == 0:
return np.empty(x.shape), np.empty(x.shape)
if axis != -1:
x = np.rollaxis(x, axis, len(x.shape))
if not same_data:
y = np.rollaxis(y, axis, len(y.shape))
if x.shape[-1] < nperseg:
warnings.warn('nperseg = %d, is greater than x.shape[%d] = %d, using '
'nperseg = x.shape[%d]' %
(nperseg, axis, x.shape[axis], axis))
nperseg = x.shape[-1]
if isinstance(window, string_types) or type(window) is tuple:
win = get_window(window, nperseg)
else:
win = np.asarray(window)
if len(win.shape) != 1:
raise ValueError('window must be 1-D')
if win.shape[0] > x.shape[-1]:
raise ValueError('window is longer than x.')
nperseg = win.shape[0]
if scaling == 'density':
scale = 1.0 / (fs * (win * win).sum())
elif scaling == 'spectrum':
scale = 1.0 / win.sum()**2
else:
raise ValueError('Unknown scaling: %r' % scaling)
if noverlap is None:
noverlap = nperseg // 2
elif noverlap >= nperseg:
raise ValueError('noverlap must be less than nperseg.')
if nfft is None:
nfft = nperseg
elif nfft < nperseg:
raise ValueError('nfft must be greater than or equal to nperseg.')
if not hasattr(detrend, '__call__'):
detrend_func = lambda seg: signaltools.detrend(seg, type=detrend)
elif axis != -1:
# Wrap this function so that it receives a shape that it could
# reasonably expect to receive.
def detrend_func(seg):
seg = np.rollaxis(seg, -1, axis)
seg = detrend(seg)
return np.rollaxis(seg, axis, len(seg.shape))
else:
detrend_func = detrend
step = nperseg - noverlap
indices = np.arange(0, x.shape[-1] - nperseg + 1, step)
for k, ind in enumerate(indices):
x_dt = detrend_func(x[..., ind:ind + nperseg])
xft = fftpack.fft(x_dt * win, nfft)
if same_data:
yft = xft
else:
y_dt = detrend_func(y[..., ind:ind + nperseg])
yft = fftpack.fft(y_dt * win, nfft)
if k == 0:
Pxy = (xft * yft.conj())
else:
Pxy *= k / (k + 1.0)
Pxy += (xft * yft.conj()) / (k + 1.0)
Pxy *= scale
f = fftpack.fftfreq(nfft, 1.0 / fs)
if axis != -1:
Pxy = np.rollaxis(Pxy, -1, axis)
return f, Pxy
def resample(x, up, down, npad=100, window='boxcar', n_jobs=1, verbose=None):
"""Resample the array x
Operates along the last dimension of the array.
Parameters
----------
x : n-d array
Signal to resample.
up : float
Factor to upsample by.
down : float
Factor to downsample by.
npad : integer
Number of samples to use at the beginning and end for padding.
window : string or tuple
See scipy.signal.resample for description.
n_jobs : int | str
Number of jobs to run in parallel. Can be 'cuda' if scikits.cuda
is installed properly and CUDA is initialized.
verbose : bool, str, int, or None
If not None, override default verbose level (see mne.verbose).
Returns
-------
xf : array
x resampled.
Notes
-----
This uses (hopefully) intelligent edge padding and frequency-domain
windowing improve scipy.signal.resample's resampling method, which
we have adapted for our use here. Choices of npad and window have
important consequences, and the default choices should work well
for most natural signals.
Resampling arguments are broken into "up" and "down" components for future
compatibility in case we decide to use an upfirdn implementation. The
current implementation is functionally equivalent to passing
up=up/down and down=1.
"""
# make sure our arithmetic will work
ratio = float(up) / down
x, orig_shape = _prep_for_filtering(x, False)[:2]
x_len = x.shape[1]
if x_len > 0:
# prep for resampling now
orig_len = x_len + 2 * npad # length after padding
new_len = ratio * orig_len # length after resampling padded signal
to_remove = np.round(ratio * npad).astype(int)
# figure out windowing function
if window is not None:
if callable(window):
W = window(fftfreq(orig_len))
elif isinstance(window, np.ndarray) and \
window.shape == (orig_len,):
W = window
else:
W = ifftshift(get_window(window, orig_len))
else:
W = np.ones(orig_len)
W *= (float(new_len) / float(orig_len))
W = W.astype(np.complex128)
# figure out if we should use CUDA
n_jobs, cuda_dict, W = setup_cuda_fft_resample(n_jobs, W, new_len)
# do the resampling using an adaptation of scipy's FFT-based resample()
# use of the 'flat' window is recommended for minimal ringing
if n_jobs == 1:
y = np.zeros((len(x), new_len - 2 * to_remove), dtype=x.dtype)
for xi, x_ in enumerate(x):
y[xi] = fft_resample(x_, W, new_len, npad, to_remove,
cuda_dict)
else:
if not isinstance(n_jobs, int):
raise ValueError('n_jobs must be an integer, or "cuda"')
parallel, p_fun, _ = parallel_func(fft_resample, n_jobs)
y = parallel(p_fun(x_, W, new_len, npad, to_remove, cuda_dict)
for x_ in x)
y = np.array(y)
# Restore the original array shape (modified for resampling)
orig_shape = list(orig_shape)
orig_shape[-1] = y.shape[1]
y.shape = tuple(orig_shape)
else:
warnings.warn('x has zero length along last axis, returning a copy of '
'x')
y = x.copy()
return y
def _weight_data_matrix(data_matrix, window_type, data_transformation=None):
window_length_in_samp = data_matrix[0].shape[0]
window = get_window(window_type, window_length_in_samp, fftbins=False)
if (data_transformation == 'group_delay'):
window *= np.arange(window_length_in_samp)
return data_matrix * window
def welch(x, fs=1.0, window='hanning', nperseg=256, noverlap=None, nfft=None,
detrend='constant', return_onesided=True, scaling='density', axis=-1):
"""
Estimate power spectral density using Welch's method.
Welch's method [1]_ computes an estimate of the power spectral density
by dividing the data into overlapping segments, computing a modified
periodogram for each segment and averaging the periodograms.
Parameters
----------
x : array_like
Time series of measurement values
fs : float, optional
Sampling frequency of the `x` time series in units of Hz. Defaults
to 1.0.
window : str or tuple or array_like, optional
Desired window to use. See `get_window` for a list of windows and
required parameters. If `window` is array_like it will be used
directly as the window and its length will be used for nperseg.
Defaults to 'hanning'.
nperseg : int, optional
Length of each segment. Defaults to 256.
noverlap: int, optional
Number of points to overlap between segments. If None,
``noverlap = nperseg / 2``. Defaults to None.
nfft : int, optional
Length of the FFT used, if a zero padded FFT is desired. If None,
the FFT length is `nperseg`. Defaults to None.
detrend : str or function or False, optional
Specifies how to detrend each segment. If `detrend` is a string,
it is passed as the ``type`` argument to `detrend`. If it is a
function, it takes a segment and returns a detrended segment.
If `detrend` is False, no detrending is done. Defaults to 'constant'.
return_onesided : bool, optional
If True, return a one-sided spectrum for real data. If False return
a two-sided spectrum. Note that for complex data, a two-sided
spectrum is always returned.
scaling : { 'density', 'spectrum' }, optional
Selects between computing the power spectral density ('density')
where Pxx has units of V**2/Hz if x is measured in V and computing
the power spectrum ('spectrum') where Pxx has units of V**2 if x is
measured in V. Defaults to 'density'.
axis : int, optional
Axis along which the periodogram is computed; the default is over
the last axis (i.e. ``axis=-1``).
Returns
-------
f : ndarray
Array of sample frequencies.
Pxx : ndarray
Power spectral density or power spectrum of x.
See Also
--------
periodogram: Simple, optionally modified periodogram
lombscargle: Lomb-Scargle periodogram for unevenly sampled data
Notes
-----
An appropriate amount of overlap will depend on the choice of window
and on your requirements. For the default 'hanning' window an
overlap of 50% is a reasonable trade off between accurately estimating
the signal power, while not over counting any of the data. Narrower
windows may require a larger overlap.
If `noverlap` is 0, this method is equivalent to Bartlett's method [2]_.
.. versionadded:: 0.12.0
References
----------
.. [1] P. Welch, "The use of the fast Fourier transform for the
estimation of power spectra: A method based on time averaging
over short, modified periodograms", IEEE Trans. Audio
Electroacoust. vol. 15, pp. 70-73, 1967.
.. [2] M.S. Bartlett, "Periodogram Analysis and Continuous Spectra",
Biometrika, vol. 37, pp. 1-16, 1950.
Examples
--------
>>> from scipy import signal
>>> import matplotlib.pyplot as plt
Generate a test signal, a 2 Vrms sine wave at 1234 Hz, corrupted by
0.001 V**2/Hz of white noise sampled at 10 kHz.
>>> fs = 10e3
>>> N = 1e5
>>> amp = 2*np.sqrt(2)
>>> freq = 1234.0
>>> noise_power = 0.001 * fs / 2
>>> time = np.arange(N) / fs
>>> x = amp*np.sin(2*np.pi*freq*time)
>>> x += np.random.normal(scale=np.sqrt(noise_power), size=time.shape)
Compute and plot the power spectral density.
>>> f, Pxx_den = signal.welch(x, fs, nperseg=1024)
>>> plt.semilogy(f, Pxx_den)
>>> plt.ylim([0.5e-3, 1])
>>> plt.xlabel('frequency [Hz]')
>>> plt.ylabel('PSD [V**2/Hz]')
>>> plt.show()
If we average the last half of the spectral density, to exclude the
peak, we can recover the noise power on the signal.
>>> np.mean(Pxx_den[256:])
0.0009924865443739191
Now compute and plot the power spectrum.
>>> f, Pxx_spec = signal.welch(x, fs, 'flattop', 1024, scaling='spectrum')
>>> plt.figure()
>>> plt.semilogy(f, np.sqrt(Pxx_spec))
>>> plt.xlabel('frequency [Hz]')
>>> plt.ylabel('Linear spectrum [V RMS]')
>>> plt.show()
The peak height in the power spectrum is an estimate of the RMS amplitude.
>>> np.sqrt(Pxx_spec.max())
2.0077340678640727
"""
x = np.asarray(x)
if x.size == 0:
return np.empty(x.shape), np.empty(x.shape)
if axis != -1:
x = np.rollaxis(x, axis, len(x.shape))
if x.shape[-1] < nperseg:
warnings.warn('nperseg = %d, is greater than x.shape[%d] = %d, using '
'nperseg = x.shape[%d]'
% (nperseg, axis, x.shape[axis], axis))
nperseg = x.shape[-1]
if isinstance(window, string_types) or type(window) is tuple:
win = get_window(window, nperseg)
else:
win = np.asarray(window)
if len(win.shape) != 1:
raise ValueError('window must be 1-D')
if win.shape[0] > x.shape[-1]:
raise ValueError('window is longer than x.')
nperseg = win.shape[0]
# numpy 1.5.1 doesn't have result_type.
outdtype = (np.array([x[0]]) * np.array([1], 'f')).dtype.char.lower()
if win.dtype != outdtype:
win = win.astype(outdtype)
if scaling == 'density':
scale = 1.0 / (fs * (win*win).sum())
elif scaling == 'spectrum':
scale = 1.0 / win.sum()**2
else:
raise ValueError('Unknown scaling: %r' % scaling)
if noverlap is None:
noverlap = nperseg // 2
elif noverlap >= nperseg:
raise ValueError('noverlap must be less than nperseg.')
if nfft is None:
nfft = nperseg
elif nfft < nperseg:
raise ValueError('nfft must be greater than or equal to nperseg.')
if not detrend:
detrend_func = lambda seg: seg
elif not hasattr(detrend, '__call__'):
detrend_func = lambda seg: signaltools.detrend(seg, type=detrend)
elif axis != -1:
# Wrap this function so that it receives a shape that it could
# reasonably expect to receive.
def detrend_func(seg):
seg = np.rollaxis(seg, -1, axis)
seg = detrend(seg)
return np.rollaxis(seg, axis, len(seg.shape))
else:
detrend_func = detrend
step = nperseg - noverlap
indices = np.arange(0, x.shape[-1]-nperseg+1, step)
if np.isrealobj(x) and return_onesided:
outshape = list(x.shape)
if nfft % 2 == 0: # even
outshape[-1] = nfft // 2 + 1
Pxx = np.empty(outshape, outdtype)
for k, ind in enumerate(indices):
x_dt = detrend_func(x[..., ind:ind+nperseg])
xft = my_rfft(x_dt*win, nfft)
#xft = fftpack.rfft(x_dt*win, nfft)
# fftpack.rfft returns the positive frequency part of the fft
# as real values, packed r r i r i r i ...
# this indexing is to extract the matching real and imaginary
# parts, while also handling the pure real zero and nyquist
# frequencies.
if k == 0:
Pxx[..., (0,-1)] = xft[..., (0,-1)]**2
Pxx[..., 1:-1] = xft[..., 1:-1:2]**2 + xft[..., 2::2]**2
else:
Pxx *= k/(k+1.0)
Pxx[..., (0,-1)] += xft[..., (0,-1)]**2 / (k+1.0)
Pxx[..., 1:-1] += (xft[..., 1:-1:2]**2 + xft[..., 2::2]**2) \
/ (k+1.0)
else: # odd
outshape[-1] = (nfft+1) // 2
Pxx = np.empty(outshape, outdtype)
for k, ind in enumerate(indices):
x_dt = detrend_func(x[..., ind:ind+nperseg])
xft = my_rfft(x_dt*win, nfft)
#xft = fftpack.rfft(x_dt*win, nfft)
#print (nfft, xft.shape, xft_2.shape)
if k == 0:
Pxx[..., 0] = xft[..., 0]**2
Pxx[..., 1:] = xft[..., 1::2]**2 + xft[..., 2::2]**2
else:
Pxx *= k/(k+1.0)
Pxx[..., 0] += xft[..., 0]**2 / (k+1)
Pxx[..., 1:] += (xft[..., 1::2]**2 + xft[..., 2::2]**2) \
/ (k+1.0)
Pxx[..., 1:-1] *= 2*scale
Pxx[..., (0,-1)] *= scale
f = np.arange(Pxx.shape[-1]) * (fs/nfft)
else:
for k, ind in enumerate(indices):
x_dt = detrend_func(x[..., ind:ind+nperseg])
xft = my_rfft(x_dt*win, nfft)
#xft = fftpack.rfft(x_dt*win, nfft)
if k == 0:
Pxx = (xft * xft.conj()).real
else:
Pxx *= k/(k+1.0)
Pxx += (xft * xft.conj()).real / (k+1.0)
Pxx *= scale
f = fftpack.fftfreq(nfft, 1.0/fs)
if axis != -1:
Pxx = np.rollaxis(Pxx, -1, axis)
return f, Pxx
def test_kaiser_float(self):
win1 = windows.get_window(7.2, 64)
win2 = windows.kaiser(64, 7.2, False)
assert_allclose(win1, win2)
#
window_data = {}
figs = {}
figs_response = {}
#
# Generate window data
#
for window in window_names:
window_name = str(window)
try:
# Generate signal data
window_data[window_name] = windows.get_window(window, Nx)
except ValueError as e:
# Call the function itself if it exists
if window in windows.__all__:
window_data[window_name] = getattr(windows, window)(Nx)
elif type(window) is tuple and window[0] in windows.__all__:
window_data[window_name] = getattr(windows, window[0])(Nx,
*window[1:])
else:
print(f"{window} failed: {e}")
continue
# Generate frequency response
def _spectral_helper(x, y, fs=1.0, window='hann', nperseg=256,
noverlap=None, nfft=None, detrend='constant',
return_onesided=True, scaling='spectrum', axis=-1,
mode='psd'):
"""
Calculate various forms of windowed FFTs for PSD, CSD, etc.
This is a helper function that implements the commonality between the
psd, csd, and spectrogram functions. It is not designed to be called
externally. The windows are not averaged over; the result from each window
is returned.
Parameters
---------
x : array_like
Array or sequence containing the data to be analyzed.
y : array_like
Array or sequence containing the data to be analyzed. If this is
the same object in memoery as x (i.e. _spectral_helper(x, x, ...)),
the extra computations are spared.
fs : float, optional
Sampling frequency of the time series. Defaults to 1.0.
window : str or tuple or array_like, optional
Desired window to use. See `get_window` for a list of windows and
required parameters. If `window` is array_like it will be used
directly as the window and its length will be used for nperseg.
Defaults to 'hann'.
nperseg : int, optional
Length of each segment. Defaults to 256.
noverlap : int, optional
Number of points to overlap between segments. If None,
``noverlap = nperseg // 2``. Defaults to None.
nfft : int, optional
Length of the FFT used, if a zero padded FFT is desired. If None,
the FFT length is `nperseg`. Defaults to None.
detrend : str or function or False, optional
Specifies how to detrend each segment. If `detrend` is a string,
it is passed as the ``type`` argument to `detrend`. If it is a
function, it takes a segment and returns a detrended segment.
If `detrend` is False, no detrending is done. Defaults to 'constant'.
return_onesided : bool, optional
If True, return a one-sided spectrum for real data. If False return
a two-sided spectrum. Note that for complex data, a two-sided
spectrum is always returned.
scaling : { 'density', 'spectrum' }, optional
Selects between computing the cross spectral density ('density')
where `Pxy` has units of V**2/Hz and computing the cross spectrum
('spectrum') where `Pxy` has units of V**2, if `x` and `y` are
measured in V and fs is measured in Hz. Defaults to 'density'
axis : int, optional
Axis along which the periodogram is computed; the default is over
the last axis (i.e. ``axis=-1``).
mode : str, optional
Defines what kind of return values are expected. Options are ['psd',
'complex', 'magnitude', 'angle', 'phase'].
Returns
-------
freqs : ndarray
Array of sample frequencies.
t : ndarray
Array of times corresponding to each data segment
result : ndarray
Array of output data, contents dependent on *mode* kwarg.
References
----------
.. [1] Stack Overflow, "Rolling window for 1D arrays in Numpy?",
http://stackoverflow.com/a/6811241
.. [2] Stack Overflow, "Using strides for an efficient moving average
filter", http://stackoverflow.com/a/4947453
Notes
-----
Adapted from matplotlib.mlab
.. versionadded:: 0.16.0
"""
from scipy import fftpack
from scipy.signal import signaltools
from scipy.signal.windows import get_window
if mode not in ['psd', 'complex', 'magnitude', 'angle', 'phase']:
raise ValueError("Unknown value for mode %s, must be one of: "
"'default', 'psd', 'complex', "
"'magnitude', 'angle', 'phase'" % mode)
# If x and y are the same object we can save ourselves some computation.
same_data = y is x
if not same_data and mode != 'psd':
raise ValueError("x and y must be equal if mode is not 'psd'")
axis = int(axis)
# Ensure we have np.arrays, get outdtype
x = np.asarray(x)
if not same_data:
y = np.asarray(y)
outdtype = np.result_type(x,y,np.complex64)
else:
outdtype = np.result_type(x,np.complex64)
if not same_data:
# Check if we can broadcast the outer axes together
xouter = list(x.shape)
youter = list(y.shape)
xouter.pop(axis)
youter.pop(axis)
try:
outershape = np.broadcast(np.empty(xouter), np.empty(youter)).shape
except ValueError:
raise ValueError('x and y cannot be broadcast together.')
if same_data:
if x.size == 0:
return np.empty(x.shape), np.empty(x.shape), np.empty(x.shape)
else:
if x.size == 0 or y.size == 0:
outshape = outershape + (min([x.shape[axis], y.shape[axis]]),)
emptyout = np.rollaxis(np.empty(outshape), -1, axis)
return emptyout, emptyout, emptyout
if x.ndim > 1:
if axis != -1:
x = np.rollaxis(x, axis, len(x.shape))
if not same_data and y.ndim > 1:
y = np.rollaxis(y, axis, len(y.shape))
# Check if x and y are the same length, zero-pad if necessary
if not same_data:
if x.shape[-1] != y.shape[-1]:
if x.shape[-1] < y.shape[-1]:
pad_shape = list(x.shape)
pad_shape[-1] = y.shape[-1] - x.shape[-1]
x = np.concatenate((x, np.zeros(pad_shape)), -1)
else:
pad_shape = list(y.shape)
pad_shape[-1] = x.shape[-1] - y.shape[-1]
y = np.concatenate((y, np.zeros(pad_shape)), -1)
# X and Y are same length now, can test nperseg with either
if x.shape[-1] < nperseg:
warnings.warn('nperseg = {0:d}, is greater than input length = {1:d}, '
'using nperseg = {1:d}'.format(nperseg, x.shape[-1]))
nperseg = x.shape[-1]
nperseg = int(nperseg)
if nperseg < 1:
raise ValueError('nperseg must be a positive integer')
if nfft is None:
nfft = nperseg
elif nfft < nperseg:
raise ValueError('nfft must be greater than or equal to nperseg.')
else:
nfft = int(nfft)
if noverlap is None:
noverlap = nperseg//2
elif noverlap >= nperseg:
raise ValueError('noverlap must be less than nperseg.')
else:
noverlap = int(noverlap)
# Handle detrending and window functions
if not detrend:
def detrend_func(d):
return d
elif not hasattr(detrend, '__call__'):
def detrend_func(d):
return signaltools.detrend(d, type=detrend, axis=-1)
elif axis != -1:
# Wrap this function so that it receives a shape that it could
# reasonably expect to receive.
def detrend_func(d):
d = np.rollaxis(d, -1, axis)
d = detrend(d)
return np.rollaxis(d, axis, len(d.shape))
else:
detrend_func = detrend
if isinstance(window, string_types) or type(window) is tuple:
win = get_window(window, nperseg)
else:
win = np.asarray(window)
if len(win.shape) != 1:
raise ValueError('window must be 1-D')
if win.shape[0] != nperseg:
raise ValueError('window must have length of nperseg')
if np.result_type(win,np.complex64) != outdtype:
win = win.astype(outdtype)
if mode == 'psd':
if scaling == 'density':
scale = 1.0 / (fs * (win*win).sum())
elif scaling == 'spectrum':
scale = 1.0 / win.sum()**2
else:
raise ValueError('Unknown scaling: %r' % scaling)
else:
scale = 1
if return_onesided is True:
if np.iscomplexobj(x):
sides = 'twosided'
else:
sides = 'onesided'
if not same_data:
if np.iscomplexobj(y):
sides = 'twosided'
else:
sides = 'twosided'
if sides == 'twosided':
num_freqs = nfft
elif sides == 'onesided':
if nfft % 2:
num_freqs = (nfft + 1)//2
else:
num_freqs = nfft//2 + 1
# Perform the windowed FFTs
result = _fft_helper(x, win, detrend_func, nperseg, noverlap, nfft)
result = result[..., :num_freqs]
freqs = fftpack.fftfreq(nfft, 1/fs)[:num_freqs]
if not same_data:
# All the same operations on the y data
result_y = _fft_helper(y, win, detrend_func, nperseg, noverlap, nfft)
result_y = result_y[..., :num_freqs]
result = np.conjugate(result) * result_y
elif mode == 'psd':
result = np.conjugate(result) * result
elif mode == 'magnitude':
result = np.absolute(result)
elif mode == 'angle' or mode == 'phase':
result = np.angle(result)
elif mode == 'complex':
pass
result *= scale
if sides == 'onesided':
if nfft % 2:
result[...,1:] *= 2
else:
# Last point is unpaired Nyquist freq point, don't double
result[...,1:-1] *= 2
t = np.arange(nperseg/2, x.shape[-1] - nperseg/2 + 1, nperseg - noverlap)/float(fs)
if sides != 'twosided' and not nfft % 2:
# get the last value correctly, it is negative otherwise
freqs[-1] *= -1
# we unwrap the phase here to handle the onesided vs. twosided case
if mode == 'phase':
result = np.unwrap(result, axis=-1)
result = result.astype(outdtype)
# All imaginary parts are zero anyways
if same_data and mode != 'complex':
result = result.real
# Output is going to have new last axis for window index
if axis != -1:
# Specify as positive axis index
if axis < 0:
axis = len(result.shape)-1-axis
# Roll frequency axis back to axis where the data came from
result = np.rollaxis(result, -1, axis)
else:
# Make sure window/time index is last axis
result = np.rollaxis(result, -1, -2)
return freqs, t, result
def tocovdata(self, timeseries):
'''
Return auto covariance function from data.
Return
-------
R : CovData1D object
with attributes:
data : ACF vector length L+1
args : time lags length L+1
sigma : estimated large lag standard deviation of the estimate
assuming x is a Gaussian process:
if R(k)=0 for all lags k>q then an approximation
of the variance for large samples due to Bartlett
var(R(k))=1/N*(R(0)^2+2*R(1)^2+2*R(2)^2+ ..+2*R(q)^2)
for k>q and where N=length(x). Special case is
white noise where it equals R(0)^2/N for k>0
norm : bool
If false indicating that R is not normalized
Example:
--------
>>> import wafo.data
>>> import wafo.objects as wo
>>> x = wafo.data.sea()
>>> ts = wo.mat2timeseries(x)
>>> acf = ts.tocovdata(150)
>>> h = acf.plot()
'''
lag = self.lag
window = self.window
detrend = self.detrend
try:
x = timeseries.data.flatten('F')
dt = timeseries.sampling_period()
except Exception:
x = timeseries[:, 1:].flatten('F')
dt = sampling_period(timeseries[:, 0])
if not (self.dt is None):
dt = self.dt
if not (self.tr is None):
x = self.tr.dat2gauss(x)
n = len(x)
indnan = np.isnan(x)
if any(indnan):
x = x - x[1 - indnan].mean()
Ncens = n - indnan.sum()
x[indnan] = 0.
else:
Ncens = n
x = x - x.mean()
if hasattr(detrend, '__call__'):
x = detrend(x)
nfft = 2 ** nextpow2(n)
Rper = abs(fft(x, nfft)) ** 2 / Ncens # Raw periodogram
R = np.real(fft(Rper)) / nfft # ifft = fft/nfft since Rper is real!
if self.flag.startswith('unbiased'):
# unbiased result, i.e. divide by n-abs(lag)
R = R[:Ncens] * Ncens / np.arange(Ncens, 1, -1)
if self.norm:
R = R / R[0]
if lag is None:
lag = self._estimate_lag(R, Ncens)
lag = min(lag, n - 2)
if isinstance(window, str) or type(window) is tuple:
win = get_window(window, 2 * lag - 1)
else:
win = np.asarray(window)
R[:lag] = R[:lag] * win[lag - 1::]
R[lag] = 0
lags = slice(0, lag + 1)
t = np.linspace(0, lag * dt, lag + 1)
acf = CovData1D(R[lags], t)
acf.sigma = np.sqrt(np.r_[0, R[0] ** 2,
R[0] ** 2 + 2 * np.cumsum(R[1:] ** 2)] / Ncens)
acf.children = [PlotData(-2. * acf.sigma[lags], t),
PlotData(2. * acf.sigma[lags], t)]
acf.plot_args_children = ['r:']
acf.norm = self.norm
return acf
