def _stft(self):
if not self._have_x:
print "Error: You need to load a sound file first: use self.load_audio('filename.wav')"
return False
fp = self._check_feature_params()
num_frames = len(self.x)
self.STFT = P.zeros((self.nfft/2+1, num_frames), dtype='complex')
self.win = P.ones(self.wfft) if self.window=='rect' else P.np.sqrt(P.hanning(self.wfft))
x = P.zeros(self.wfft)
buf_frames = 0
for k, nex in enumerate(self.x):
x = self._shift_insert(x, nex, self.nhop)
if self.nhop >= self.wfft - k*self.nhop : # align buffer on start of audio
self.STFT[:,k-buf_frames]=P.rfft(self.win*x, self.nfft).T
else:
buf_frames+=1
self.STFT = self.STFT / self.nfft
self._fftfrqs = P.arange(0,self.nfft/2+1) * self.sample_rate/float(self.nfft)
self._have_stft=True
if self.verbosity:
print "Extracted STFT: nfft=%d, hop=%d" %(self.nfft, self.nhop)
self.inverse=self._istftm
self.X = abs(self.STFT)
if not self.magnitude:
self.X = self.X**2
return True
def _stft(self):
if not self._have_x:
print(
"Error: You need to load a sound file first: use self.load_audio('filename.wav')"
)
return False
fp = self._check_feature_params()
num_frames = len(self.x)
self.STFT = P.zeros((int(self.nfft / 2 + 1), num_frames),
dtype='complex')
self.win = P.ones(self.wfft) if self.window == 'rect' else P.np.sqrt(
P.hanning(self.wfft))
x = P.zeros(self.wfft)
buf_frames = 0
for k, nex in enumerate(self.x):
x = self._shift_insert(x, nex, self.nhop)
# align buffer on start of audio
if self.nhop >= self.wfft - k * self.nhop:
self.STFT[:, k - buf_frames] = P.rfft(self.win * x,
self.nfft).T
else:
buf_frames += 1
self.STFT = self.STFT / self.nfft
self._fftfrqs = P.arange(
0, self.nfft / 2 + 1) * self.sample_rate / float(self.nfft)
self._have_stft = True
if self.verbosity:
print("Extracted STFT: nfft=%d, hop=%d" % (self.nfft, self.nhop))
self.inverse = self._istftm
self.X = abs(self.STFT)
if not self.magnitude:
self.X = self.X**2
return True
def plot_signal(s,sr,start=0,end=440,N=32768):
'''plots the waveform and spectrum of a signal s
with sampling rate sr, from a start to an
end position (waveform), and from a start
position with a N-point DFT (spectrum)'''
pl.figure(figsize=(8,5))
pl.subplot(211)
sig = s[start:end]
time = pl.arange(0,len(sig))/sr
pl.plot(time,sig, 'k-')
pl.ylim(-1.1,1.1)
pl.xlabel("time (s)")
pl.subplot(212)
N = 32768
win = pl.hanning(N)
scal = N*pl.sqrt(pl.mean(win**2))
sig = s[start:start+N]
window = pl.rfft(sig*win/max(sig))
f = pl.arange(0,len(window))
bins = f*sr/N
mags = abs(window/scal)
spec = 20*pl.log10(mags/max(mags))
pl.plot(bins,spec, 'k-')
pl.ylim(-60, 1)
pl.ylabel("amp (dB)", size=16)
pl.xlabel("freq (Hz)", size=16)
pl.yticks()
pl.xticks()
pl.xlim(0,sr/2)
pl.tight_layout()
pl.show()
def spec_env(frame, coefs):
mags = abs(frame)
N = len(mags)
ceps = pl.rfft(pl.log(mags[:N - 1]))
ceps[coefs:] = 0
mags[:N - 1] = pl.exp(pl.irfft(ceps))
return mags
def pconv(input, ir, S=1024):
N = len(ir)
L = len(input)
M = S * 2
P = int(pl.ceil(N / S))
H = pl.zeros((P, M // 2 + 1)) + 0j
X = pl.zeros((P, M // 2 + 1)) + 0j
x = pl.zeros(M)
o = pl.zeros(M)
s = pl.zeros(S)
for i in range(0, P):
p = (P - 1 - i) * S
ll = len(ir[p:p + S])
x[:ll] = ir[p:p + S]
H[i] = pl.rfft(x)
y = pl.zeros(L + N - 1)
n = S
i = 0
for p in range(0, L + N - S, S):
if p > L:
x = pl.zeros(M)
else:
if n + p > L:
n = L - p
x[n:] = pl.zeros(M - n)
x[:n] = input[p:p + n]
X[i] = pl.rfft(x)
i = (i + 1) % P
O = pl.zeros(M // 2 + 1) + 0j
for j in range(0, P):
k = (j + i) % P
O += H[j] * X[k]
o = pl.irfft(O)
y[p:p + S] = o[:S] + s
s = o[S:]
return y
def _stft(self):
if not self._have_x:
print "Error: You need to load a sound file first: use self.load_audio('filename.wav')"
return False
fp = self._check_feature_params()
WX = self._win_mtx(self.x, fp['wfft'], fp['nhop'], fp['nsamples'])
self.STFT=pylab.rfft(WX, fp['nfft']).T
self.STFT/=fp['nfft']
self._have_stft=True
if fp['verbosity']:
print "Extracted STFT: nfft=%d, hop=%d" %(fp['nfft'], fp['nhop'])
return True
def conv(input, ir):
N = len(ir)
L = len(input)
M = 2
while (M <= 2 * N - 1):
M *= 2
h = pl.zeros(M)
x = pl.zeros(M)
y = pl.zeros(L + N - 1)
h[0:N] = ir
H = pl.rfft(h)
n = N
for p in range(0, L, N):
if p + n > L:
n = L - p
x[n:] = pl.zeros(M - n)
x[0:n] = input[p:p + n]
y[p:p + 2 * n - 1] += pl.irfft(H * pl.rfft(x))[0:2 * n - 1]
return y
def TemporalSpectPlot(self, d,wmin,wmax):
self.fig.clear()
sp = Axes3D(self.fig)
ndav, int_step, cc = self.more_opts_extr()
ll = len(d[0])-1
for i in ll-np.arange(ll):
y=i*np.ones(len(d)/2)
sp.plot(fftfreq(len(d), (d[1,0]-d[0,0])/(ndav/cc))[:len(d)/2], y ,abs(rfft(d[:,i]))[:-1])
sp.set_xlim3d(wmin,wmax)
sp.set_xlabel(r'$\omega/\omega_0$',fontsize=14)
sp.set_ylabel('node index',fontsize=14)
sp.set_zlabel(r'$\langle E_z \rangle_\omega$',fontsize=14)
self.canvas.show()
def run_WN(photoreceptor, I):
Z_dictionary = {}
V_dictionary = {}
Vl_dictionary = {}
I_dictionary = {}
for i, V in enumerate(Vr):
DepolarisePhotoreceptor.WithLight(photoreceptor, V)
## Filtering signal
print("Processing V=", V, " mV")
## Run HH
print("Running HH...")
Vm, g = Experiment.inject_current(photoreceptor, I, dt)
Vl = Linearise.inject_current(photoreceptor, I, dt)
print("Variance of voltage is ", var(Vm), "mV2")
## Calculate impedance
Vmm = (Vm - mean(Vm)) #without DC
CPSD = zeros(round(len(time) / 2 + 1), dtype=complex_)
PSD = zeros(round(len(time) / 2 + 1))
window = hamming(len(time))
print("len(window)", len(window))
for ii in range(repetitions):
I_cut = I[ii * len(time):(ii + 1) * len(time)]
V_cut = Vmm[ii * len(time):(ii + 1) * len(time)]
PSD += power(absolute(rfft(I_cut * window)), 2)
CPSD += rfft(V_cut * window) * rfft(I_cut * window).conjugate()
Z = CPSD / PSD #MOhm
Z_dictionary[V] = Z
I_dictionary[V] = I[sample * len(time):(sample + 1) * len(time)]
V_dictionary[V] = Vm[sample * len(time):(sample + 1) * len(time)]
Vl_dictionary[V] = Vl[sample * len(time):(sample + 1) * len(time)]
return (Z_dictionary, V_dictionary, Vl_dictionary, I_dictionary)
def true_spec_env(frame, coefs, thresh):
N = len(frame)
form = pl.zeros(N)
lmags = pl.log(abs(frame[:N - 1]))
mags = lmags
go = True
while (go):
ceps = pl.rfft(lmags)
ceps[coefs:] = 0
form[:N - 1] = pl.irfft(ceps)
go = False
for i in range(0, N - 1):
if lmags[i] < form[i]: lmags[i] = form[i]
diff = mags[i] - form[i]
if diff > thresh: go = True
return pl.exp(form)
def true_spec_env(amps,coefs,thresh):
N = len(amps)
sm = 10e-15
form = pl.zeros(N)
lmags = pl.log(amps[:N-1]+sm)
mags = lmags
check = True
while(check):
ceps = pl.rfft(lmags)
ceps[coefs:] = 0
form[:N-1] = pl.irfft(ceps)
for i in range(0,N-1):
if lmags[i] < form[i]: lmags[i] = form[i]
diff = mags[i] - form[i]
if diff > thresh: check = True
else: check = False
return pl.exp(form)+sm
def fft_filter(time, signal, result, window_func=pl.hanning, \
samp_freq=8, freq_limit = (0.0033, 0.04)):
"""
Applies an Band Pass filter to signal in the frequency domain and returns
filterd PSD.
In:
time : ndarray, relative R occurance times vector
signal : ndarray, HRV vector
result : dict, current fragment info (see process function)
samp_freq : int, signal sample frequency
freq_limit : tuple, (min, max) frequencies for band pass filter
Out:
freq_filt_log : ndarray, frequency in logarithm scale
spec_filt_log : ndarray, PSD in logarithm scale
"""
window = window_func(len(signal)) # window
signal_wed = signal * window
n = len(signal)/2
spec = (pl.absolute(pl.rfft(signal_wed))[:-1]/(n))**2 # entering to frequency domain
freq = pl.fftfreq(len(signal), 1.0/samp_freq)[:n] # get freqs axis values
spec_filt = pl.zeros(spec.shape, dtype='float')
freq_filt = pl.zeros(freq.shape, dtype='float') # same for filtered
for i in range(n): # filter by frequency
if pl.logical_and(freq_limit[0] <= abs(freq[i]), abs(freq[i]) <= freq_limit[1]):
freq_filt[i] = freq[i]
spec_filt[i] = spec[i]
else:
freq_filt[i] = 0 # fill invalid frequencies with 0 value
spec_filt[i] = 0
spec_filt = pl.ma.masked_equal(spec_filt,0) # cutt off invalid values
spec_filt = pl.ma.compressed(spec_filt)
freq_filt = pl.ma.masked_equal(freq_filt,0)
freq_filt = pl.ma.compressed(freq_filt)
spec_filt_log = pl.log10(spec_filt) # for output
freq_filt_log = pl.log10(freq_filt)
return freq_filt_log, spec_filt_log
def SpecEvolPlot(self, d,windwidth,step,wmin,wmax, rowind):
dd=np.array([])
ndav, int_step, cc = self.more_opts_extr()
Tim=np.arange(windwidth, len(d)-windwidth,step)
for n in Tim:
dd=np.append(dd, np.abs(rfft(d[n-windwidth:n+windwidth,rowind]))[:-1])
Freq=fftfreq(2*windwidth, (d[1,0]-d[0,0])/(ndav/cc))[:windwidth]
if wmax!=0:
nFrqMax= int(wmax/(Freq[1]-Freq[0]))
else:
nFrqMax=len (Freq)
nFrqMin= int(wmin/(Freq[1]-Freq[0]))
dd=dd.reshape(len(Tim),len(Freq))[:,nFrqMin:nFrqMax].transpose()
self.fig.clear()
sp = self.fig.add_subplot(111)
pplt = sp.imshow(dd, origin='lower',aspect='auto',extent=[d[Tim[0],0]/(ndav/cc), d[Tim[-1],0]/(ndav/cc),Freq[nFrqMin],Freq[nFrqMax]])
sp.set_xlabel(r'$t/\tau_0$',fontsize=14)
sp.set_ylabel(r'$\omega/\omega_0$',fontsize=14)
self.fig.colorbar(pplt)
self.canvas.show()
return
#! /usr/bin/env python
from pylab import figure, loadtxt, arange, ones, fftfreq, rfft
from mpl_toolkits.mplot3d import Axes3D
fig = figure()
ax = Axes3D(fig)
inp = raw_input("input xfel_xx number xx and node index:").split(' ')
xx = inp[0]
node = inp[1]
d = loadtxt('../xfel_' + xx + '/etc/fld_00' + node + '_000')
ll = len(d[0]) - 1
for i in ll - arange(ll):
y = i * ones(len(d) / 2)
ax.plot(fftfreq(len(d), 0.04)[:len(d) / 2], y, abs(rfft(d[:, i]))[:-1])
ax.set_xlim3d(0.9, 1.1)
fig.show()
fig.show()
raw_input("Press Enter to continue...")
import pylab as pl pi = pl.pi T = 1000 t = pl.arange(0, T) x = 0.51 - 0.49 * pl.cos(2 * pi * t / T) y = pl.sin(2 * pi * 10 * t / T) X = pl.rfft(y * x) pl.plot(pl.irfft(X)) pl.show()
import pylab as pl T = 64 t = pl.arange(0, T) env = pl.zeros(T) pl.figure(figsize=(8, 3)) sig = pl.sin(2 * pl.pi * (T / 4) * t / T) * pl.hanning(T) pl.ylim(0, 1.1) pl.xlim(0, T // 2) pl.xticks([x for x in range(0, T // 2 + 1, 4)]) pl.stem(t[0:T // 2], abs(pl.rfft(sig) / (T / 4))[0:T // 2], 'k-', linewidth=1) pl.tight_layout() pl.show()
def speriodogram(x, NFFT=None, detrend=True, sampling=1.,
scale_by_freq=True, window='hamming', axis=0):
"""Simple periodogram, but matrices accepted.
:param x: an array or matrix of data samples.
:param NFFT: length of the data before FFT is computed (zero padding)
:param bool detrend: detrend the data before co,puteing the FFT
:param float sampling: sampling frequency of the input :attr:`data`.
:param scale_by_freq:
:param str window:
:return: 2-sided PSD if complex data, 1-sided if real.
if a matrix is provided (using numpy.matrix), then a periodogram
is computed for each row. The returned matrix has the same shape as the input
matrix.
The mean of the input data is also removed from the data before computing
the psd.
.. plot::
:width: 80%
:include-source:
from pylab import *
from spectrum import *
data = data_cosine(N=1024, A=0.1, sampling=1024, freq=200)
semilogy(speriodogram(data, detrend=False, sampling=1024), marker='o')
grid(True)
.. plot::
:width: 80%
:include-source:
from spectrum import *
from pylab import *
# create N data sets and make the frequency dependent on the time
N = 100
m = numpy.concatenate([data_cosine(N=1024, A=0.1, sampling=1024, freq=x) for x in range(1, N)]);
m.resize(N, 1024)
res = speriodogram(m)
figure(1)
semilogy(res)
figure(2)
imshow(res.transpose(), aspect='auto')
.. todo:: a proper spectrogram class/function that takes care of normalisation
"""
x = array(x)
# array with 1 dimension case
if x.ndim == 1:
axis = 0
r = x.shape[0]
w = Window(r, window) #same size as input data
# matrix case
elif x.ndim == 2:
print('2D array. each row is a 1D array')
[r, c] = x.shape
if r==1:
r = c
w = Window(x.size, window) #same size as input data
else:
w = Window(c, window) #same size as input data
if NFFT is None:
NFFT = len(x)
isreal = numpy.isrealobj(x)
if detrend == True:
m = mean(x, axis=axis)
else:
m = 0
if isreal == True:
res = (abs (rfft (x*w.data - m, NFFT))) ** 2. / r
else:
res = (abs (fft (x*w.data - m, NFFT ))) ** 2. / r
if scale_by_freq is True:
df = sampling / float(NFFT)
res*= 2*pi/df
return res.transpose()
def stft(x, w, n):
N = len(w)
X = pl.rfft(w * x)
k = pl.arange(0, N / 2 + 1)
return X * pl.exp(2 * pl.pi * 1j * k * n / N)
sampling_frequency = wf.GetSamplingFrequency()
sampling_period = wf.GetSamplingPeriod()
if wf3count == 0 and channel == OPPI3ChannelNumber:
avgwf3.MakeSimilarTo(rawwf)
if wf4count == 0 and channel == OPPI4ChannelNumber:
avgwf4.MakeSimilarTo(rawwf)
if goodOPPI3Waveform:
rawwf.MakeSimilarTo(avgwf3)
avgwf3 += rawwf
# Sample baseline from baseline window
a3base = numpy.ndarray(
int(baselineAverageTime / sampling_period), dtype="float", buffer=rawwf.GetData()
)
a3base = a3base / a3base.max()
if wf3count == 0:
a3bFFT = pylab.rfft(a3base)
else:
a3bFFT += pylab.rfft(a3base)
print a3bFFT * pylab.conj(a3bFFT)
raw_input("Press Enter to continue")
pylab.subplot(2, 1, 1)
X3 = numpy.arange(0, numpy.size(a3bFFT))
pylab.plot(X3, sampling_period * sampling_period * a3bFFT * pylab.conj(a3bFFT), label="OPPI3")
pylab.subplot(2, 1, 2)
pylab.psd(a3base, Fs=100e6)
fig.canvas.draw()
# pylab.plot(X3,a3bFFT,label='OPPI3')
# print "OPPI3 ", eventTime , avgwf3.GetLength(), a3.size , a3
wf3count += 1
if goodOPPI4Waveform:
rawwf.MakeSimilarTo(avgwf4)
import pylab as pl
import numpy as np
import seaborn as sns
WD721 = np.loadtxt("/home/russell/Desktop/matplotlib/WaveData721.csv", skiprows = 3, delimiter = ',')
pl.figure()
pl.clf()
fx2 = np.arange(0, 2001)
fx2 = fx2/0.02 #scaling factor
sns.set_context("notebook", rc={"lines.linewidth": 1.5})
sns.set_style("darkgrid")
sns.set_palette("husl")
wave = WD721[0:4000, 1] #The wave off the osciliscope
pl.subplot(2,2,1)
pl.title("Wave")
pl.plot(wave) #Plotting The Wave
pl.xlim([1050, 1600])
pl.subplot(2,2,2)
pl.title("Square of wave")
pl.plot(wave**2) #Plotting the square of The Wave
pl.xlim([1050, 1600])
pl.subplot(2,2,3)
pl.title("Spectrum of wave")
pl.plot(fx2, abs(pl.rfft(wave))) #Plotting the Fourier transform of The Wave
pl.xlim([36000, 44000])
pl.xlabel("frequency")
pl.subplot(2,2,4)
pl.title("Spectrum of wave^2")
pl.plot(fx2, abs(pl.rfft(wave**2)))
pl.xlim([-100, 4500])
pl.xlabel("frequency")
pl.tight_layout()
pl.figure(figsize=(8, 5))
pl.subplot(211)
pl.title('waveform')
end = N
t = pl.arange(0, end) / float(sr)
pl.xlim(0, t[end - 1])
pl.ylim(-1.1, 1.1)
pl.xlabel("time (s)")
pl.plot(t, sig[0:end], 'k')
pl.subplot(212)
pl.title('spectral envelope')
x = pl.arange(0, N / 2)
bins = x * sr / N
spec = pl.fft(sig / max(sig))
mags = abs(spec / N)
logm = pl.log(mags)
ceps = pl.rfft(logm)
ceps2 = pl.zeros(len(ceps))
ceps2[0:50] = ceps[0:50]
specenv = pl.exp(pl.irfft(ceps2))
spec1 = 20 * pl.log10(mags / max(mags))
pl.ylim(-40, 1)
pl.ylabel("amp (dB)")
pl.xlabel("freq (Hz)")
pl.xlim(0, 2000)
pl.tight_layout()
pl.plot(bins[0:N / 2], spec1[0:N / 2], 'k-', linewidth=2)
pl.show()
def stft(x, w):
X = pl.rfft(x * w)
return X
#! /usr/bin/env python
from pylab import figure, loadtxt, arange, ones,fftfreq, rfft
from mpl_toolkits.mplot3d import Axes3D
fig = figure()
ax = Axes3D(fig)
inp = raw_input("input xfel_xx number xx and node index:").split(' ')
xx = inp[0]
node=inp[1]
d = loadtxt('../xfel_'+xx+'/etc/fld_00'+node+'_000')
ll = len(d[0])-1
for i in ll-arange(ll):
y=i*ones(len(d)/2)
ax.plot(fftfreq(len(d), 0.04)[:len(d)/2], y ,abs(rfft(d[:,i]))[:-1])
ax.set_xlim3d(0.9,1.1)
fig.show()
fig.show()
raw_input("Press Enter to continue...")
def SpectSpacePlotter(self,f,Res, reg, TIM = None, out_name = None, flag3D='2D', frmt='png'):
Nx, Ny, Lx, Ly, N_nodes = self.OptsExtr()
ndav, int_step, cc = self.more_opts_extr()
actual_time = int(TIM)*int_step/ndav
Nx1, Nx2, Ny1, Ny2, Fmin, Fmax, res, res_x, res_y, clrmap, freq_min, freq_max,freq2_min, freq2_max, ax = Res
res, res_x, res_y, Nx1, Nx2, Ny1, Ny2,freq_max = int(res), int(res_x), int(res_y), float(Nx1),float(Nx2),float(Ny1),float(Ny2), float(freq_max)
Nx1=int(np.rint(Nx1*Nx/Lx));Nx2=int(np.rint(Nx2*Nx/Lx))
Ny1=int(np.rint(Ny1*Ny/Ly));Ny2=int(np.rint(Ny2*Ny/Ly))
dX=Lx/Nx
dY=Ly/Ny
X=np.linspace(0,Lx,Nx)
Y=np.linspace(0,Ly,Ny)
dx, dy = self.StepDef(Res[:-5])
if Fmin == 0.:
Fmin = None
if Fmax == 0.:
Fmax= None
if Nx2==0:
Nx2 = Nx-1
if Ny2==0:
Ny2 = Ny-1
if ax=='x':
Nn1=Nx
Nn2=Ny
Xx=X
Yy=Y
Ll=Ly
ddx=int(np.ceil(dx/2.))
ddy=dy
freq_min, freq_max = freq_min, freq_max
elif ax=='y':
f=f.transpose()
Nn1=np.shape(f)[1]
Nn2=Nx
Xx=Y
Yy=X
Ll=Lx
ddx=int(np.ceil(dy/2.))
ddy=dx
freq_min, freq_max = freq2_min, freq2_max
datFFT = np.zeros((Nn2, Nn1/2+1))
XFFT = abs(fftfreq( Nn1, Xx[1])[:Nn1/2+1])
for j in np.arange(Nn2):
datFFT[j] = abs(rfft(f[j]))
myextent=[XFFT[0],XFFT[-1],Yy[0],Yy[-1]]
self.fig.clear()
if flag3D == '3D':
XX, YY = np.meshgrid(XFFT, Yy)
sp = Axes3D(self.fig, title='time step = '+str(actual_time)+r'$ \tau_0$')
sp.plot_surface(XX,YY,datFFT, cmap = clrmap, linewidth=0)
sp.set_zlim3d(Fmin,Fmax)
if flag3D == '2D':
sp = self.fig.add_subplot(111, title='time step = '+str(actual_time)+r'$ \tau_0$')
if ax=='x':
sp.set_ylabel(r'$y/\lambda_0$',fontsize=14)
sp.set_xlabel(r'$k_x/k_0$',fontsize=14)
elif ax=='y':
sp.set_ylabel(r'$x/\lambda_0$',fontsize=14)
sp.set_xlabel(r'$k_y/k_0$', fontsize=14)
pplt = sp.imshow(datFFT[::ddy,::ddx],extent=myextent, aspect = 'auto', interpolation=intrp,origin='lower',vmin=Fmin, vmax=Fmax, cmap = clrmap)
sp.axis((freq_min,freq_max,0,Ll))
self.fig.colorbar(pplt)
self.canvas.show()
if reg == 'wrt':
self.fig.savefig('./img/'+self.out_name_add+'_'+out_name+'_s'+ax+'_'+TIM+'.'+frmt)
clf()
return
