def modelPulses(**kwargs):
ampl = makeNP(kwargs.get('amplitude', 1))
decay = makeNP(kwargs.get('decay', 0))
t = makeNP(kwargs.get('t', 1))
Fs = kwargs.get('Fs', 1)
t = genTime(maxTime=t, Fs=Fs) if t.size == 1 else t
timeSituations = np.arange(start=kwargs.get('start', t[0]),
stop=kwargs.get('stop', t[-1]),
step=kwargs.get('step', (t[-1] - t[0]) / 20))
timeSamples = closeInVect(t, timeSituations)[0]
envel = expPulses(t, timeSamples, decay, ampl, kwargs)[0] - np.max(ampl)
ams = AMsign(t, kwargs.get('carrier', 0.1 * Fs), envel, depth=1)
signalTupl = pa.awgn(ams[0], SNRdB=makeNP(kwargs.get('SNR', np.inf)))
saveAudio(kwargs, signalTupl[0])
return (signalTupl[0], ) + (timeSamples, t)
def modelModulated(**kwargs):
t = makeNP(kwargs.get('t', 1))
Fs = kwargs.get('Fs', 1)
t = genTime(maxTime=t, Fs=Fs) if t.size == 1 else t
(FMcomp, freq) = frequencyModulated(kwargs.get('FMfreq', 0),
t,
f0=kwargs.get('carrier', 0.1 * Fs),
depth=kwargs.get('FMdepth', 0),
phi0=kwargs.get('phi0', 0),
linear=kwargs.get('linear', False))
ams = AMsign(t,
FMcomp,
kwargs.get('AMfreq', 0),
depth=kwargs.get('AMdepth', 0))
signalTupl = pa.awgn(ams[0], SNRdB=makeNP(kwargs.get('SNR', np.inf)))
return (signalTupl[0], t, freq)
def main():
matplotlib.rc('font', family='Times New Roman', size=12)
Fs = 2000
ampl = 1
decay = 50
t = genTime(maxTime=1, Fs=Fs)
dFmax = 2
freq = np.linspace(start=0, stop=dFmax, num=t.size)
plotGraphs = 0
roughFreq = (50, 200)
kwargs = {
'Fs': Fs,
'SNRvals': np.arange(-2.5, -5.5, -0.5),
'carrier': 100,
'plotGraphs': plotGraphs,
'experiences': 5,
'processes': 5,
'asyncLoop': True
}
kwargs.update({
'roughFreqs': roughFreq,
'formFactor': 128,
'hold': 0,
'secondsNum': 0.025,
'percentOverlap': 75
}) # (64, 128)(0.075, 0.02)
# kwargs.update({'paramNames': ['amplitude', 'decay', 'timeSamples'], 'percentDeviation': [100, 50, 0]})
# kwargs.update({'paramNames': 'amplitude', 'percentDeviation': 100})
kwargs.update({'writeAudio': True, 'hold': 0})
# kwargs.update({'start': 1, 'stop': 2, 't': 5, 'decay': decay, 'SNR': -5, 'fileName': 'aFewPulses.pkl'})
# o = modelPulses(**kwargs)
# (alpha, f, A, theta, res, timeSamplesIdsEst) = pulsesParamsEst(o[0], **kwargs)[0:6]
# modTest(Fs=Fs, t=1, SNRvals=np.arange(4, -16.5, -0.5), fileName='linear02AM0.pkl',
#carrier=100, FMfreq=freq, FMdepth=0.1, AMfreq=5, AMdepth=0.0, plotGraphs=plotGraphs, experiences=102, processes=6, asyncLoop=True) # 6, -12
#modTest(Fs=Fs, t=1, SNRvals=np.arange(-3, -16.5, -0.5), fileName='AMf5d025.pkl',
# carrier=100, FMfreq=5, FMdepth=0.1, AMfreq=5, AMdepth=0.25, plotGraphs=plotGraphs, experiences=1) # 6, -12
#return
pulsesTest(decay=decay,
t=5,
fileName='pulses_5trials__1.4hold.pkl',
**kwargs)
return
modTest(t=1,
fileName='pulses_104trials.pkl',
FMfreq=5,
FMdepth=0.1,
AMfreq=5,
AMdepth=0.25,
**kwargs) # 6, -12
# np.hstack((np.arange(4, -5, -1), np.arange(-5, -7.5, -0.5), np.arange(-8, -22, -2)))
#modTest(t=1, fileName='AMf5d02.pkl', FMfreq=5, FMdepth=0.1, AMfreq=5, AMdepth=0.2, **kwargs) # 6, -12
#modTest(t=1, fileName='AMf3d02.pkl', FMfreq=5, FMdepth=0.1, AMfreq=3, AMdepth=0.2, **kwargs) # 6, -12
return
h0Sam = np.random.normal(loc=0.0, scale=1 / 3,
size=(1, 100)) # np.arange(0.1, 1.1, 0.05)
h1Sam = np.random.normal(loc=1.0, scale=1 / 2.5,
size=(1, 100)) # np.arange(0.9, 2, 0.1)
autoThresholding(h1Sam, h0Sam)
from timeit import default_timer as timer
#modTest(Fs=Fs, t=1, SNRvals=(0,), carrier=100, FMfreq=5, FMdepth=0.1, AMfreq=4, AMdepth=0.25, plotGraphs=plotGraphs, experiences=1) # 6, -12
tStart = timer()
modTest(Fs=Fs,
t=1,
SNRvals=np.arange(4, 1, -1),
fileName='',
carrier=100,
FMfreq=5,
FMdepth=0.1,
AMfreq=5,
AMdepth=0,
plotGraphs=plotGraphs,
experiences=16,
processes=4,
asyncLoop=True) # 6, -12
elapsed2 = timer() - tStart
print(('2 async', elapsed2))
tStart = timer()
modTest(Fs=Fs,
t=1,
SNRvals=np.arange(4, 1, -1),
fileName='',
carrier=100,
FMfreq=5,
FMdepth=0.1,
AMfreq=5,
AMdepth=0,
plotGraphs=plotGraphs,
experiences=16,
processes=4) # 6, -12
elapsed1 = timer() - tStart
print(('2 map', elapsed1))
tStart = timer()
modTest(Fs=Fs,
t=1,
SNRvals=np.arange(4, 1, -1),
fileName='',
carrier=100,
FMfreq=5,
FMdepth=0.1,
AMfreq=5,
AMdepth=0,
plotGraphs=plotGraphs,
experiences=16) # 6, -12
elapsed3 = timer() - tStart
print(('Consequent', elapsed3))
modTest(Fs=Fs,
t=1,
SNRvals=np.hstack((np.arange(4, -5, -1), np.arange(-5, -7.5, -0.5),
np.arange(-8, -16, -2))),
fileName='',
carrier=100,
FMfreq=5,
FMdepth=0.1,
AMfreq=5,
AMdepth=0,
plotGraphs=plotGraphs,
experiences=16) # 6, -12
pulsesTest(Fs=Fs,
decay=decay,
t=5,
SNRvals=np.arange(0, -12, -2),
carrier=100,
plotGraphs=plotGraphs,
experiences=5)
# freq = np.ones((1, t.size))
# freq[0, :] *= 100 * (1 + np.linspace(start=0, stop=dFmax, num=t.size))
# freq = AMsign(t, np.ones((1, t.size))*100, 5, depth=0.1)[0]
(FMcomp, freq) = frequencyModulated(
5, t, f0=100, depth=0.1,
phi0=0) # chirp(t, freq[0, 0], t[-1], freq[0, -1])
FMsignal = pa.awgn(FMcomp, SNRdB=60)[0]
(representation, t0, fVectNew) = pa.DFTbank(FMsignal,
rect=2,
level=0.2,
Fs=Fs,
mirrorLen=0.15,
df=0.05,
freqLims=(50, 200),
formFactor=128)
tEnd = int(closeInVect(t, 0.15)[1])
fInd = int(closeInVect(fVectNew, 100)[1])
fig4 = plotRepresentation(t, representation, fVectNew, None)
fig100 = plotSignal(representation[fInd, :], t, specKind='amplitude')
(alpha, f, A, theta, resid,
coefficient) = pa.thresholdRepreProny(representation,
fVectNew,
Fs=Fs,
percentLength=2,
percentOverlap=75,
lowFreq=85,
highFreq=120,
hold=1.4)
figFull = plotUnder(t, (FMcomp, alpha, (f, freq)),
yLims=[(-1.2, 1.2), (-30, 30), (85, 125)],
labels=(None, 'decay', ('estimated frequency',
'initial frequency')),
secondParam=resid,
secondLim=None) # (0, 2*10 ** -6)
ams = AMsign(t, FMcomp, 5, depth=0.25)
signal = pa.awgn(ams[0], SNRdB=60)[0]
# figAM = plotSignal(signal, t, specKind='amplitude')
(representation, t0, fVectNew) = pa.DFTbank(signal,
rect=2,
level=0.2,
Fs=Fs,
mirrorLen=0.15,
df=1,
freqLims=(50, 200),
formFactor=128)
fig4 = plotRepresentation(t, representation, fVectNew, None)
fInd = int(closeInVect(fVectNew, 105)[1])
fig100 = plotSignal(representation[fInd, :], t, specKind='amplitude')
(alpha, f, A, theta, resid,
coefficient) = pa.thresholdRepreProny(representation,
fVectNew,
Fs=Fs,
percentLength=2,
percentOverlap=75,
lowFreq=80,
highFreq=125,
hold=1.4)
figFull = plotUnder(t, (FMcomp, alpha, (f, freq)),
yLims=[(-1.2, 1.2), (-30, 30), (85, 125)],
secondParam=resid,
secondLim=None) # (0, 2*10 ** -6)
timeSituations = np.arange(start=0, stop=5, step=0.5)
timeSamples = closeInVect(t, timeSituations)[0]
envel = expPulses(t, timeSamples, decay, ampl)[0] - np.max(ampl)
ams = AMsign(t, 100, envel, depth=1)
signal = pa.awgn(ams[0], SNRdB=-2)[0]
# fig = plotSignal(signal, t, specKind='amplitude')
(representation, t0, fVectNew) = pa.DFTbank(signal,
rect=2,
level=0.2,
Fs=Fs,
mirrorLen=0.15,
df=1,
freqLims=(50, 200),
formFactor=128)
# fig4 = plotRepresentation(t, representation, fVectNew, None)
(alpha, f, A, theta, resid,
coefficient) = pa.thresholdRepreProny(representation,
fVectNew,
Fs=Fs,
periodsNum=10,
lowFreq=95,
highFreq=105,
hold=1.5)
'''''
fig3 = plt.figure()
plt.plot(t, alpha)
fig2 = plt.figure()
plt.plot(t, f)
fig4 = plt.figure()
plt.plot(t, resid)
fig4.show()
''' ''
print(np.mean((f[:int(f.size * 0.85)] - 100)**2))
# figFull = plotUnder(t, (representation[fInd, :], alpha, f), yLims=[(-0.6, 0.6), (12.5, 17.5), (95, 105)], secondParam=resid, secondLim=(0, 7*10**-6))
# fig100 = plotSignal(representation[fInd, :], t, specKind='amplitude')
(alpha1, f1, A1, theta1,
resid1) = pa.pronyDecomp(representation[fInd, :tEnd], 2, Fs=Fs)
print(np.sum(resid1.bse**2))
ams = AMsign(t, 100, 5, depth=0.25)
signal = pa.awgn(ams[0], SNRdB=-2)[0]
# figAM = plotSignal(signal, t, specKind='amplitude')
(representation, t0, fVectNew) = pa.DFTbank(signal,
rect=2,
level=0.2,
Fs=Fs,
mirrorLen=0.15,
df=1,
freqLims=(50, 200),
formFactor=1024)
fig4 = plotRepresentation(t, representation, fVectNew, None)
fig100 = plotSignal(representation[fInd, :], t, specKind='amplitude')
(alpha, f, A, theta, resid,
coefficient) = pa.thresholdRepreProny(representation,
fVectNew,
Fs=Fs,
percentLength=5,
percentOverlap=75,
lowFreq=95,
highFreq=105,
hold=1.4)
figFull = plotUnder(t, (representation[fInd, :], alpha, f),
yLims=[(-2, 2), (-20, 20), (95, 105)],
secondParam=resid,
secondLim=None) # (0, 2*10 ** -6)
pass
def waveProc(kwargs):
signal = kwargs.get('signal')
SNR = kwargs.get('SNR')
dFmax = kwargs.get('dFmax')
Fs = kwargs.get('Fs')
freq = kwargs.get('freq')
selectWinds = kwargs.get('selectWinds')
signalN = pa.awgn(signal, SNRdB=SNR)[0]
(representation, t0, fVectNew) = pa.DFTbank(signalN,
rect=2,
mirrorLen=0.15,
level=0.2,
freqLims=(50, 200),
Fs=Fs,
df=1,
formFactor=1024)
'''''
if plotGraphs == 2:
fig4 = plt.figure()
grid = matplotlib.gridspec.GridSpec(1, 1)
ax_specWav = fig4.add_subplot(grid[0])
extent = t[0], t[-1], fVectNew[0], fVectNew[-1]
ax_specWav.imshow(np.flipud(np.abs(representation)), extent=extent)
ax_specWav.axis('auto')
plt.ylim(int(np.min(freq) * 0.9), int(np.max(freq) * 1.1))
fig4.show()
''' ''
repFreq = pa.representationTrack(representation, fVectNew)
(representation, t0, fVectNew) = pa.DFTbank(signalN,
rect=2,
mirrorLen=0.15,
level=0.2,
freqLims=(50, 200),
Fs=Fs,
df=1,
formFactor=128)
(alpha, f, A, theta, resid,
coefficient) = pa.thresholdRepreProny(representation,
fVectNew,
Fs=Fs,
centralFreq=0,
lowFreq=95,
highFreq=130,
hold=1.4)
'''''
if plotGraphs:
ax_FreqDev.plot(t, freq[0, :], linestyle='-', color='k', label='Мгновенная частота')
ax_FreqR.plot(t, freq[0, :], linestyle='-', color='k', label='Мгновенная частота')
''' ''
indexes = np.hstack(np.nonzero(f))
# Select long stable windows.
if selectWinds:
errIdxs = pa.SM.validateTrack(f)
errIdxsRepr = pa.SM.validateTrack(repFreq)
else:
errIdxs = np.hstack(np.nonzero(f))
errIdxsRepr = np.hstack(np.nonzero(repFreq))
validLen = errIdxs.size / f.size
if validLen < 0.85:
errIdxs = indexes
validLenRepr = errIdxsRepr.size / f.size
if validLenRepr < 0.85:
errIdxsRepr = indexes
'''''
if plotGraphs:
ax_FreqDev.plot(t[indexes], f[indexes], linestyle=':', color=cmap1(bi), label='Оценка частоты')
# ax_FreqDev.plot(t[errIdxs], f[errIdxs], linestyle=':', color='r', label='Поправленная оценка')
ax_FreqDev.set_xlabel('Время, сек')
ax_FreqDev.set_ylabel('Частота, Гц')
ax_FreqR.plot(t, repFreq, linestyle=':', color=cmap1(bi), label='Оценка частоты')
ax_FreqR.set_xlabel('Время, сек')
ax_FreqR.set_ylabel('Частота, Гц')
''' ''
diff = freq[0, errIdxs] - f[errIdxs]
err = pa.rms(diff)
errRep = pa.rms(freq[0, errIdxsRepr] - repFreq[errIdxsRepr])
#print('Coefficient {}, error {}, valid track length {} percents, representation error {}, repres. valid track len = {}'.format(
#dFmax, err, validLen * 100, errRep, validLenRepr * 100))
'''''
if ai < 1 and bi < 1 and plotGraphs:
ax_FreqDev.legend()
ax_FreqR.legend()
''' ''
return err, errRep, validLen, validLenRepr
harmDisp = np.zeros_like(SNRs)
noisDisp = np.zeros_like(SNRs)
dF = np.zeros_like(SNRs)
dF2 = np.zeros((SNRs.size, distances.size))
freq = np.ones((1, t.size))
freq[0, :] = freq[0, :] * 100 * (
1 + np.linspace(start=0, stop=coefficients[ai], num=t.size))
(frq, frqWinIdxs) = pa.windowing(
freq[0, :], int(np.round(freq[0, :].size * percentLength / 100)),
percentOverlap)
fOrig = frq[:, 0]
for bi in range(SNRs.size):
for ci in range(experiences):
signal = chirp(t, freq[0, 0], t[-1], freq[0, -1])
#(alpha, f, A, theta, resid, coefficient) = pa.timeSegmentedProny(signal, Fs=Fs, percentLength=10, percentOverlap=50, order=2)
(signal, noise) = pa.awgn(signal, SNRdB=SNRs[bi])[0:2]
spec = np.fft.rfft(signal) / signal.size
spec[fVect < 95] = 0 # freq[0, 0] * 0.95
spec[fVect > 135] = 0 # freq[0, -1] * 1.05
cmp = np.fft.irfft(spec) * spec.size * 2
(alpha, f, A, theta, resid, coefficient) = pa.timeSegmentedProny(
cmp,
Fs=Fs,
percentLength=percentLength,
percentOverlap=percentOverlap,
order=2)
fTemp = np.array(f)[:, 1]
dF[bi] += pa.rms(fOrig - fTemp)
harmDisp += np.std(fTemp)
spec2 = np.fft.rfft(noise) / noise.size
spec2[fVect < 95] = 0 # freq[0, 0] * 0.95
comp = amplitude * chirp(t, freq[ai, 0], t[-1],
freq[ai, -1]) #*np.sin(fullPhase)
#comp = pa.abwgn(comp, SNRdB=-7, Fs=Fs, band=(freq[ai, 0]*0.95, freq[ai, -1]*1.05))[0]
signal = signal + comp
k[ai] = (freq[ai, -1] - freq[ai, 0]) / (t[-1] - t[0]) / freq[ai, 0]
#Plot instantateous frequency
if plotGraphs:
analytic_signal = hilbert(comp)
instantaneous_phase = np.unwrap(np.angle(analytic_signal))
instantaneous_frequency = (np.diff(instantaneous_phase) /
(2.0 * np.pi) * Fs)
plt.plot(t[1:], instantaneous_frequency)
# signal = chirp(t, freq[0, 0], t[-1], freq[0, -1])
# signal = pa.abwgn(signal, SNRdB=10, Fs=Fs, band=(95, 105))[0]
signal = pa.awgn(signal, SNRdB=-2)[0]
if plotGraphs:
spec = np.fft.rfft(signal) / signal.size
grid = matplotlib.gridspec.GridSpec(2, 1)
fig = plt.figure(figsize=(16, 10))
ax_signal = fig.add_subplot(grid[0])
ax_spectr = fig.add_subplot(grid[1])
ax_phase = ax_spectr.twinx()
ax_signal.plot(t, signal)
ax_spectr.plot(fVect, np.abs(spec))
ax_phase.plot(fVect, np.angle(spec) / np.pi, '1k')
ax_spectr.grid(color='r', linestyle='--')
ax_phase.grid(color='k', linestyle=':')
fig.show()
fig2 = plt.figure(figsize=(16, 10))