def correct_response(st, removeResp=False, removePAZ=False, simPAZ=False, pre_filt=None, cornFreq=0.0083):
"""
Correct the seismometer response.
Seismometer response is given in either a dictionary ``removeResp''
or a dictionary ``removePAZ''. ``removeResp has precedence. The
dictionaries have the following structure
removeResp: dictionary with Response information to be removed
has the following keys:
respfile: (str) filename of evalresp response file.
units: (str) Units to return response in. Can be either DIS, VEL or ACC
start_stage: (int) integer stage numbers of start stage (<0 causes
default evalresp bahaviour).
stop_stage: (int) integer stage numbers of stop stage
removePAZ: dictionary with poles and zeros to be removed has the following
keys:
poles: (list of complex numbers) location of poles
zeros: (list of complex numbers) location of zeros
gain: (float) gain
sensitivity: (float) sensitivity
It can easily be retrieved with obspy.arclink.client.Client.getPAZ
if ``removeResp'' is given the response of each trace must be present in
the respfile. If ``removePAZ'' is used the response is assumed to be the
same for all traces in the stream.
A filter specified in pre_filt can be applied in to avoid amplification of
noise.
The instrument to be simulated is either described in the dictionary simPAZ
or if simPAZ is False by the corner frequency ``cornFreq''. Response
correction is done in place and original data is overwritten.
The input stream ``st'' should be demeaned and tapered.
:type st: obspy.core.stream.Stream
:param st: data stream to be corrected
:type removeResp: dict
:param removeResp: Response information to be removed
:type removePAZ: dict
:param removePAZ: Response information to be removed
:type simPAZ: dict
:param simPAZ: Response information to be simulated
:type cornFreq: float
:param cornFreq: corner frequency of instrument to be simulated
:type pre_filt: list
:param pre_filt: 4 corners of the filter
"""
for tr in st:
starttime = tr.stats['starttime']
endtime = tr.stats['endtime']
network = tr.stats['network']
station = tr.stats['station']
channel = tr.stats['channel']
location = tr.stats['location']
length = tr.stats['npts']
sampling_rate = tr.stats['sampling_rate']
np2l = nextpow2(2.*length)
if not simPAZ:
simPAZ = cornFreq2Paz(cornFreq, damp=0.70716)
simresp, freqs = np.conj(pazToFreqResp(simPAZ['poles'], simPAZ['zeros'],
scale_fac=simPAZ['gain']*simPAZ['sensitivity'],
t_samp=1./sampling_rate,
nfft=np2l, freq=True)) #see Doc of pazToFreqResp for reason of conj()
if removeResp:
freqresp, freqs = evalresp(1./sampling_rate,np2l,removeResp['respfile'],
starttime, network=network, station=station,
channel=channel, locid=location,
start_stage=removeResp['start_stage'],
stop_stage=removeResp['stop_stage'],
units=removeResp['units'], freq=True)
else:
freqresp, freqs = np.conj(pazToFreqResp(removePAZ['poles'], removePAZ['zeros'],
scale_fac=removePAZ['gain']*removePAZ['sensitivity'],
t_samp=1./sampling_rate,
nfft=np2l, freq=True)) #see Doc of pazToFreqResp for reason of conj()
ftr = np.fft.rfft(tr.data,n=np2l)
ftr /= freqresp
ftr[0] = 0.j # correct the NaN in the DC component
ftr *= simresp
if pre_filt:
ftr *= c_sac_taper(freqs, flimit=pre_filt)
tr.data = np.fft.irfft(ftr)
tr.trim(starttime,endtime)
return
def time_shift_apply(corr_data, shift):
""" Apply time shift to traces.
Apply time shifts to traces e.g. to align them to a common time base.
Such shifts can occur in corrlation traces in case of a drifting clock.
This function ``applies`` the shifts. To correct for shift estimated with
:class:`~miic.core.stretch_mod.time_shift_estimate` you need to apply
negative shifts.
Shifting is done in frequency domain with 5% tapering.
:type corr_data: :py:class:`~numpy.ndarray`
:param corr_data: 2d ndarray containing the correlation functions that are
to be shifted.
One for each row.
:type shift: :py:class:`~numpy.ndarray`
:param shift: ndarray with shift.shape[0] = corr_data.shape[0] containing
the shifts in units of the sampling interval by which the trace are to
be shifted
:rtype: :py:class:`~numpy.ndarray`
:return: **shifted_mat**: shifted version of the input matrix
"""
mat = corr_data
# check input
# shift is just a 1d array
if len(shift.shape) == 1:
t_shift = np.zeros([shift.shape[0], 1])
t_shift[:, 0] = shift
shift = t_shift
# shift has the wrong length
elif shift.shape[0] != mat.shape[0]:
print 'InputError: shift.shape[0] must be equal corr_data.shape[0]'
return 0
# shift has multiple columns (multiple measurements for the same time)
if shift.shape[1] > 1:
shift = np.delete(shift, np.arange(1, shift.shape[1]), axis=1)
# taper the reference matrix to avoid interpolation
taper = cosTaper(mat.shape[1], 0.05)
mat *= np.tile(taper, [mat.shape[0], 1])
# find a suitable length for the FFT
N = nextpow2(2 * mat.shape[1])
w = np.zeros([1, N / 2 + 1])
# original and shifted phase
w[0, :] = np.linspace(0, np.pi, N / 2 + 1)
pha = np.exp(-1j * (shift) * w)
# Fourier Transform
F = np.fft.rfft(mat, N, 1)
# apply the phase shift
sF = F * pha
# transform to time domain
smat = np.fft.irfft(sF)
# cut to original size
shifted_mat = smat[:, 0:mat.shape[1]]
return shifted_mat
def correct_response(st,
removeResp=False,
removePAZ=False,
simPAZ=False,
pre_filt=None,
cornFreq=0.0083):
"""
Correct the seismometer response.
Seismometer response is given in either a dictionary ``removeResp''
or a dictionary ``removePAZ''. ``removeResp has precedence. The
dictionaries have the following structure
removeResp: dictionary with Response information to be removed
has the following keys:
respfile: (str) filename of evalresp response file.
units: (str) Units to return response in. Can be either DIS, VEL or ACC
start_stage: (int) integer stage numbers of start stage (<0 causes
default evalresp bahaviour).
stop_stage: (int) integer stage numbers of stop stage
removePAZ: dictionary with poles and zeros to be removed has the following
keys:
poles: (list of complex numbers) location of poles
zeros: (list of complex numbers) location of zeros
gain: (float) gain
sensitivity: (float) sensitivity
It can easily be retrieved with obspy.arclink.client.Client.getPAZ
if ``removeResp'' is given the response of each trace must be present in
the respfile. If ``removePAZ'' is used the response is assumed to be the
same for all traces in the stream.
A filter specified in pre_filt can be applied in to avoid amplification of
noise.
The instrument to be simulated is either described in the dictionary simPAZ
or if simPAZ is False by the corner frequency ``cornFreq''. Response
correction is done in place and original data is overwritten.
The input stream ``st'' should be demeaned and tapered.
:type st: obspy.core.stream.Stream
:param st: data stream to be corrected
:type removeResp: dict
:param removeResp: Response information to be removed
:type removePAZ: dict
:param removePAZ: Response information to be removed
:type simPAZ: dict
:param simPAZ: Response information to be simulated
:type cornFreq: float
:param cornFreq: corner frequency of instrument to be simulated
:type pre_filt: list
:param pre_filt: 4 corners of the filter
"""
for tr in st:
starttime = tr.stats['starttime']
endtime = tr.stats['endtime']
network = tr.stats['network']
station = tr.stats['station']
channel = tr.stats['channel']
location = tr.stats['location']
length = tr.stats['npts']
sampling_rate = tr.stats['sampling_rate']
np2l = nextpow2(2. * length)
if not simPAZ:
simPAZ = cornFreq2Paz(cornFreq, damp=0.70716)
simresp, freqs = np.conj(
pazToFreqResp(
simPAZ['poles'],
simPAZ['zeros'],
scale_fac=simPAZ['gain'] * simPAZ['sensitivity'],
t_samp=1. / sampling_rate,
nfft=np2l,
freq=True)) #see Doc of pazToFreqResp for reason of conj()
if removeResp:
freqresp, freqs = evalresp(1. / sampling_rate,
np2l,
removeResp['respfile'],
starttime,
network=network,
station=station,
channel=channel,
locid=location,
start_stage=removeResp['start_stage'],
stop_stage=removeResp['stop_stage'],
units=removeResp['units'],
freq=True)
else:
freqresp, freqs = np.conj(
pazToFreqResp(
removePAZ['poles'],
removePAZ['zeros'],
scale_fac=removePAZ['gain'] * removePAZ['sensitivity'],
t_samp=1. / sampling_rate,
nfft=np2l,
freq=True)) #see Doc of pazToFreqResp for reason of conj()
ftr = np.fft.rfft(tr.data, n=np2l)
ftr /= freqresp
ftr[0] = 0.j # correct the NaN in the DC component
ftr *= simresp
if pre_filt:
ftr *= c_sac_taper(freqs, flimit=pre_filt)
tr.data = np.fft.irfft(ftr)
tr.trim(starttime, endtime)
return
def time_shift_apply(corr_data, shift):
""" Apply time shift to traces.
Apply time shifts to traces e.g. to align them to a common time base.
Such shifts can occur in corrlation traces in case of a drifting clock.
This function ``applies`` the shifts. To correct for shift estimated with
:class:`~miic.core.stretch_mod.time_shift_estimate` you need to apply
negative shifts.
Shifting is done in frequency domain with 5% tapering.
:type corr_data: :py:class:`~numpy.ndarray`
:param corr_data: 2d ndarray containing the correlation functions that are
to be shifted.
One for each row.
:type shift: :py:class:`~numpy.ndarray`
:param shift: ndarray with shift.shape[0] = corr_data.shape[0] containing
the shifts in units of the sampling interval by which the trace are to
be shifted
:rtype: :py:class:`~numpy.ndarray`
:return: **shifted_mat**: shifted version of the input matrix
"""
mat = corr_data
# check input
# shift is just a 1d array
if len(shift.shape) == 1:
t_shift = np.zeros([shift.shape[0], 1])
t_shift[:, 0] = shift
shift = t_shift
# shift has the wrong length
elif shift.shape[0] != mat.shape[0]:
print "InputError: shift.shape[0] must be equal corr_data.shape[0]"
return 0
# shift has multiple columns (multiple measurements for the same time)
if shift.shape[1] > 1:
shift = np.delete(shift, np.arange(1, shift.shape[1]), axis=1)
# taper the reference matrix to avoid interpolation
taper = cosTaper(mat.shape[1], 0.05)
mat *= np.tile(taper, [mat.shape[0], 1])
# find a suitable length for the FFT
N = nextpow2(2 * mat.shape[1])
w = np.zeros([1, N / 2 + 1])
# original and shifted phase
w[0, :] = np.linspace(0, np.pi, N / 2 + 1)
pha = np.exp(-1j * (shift) * w)
# Fourier Transform
F = np.fft.rfft(mat, N, 1)
# apply the phase shift
sF = F * pha
# transform to time domain
smat = np.fft.irfft(sF)
# cut to original size
shifted_mat = smat[:, 0 : mat.shape[1]]
return shifted_mat
