Coverage for /usr/local/lib/python3.11/dist-packages/pyrocko/trace.py: 76%
1745 statements
« prev ^ index » next coverage.py v6.5.0, created at 2024-03-07 11:54 +0000
1# https://pyrocko.org - GPLv3
2#
3# The Pyrocko Developers, 21st Century
4# ---|P------/S----------~Lg----------
6'''
7Basic signal processing for seismic waveforms.
8'''
10import time
11import math
12import copy
13import logging
14import hashlib
15from collections import defaultdict
17import numpy as num
18from scipy import signal
20from pyrocko import util, orthodrome, pchain, model, signal_ext
21from pyrocko.util import reuse
22from pyrocko.guts import Object, Float, Int, String, List, \
23 StringChoice, Timestamp
24from pyrocko.guts_array import Array
25from pyrocko.model import squirrel_content
27# backward compatibility
28from pyrocko.util import UnavailableDecimation # noqa
29from pyrocko.response import ( # noqa
30 FrequencyResponse, Evalresp, InverseEvalresp, PoleZeroResponse,
31 ButterworthResponse, SampledResponse, IntegrationResponse,
32 DifferentiationResponse, MultiplyResponse)
35guts_prefix = 'pf'
37logger = logging.getLogger('pyrocko.trace')
40g_tapered_coeffs_cache = {}
41g_one_response = FrequencyResponse()
44@squirrel_content
45class Trace(Object):
47 '''
48 Create new trace object.
50 A ``Trace`` object represents a single continuous strip of evenly sampled
51 time series data. It is built from a 1D NumPy array containing the data
52 samples and some attributes describing its beginning and ending time, its
53 sampling rate and four string identifiers (its network, station, location
54 and channel code).
56 :param network: network code
57 :param station: station code
58 :param location: location code
59 :param channel: channel code
60 :param extra: extra code
61 :param tmin: system time of first sample in [s]
62 :param tmax: system time of last sample in [s] (if set to ``None`` it is
63 computed from length of ``ydata``)
64 :param deltat: sampling interval in [s]
65 :param ydata: 1D numpy array with data samples (can be ``None`` when
66 ``tmax`` is not ``None``)
67 :param mtime: optional modification time
69 :param meta: additional meta information (not used, but maintained by the
70 library)
72 The length of the network, station, location and channel codes is not
73 resricted by this software, but data formats like SAC, Mini-SEED or GSE
74 have different limits on the lengths of these codes. The codes set here are
75 silently truncated when the trace is stored
76 '''
78 network = String.T(default='', help='Deployment/network code (1-8)')
79 station = String.T(default='STA', help='Station code (1-5)')
80 location = String.T(default='', help='Location code (0-2)')
81 channel = String.T(default='', help='Channel code (3)')
82 extra = String.T(default='', help='Extra/custom code')
84 tmin = Timestamp.T(default=Timestamp.D('1970-01-01 00:00:00'))
85 tmax = Timestamp.T()
87 deltat = Float.T(default=1.0)
88 ydata = Array.T(optional=True, shape=(None,), serialize_as='base64+meta')
90 mtime = Timestamp.T(optional=True)
92 cached_frequencies = {}
94 def __init__(self, network='', station='STA', location='', channel='',
95 tmin=0., tmax=None, deltat=1., ydata=None, mtime=None,
96 meta=None, extra=''):
98 Object.__init__(self, init_props=False)
100 time_float = util.get_time_float()
102 if not isinstance(tmin, time_float):
103 tmin = Trace.tmin.regularize_extra(tmin)
105 if tmax is not None and not isinstance(tmax, time_float):
106 tmax = Trace.tmax.regularize_extra(tmax)
108 if mtime is not None and not isinstance(mtime, time_float):
109 mtime = Trace.mtime.regularize_extra(mtime)
111 if ydata is not None and not isinstance(ydata, num.ndarray):
112 ydata = Trace.ydata.regularize_extra(ydata)
114 self._growbuffer = None
116 tmin = time_float(tmin)
117 if tmax is not None:
118 tmax = time_float(tmax)
120 if mtime is None:
121 mtime = time_float(time.time())
123 self.network, self.station, self.location, self.channel, \
124 self.extra = [
125 reuse(x) for x in (
126 network, station, location, channel, extra)]
128 self.tmin = tmin
129 self.deltat = deltat
131 if tmax is None:
132 if ydata is not None:
133 self.tmax = self.tmin + (ydata.size-1)*self.deltat
134 else:
135 raise Exception(
136 'fixme: trace must be created with tmax or ydata')
137 else:
138 n = int(round((tmax - self.tmin) / self.deltat)) + 1
139 self.tmax = self.tmin + (n - 1) * self.deltat
141 self.meta = meta
142 self.ydata = ydata
143 self.mtime = mtime
144 self._update_ids()
145 self.file = None
146 self._pchain = None
148 def __str__(self):
149 fmt = min(9, max(0, -int(math.floor(math.log10(self.deltat)))))
150 s = 'Trace (%s, %s, %s, %s)\n' % self.nslc_id
151 s += ' timerange: %s - %s\n' % (
152 util.time_to_str(self.tmin, format=fmt),
153 util.time_to_str(self.tmax, format=fmt))
155 s += ' delta t: %g\n' % self.deltat
156 if self.meta:
157 for k in sorted(self.meta.keys()):
158 s += ' %s: %s\n' % (k, self.meta[k])
159 return s
161 @property
162 def codes(self):
163 from pyrocko.squirrel import model
164 return model.CodesNSLCE(
165 self.network, self.station, self.location, self.channel,
166 self.extra)
168 @property
169 def time_span(self):
170 return self.tmin, self.tmax
172 @property
173 def summary_entries(self):
174 return (
175 self.__class__.__name__,
176 str(self.codes),
177 self.str_time_span,
178 '%g' % (1.0/self.deltat))
180 @property
181 def summary_stats_entries(self):
182 return tuple('%13.7g' % v for v in (
183 self.ydata.min(),
184 self.ydata.max(),
185 self.ydata.mean(),
186 self.ydata.std()))
188 @property
189 def summary(self):
190 return util.fmt_summary(self.summary_entries, (10, 20, 55, 0))
192 @property
193 def summary_stats(self):
194 return self.summary + ' | ' + util.fmt_summary(
195 self.summary_stats_entries, (12, 12, 12, 12))
197 def __getstate__(self):
198 return (self.network, self.station, self.location, self.channel,
199 self.tmin, self.tmax, self.deltat, self.mtime,
200 self.ydata, self.meta, self.extra)
202 def __setstate__(self, state):
203 if len(state) == 11:
204 self.network, self.station, self.location, self.channel, \
205 self.tmin, self.tmax, self.deltat, self.mtime, \
206 self.ydata, self.meta, self.extra = state
208 elif len(state) == 12:
209 # backward compatibility 0
210 self.network, self.station, self.location, self.channel, \
211 self.tmin, self.tmax, self.deltat, self.mtime, \
212 self.ydata, self.meta, _, self.extra = state
214 elif len(state) == 10:
215 # backward compatibility 1
216 self.network, self.station, self.location, self.channel, \
217 self.tmin, self.tmax, self.deltat, self.mtime, \
218 self.ydata, self.meta = state
220 self.extra = ''
222 else:
223 # backward compatibility 2
224 self.network, self.station, self.location, self.channel, \
225 self.tmin, self.tmax, self.deltat, self.mtime = state
226 self.ydata = None
227 self.meta = None
228 self.extra = ''
230 self._growbuffer = None
231 self._update_ids()
233 def hash(self, unsafe=False):
234 sha1 = hashlib.sha1()
235 for code in self.nslc_id:
236 sha1.update(code.encode())
237 sha1.update(self.tmin.hex().encode())
238 sha1.update(self.tmax.hex().encode())
239 sha1.update(self.deltat.hex().encode())
241 if self.ydata is not None:
242 if not unsafe:
243 sha1.update(memoryview(self.ydata))
244 else:
245 sha1.update(memoryview(self.ydata[:32]))
246 sha1.update(memoryview(self.ydata[-32:]))
248 return sha1.hexdigest()
250 def name(self):
251 '''
252 Get a short string description.
253 '''
255 s = '%s.%s.%s.%s, %s, %s' % (self.nslc_id + (
256 util.time_to_str(self.tmin),
257 util.time_to_str(self.tmax)))
259 return s
261 def __eq__(self, other):
262 return (
263 isinstance(other, Trace)
264 and self.network == other.network
265 and self.station == other.station
266 and self.location == other.location
267 and self.channel == other.channel
268 and (abs(self.deltat - other.deltat)
269 < (self.deltat + other.deltat)*1e-6)
270 and abs(self.tmin-other.tmin) < self.deltat*0.01
271 and abs(self.tmax-other.tmax) < self.deltat*0.01
272 and num.all(self.ydata == other.ydata))
274 def almost_equal(self, other, rtol=1e-5, atol=0.0):
275 return (
276 self.network == other.network
277 and self.station == other.station
278 and self.location == other.location
279 and self.channel == other.channel
280 and (abs(self.deltat - other.deltat)
281 < (self.deltat + other.deltat)*1e-6)
282 and abs(self.tmin-other.tmin) < self.deltat*0.01
283 and abs(self.tmax-other.tmax) < self.deltat*0.01
284 and num.allclose(self.ydata, other.ydata, rtol=rtol, atol=atol))
286 def assert_almost_equal(self, other, rtol=1e-5, atol=0.0):
288 assert self.network == other.network, \
289 'network codes differ: %s, %s' % (self.network, other.network)
290 assert self.station == other.station, \
291 'station codes differ: %s, %s' % (self.station, other.station)
292 assert self.location == other.location, \
293 'location codes differ: %s, %s' % (self.location, other.location)
294 assert self.channel == other.channel, 'channel codes differ'
295 assert (abs(self.deltat - other.deltat)
296 < (self.deltat + other.deltat)*1e-6), \
297 'sampling intervals differ %g, %g' % (self.deltat, other.delta)
298 assert abs(self.tmin-other.tmin) < self.deltat*0.01, \
299 'start times differ: %s, %s' % (
300 util.time_to_str(self.tmin), util.time_to_str(other.tmin))
301 assert abs(self.tmax-other.tmax) < self.deltat*0.01, \
302 'end times differ: %s, %s' % (
303 util.time_to_str(self.tmax), util.time_to_str(other.tmax))
305 assert num.allclose(self.ydata, other.ydata, rtol=rtol, atol=atol), \
306 'trace values differ'
308 def __hash__(self):
309 return id(self)
311 def __call__(self, t, clip=False, snap=round):
312 it = int(snap((t-self.tmin)/self.deltat))
313 if clip:
314 it = max(0, min(it, self.ydata.size-1))
315 else:
316 if it < 0 or self.ydata.size <= it:
317 raise IndexError()
319 return self.tmin+it*self.deltat, self.ydata[it]
321 def interpolate(self, t, clip=False):
322 '''
323 Value of trace between supporting points through linear interpolation.
325 :param t: time instant
326 :param clip: whether to clip indices to trace ends
327 '''
329 t0, y0 = self(t, clip=clip, snap=math.floor)
330 t1, y1 = self(t, clip=clip, snap=math.ceil)
331 if t0 == t1:
332 return y0
333 else:
334 return y0+(t-t0)/(t1-t0)*(y1-y0)
336 def index_clip(self, i):
337 '''
338 Clip index to valid range.
339 '''
341 return min(max(0, i), self.ydata.size)
343 def add(self, other, interpolate=True, left=0., right=0.):
344 '''
345 Add values of other trace (self += other).
347 Add values of ``other`` trace to the values of ``self``, where it
348 intersects with ``other``. This method does not change the extent of
349 ``self``. If ``interpolate`` is ``True`` (the default), the values of
350 ``other`` to be added are interpolated at sampling instants of
351 ``self``. Linear interpolation is performed. In this case the sampling
352 rate of ``other`` must be equal to or lower than that of ``self``. If
353 ``interpolate`` is ``False``, the sampling rates of the two traces must
354 match.
355 '''
357 if interpolate:
358 assert self.deltat <= other.deltat \
359 or same_sampling_rate(self, other)
361 other_xdata = other.get_xdata()
362 xdata = self.get_xdata()
363 self.ydata += num.interp(
364 xdata, other_xdata, other.ydata, left=left, right=left)
365 else:
366 assert self.deltat == other.deltat
367 ioff = int(round((other.tmin-self.tmin)/self.deltat))
368 ibeg = max(0, ioff)
369 iend = min(self.data_len(), ioff+other.data_len())
370 self.ydata[ibeg:iend] += other.ydata[ibeg-ioff:iend-ioff]
372 def mult(self, other, interpolate=True):
373 '''
374 Muliply with values of other trace ``(self *= other)``.
376 Multiply values of ``other`` trace to the values of ``self``, where it
377 intersects with ``other``. This method does not change the extent of
378 ``self``. If ``interpolate`` is ``True`` (the default), the values of
379 ``other`` to be multiplied are interpolated at sampling instants of
380 ``self``. Linear interpolation is performed. In this case the sampling
381 rate of ``other`` must be equal to or lower than that of ``self``. If
382 ``interpolate`` is ``False``, the sampling rates of the two traces must
383 match.
384 '''
386 if interpolate:
387 assert self.deltat <= other.deltat or \
388 same_sampling_rate(self, other)
390 other_xdata = other.get_xdata()
391 xdata = self.get_xdata()
392 self.ydata *= num.interp(
393 xdata, other_xdata, other.ydata, left=0., right=0.)
394 else:
395 assert self.deltat == other.deltat
396 ibeg1 = int(round((other.tmin-self.tmin)/self.deltat))
397 ibeg2 = int(round((self.tmin-other.tmin)/self.deltat))
398 iend1 = int(round((other.tmax-self.tmin)/self.deltat))+1
399 iend2 = int(round((self.tmax-other.tmin)/self.deltat))+1
401 ibeg1 = self.index_clip(ibeg1)
402 iend1 = self.index_clip(iend1)
403 ibeg2 = self.index_clip(ibeg2)
404 iend2 = self.index_clip(iend2)
406 self.ydata[ibeg1:iend1] *= other.ydata[ibeg2:iend2]
408 def max(self):
409 '''
410 Get time and value of data maximum.
411 '''
413 i = num.argmax(self.ydata)
414 return self.tmin + i*self.deltat, self.ydata[i]
416 def min(self):
417 '''
418 Get time and value of data minimum.
419 '''
421 i = num.argmin(self.ydata)
422 return self.tmin + i*self.deltat, self.ydata[i]
424 def absmax(self):
425 '''
426 Get time and value of maximum of the absolute of data.
427 '''
429 tmi, mi = self.min()
430 tma, ma = self.max()
431 if abs(mi) > abs(ma):
432 return tmi, abs(mi)
433 else:
434 return tma, abs(ma)
436 def set_codes(
437 self, network=None, station=None, location=None, channel=None,
438 extra=None):
440 '''
441 Set network, station, location, and channel codes.
442 '''
444 if network is not None:
445 self.network = network
446 if station is not None:
447 self.station = station
448 if location is not None:
449 self.location = location
450 if channel is not None:
451 self.channel = channel
452 if extra is not None:
453 self.extra = extra
455 self._update_ids()
457 def set_network(self, network):
458 self.network = network
459 self._update_ids()
461 def set_station(self, station):
462 self.station = station
463 self._update_ids()
465 def set_location(self, location):
466 self.location = location
467 self._update_ids()
469 def set_channel(self, channel):
470 self.channel = channel
471 self._update_ids()
473 def overlaps(self, tmin, tmax):
474 '''
475 Check if trace has overlap with a given time span.
476 '''
477 return not (tmax < self.tmin or self.tmax < tmin)
479 def is_relevant(self, tmin, tmax, selector=None):
480 '''
481 Check if trace has overlap with a given time span and matches a
482 condition callback. (internal use)
483 '''
485 return not (tmax <= self.tmin or self.tmax < tmin) \
486 and (selector is None or selector(self))
488 def _update_ids(self):
489 '''
490 Update dependent ids.
491 '''
493 self.full_id = (
494 self.network, self.station, self.location, self.channel, self.tmin)
495 self.nslc_id = reuse(
496 (self.network, self.station, self.location, self.channel))
498 def prune_from_reuse_cache(self):
499 util.deuse(self.nslc_id)
500 util.deuse(self.network)
501 util.deuse(self.station)
502 util.deuse(self.location)
503 util.deuse(self.channel)
505 def set_mtime(self, mtime):
506 '''
507 Set modification time of the trace.
508 '''
510 self.mtime = mtime
512 def get_xdata(self):
513 '''
514 Create array for time axis.
515 '''
517 if self.ydata is None:
518 raise NoData()
520 return self.tmin \
521 + num.arange(len(self.ydata), dtype=num.float64) * self.deltat
523 def get_ydata(self):
524 '''
525 Get data array.
526 '''
528 if self.ydata is None:
529 raise NoData()
531 return self.ydata
533 def set_ydata(self, new_ydata):
534 '''
535 Replace data array.
536 '''
538 self.drop_growbuffer()
539 self.ydata = new_ydata
540 self.tmax = self.tmin+(len(self.ydata)-1)*self.deltat
542 def data_len(self):
543 if self.ydata is not None:
544 return self.ydata.size
545 else:
546 return int(round((self.tmax-self.tmin)/self.deltat)) + 1
548 def drop_data(self):
549 '''
550 Forget data, make dataless trace.
551 '''
553 self.drop_growbuffer()
554 self.ydata = None
556 def drop_growbuffer(self):
557 '''
558 Detach the traces grow buffer.
559 '''
561 self._growbuffer = None
562 self._pchain = None
564 def copy(self, data=True):
565 '''
566 Make a deep copy of the trace.
567 '''
569 tracecopy = copy.copy(self)
570 tracecopy.drop_growbuffer()
571 if data:
572 tracecopy.ydata = self.ydata.copy()
573 tracecopy.meta = copy.deepcopy(self.meta)
574 return tracecopy
576 def crop_zeros(self):
577 '''
578 Remove any zeros at beginning and end.
579 '''
581 indices = num.where(self.ydata != 0.0)[0]
582 if indices.size == 0:
583 raise NoData()
585 ibeg = indices[0]
586 iend = indices[-1]+1
587 if ibeg == 0 and iend == self.ydata.size-1:
588 return
590 self.drop_growbuffer()
591 self.ydata = self.ydata[ibeg:iend].copy()
592 self.tmin = self.tmin+ibeg*self.deltat
593 self.tmax = self.tmin+(len(self.ydata)-1)*self.deltat
594 self._update_ids()
596 def append(self, data):
597 '''
598 Append data to the end of the trace.
600 To make this method efficient when successively very few or even single
601 samples are appended, a larger grow buffer is allocated upon first
602 invocation. The traces data is then changed to be a view into the
603 currently filled portion of the grow buffer array.
604 '''
606 assert self.ydata.dtype == data.dtype
607 newlen = data.size + self.ydata.size
608 if self._growbuffer is None or self._growbuffer.size < newlen:
609 self._growbuffer = num.empty(newlen*2, dtype=self.ydata.dtype)
610 self._growbuffer[:self.ydata.size] = self.ydata
611 self._growbuffer[self.ydata.size:newlen] = data
612 self.ydata = self._growbuffer[:newlen]
613 self.tmax = self.tmin + (newlen-1)*self.deltat
615 def chop(
616 self, tmin, tmax, inplace=True, include_last=False,
617 snap=(round, round), want_incomplete=True):
619 '''
620 Cut the trace to given time span.
622 If the ``inplace`` argument is True (the default) the trace is cut in
623 place, otherwise a new trace with the cut part is returned. By
624 default, the indices where to start and end the trace data array are
625 determined by rounding of ``tmin`` and ``tmax`` to sampling instances
626 using Python's :py:func:`round` function. This behaviour can be changed
627 with the ``snap`` argument, which takes a tuple of two functions (one
628 for the lower and one for the upper end) to be used instead of
629 :py:func:`round`. The last sample is by default not included unless
630 ``include_last`` is set to True. If the given time span exceeds the
631 available time span of the trace, the available part is returned,
632 unless ``want_incomplete`` is set to False - in that case, a
633 :py:exc:`NoData` exception is raised. This exception is always raised,
634 when the requested time span does dot overlap with the trace's time
635 span.
636 '''
638 if want_incomplete:
639 if tmax <= self.tmin-self.deltat or self.tmax+self.deltat < tmin:
640 raise NoData()
641 else:
642 if tmin < self.tmin or self.tmax < tmax:
643 raise NoData()
645 ibeg = max(0, t2ind(tmin-self.tmin, self.deltat, snap[0]))
646 iplus = 0
647 if include_last:
648 iplus = 1
650 iend = min(
651 self.data_len(),
652 t2ind(tmax-self.tmin, self.deltat, snap[1])+iplus)
654 if ibeg >= iend:
655 raise NoData()
657 obj = self
658 if not inplace:
659 obj = self.copy(data=False)
661 self.drop_growbuffer()
662 if self.ydata is not None:
663 obj.ydata = self.ydata[ibeg:iend].copy()
664 else:
665 obj.ydata = None
667 obj.tmin = obj.tmin+ibeg*obj.deltat
668 obj.tmax = obj.tmin+((iend-ibeg)-1)*obj.deltat
670 obj._update_ids()
672 return obj
674 def downsample(
675 self, ndecimate, snap=False, demean=False, ftype='fir-remez',
676 cut=False):
678 '''
679 Downsample (decimate) trace by a given integer factor.
681 Antialiasing filter details:
683 * ``iir``: A Chebyshev type I digital filter of order 8 with maximum
684 ripple of 0.05 dB in the passband.
685 * ``fir``: A FIR filter using a Hamming window and 31 taps and a
686 well-behaved phase response.
687 * ``fir-remez``: A FIR filter based on the Remez exchange algorithm
688 with ``45*ndecimate`` taps and a cutoff at 75% Nyquist frequency.
690 Comparison of the digital filters:
692 .. figure :: ../../static/downsampling-filter-comparison.png
693 :width: 60%
694 :alt: Comparison of the downsampling filters.
696 See also :py:meth:`Trace.downsample_to`.
698 :param ndecimate:
699 Decimation factor, avoid values larger than 8.
700 :type ndecimate:
701 int
703 :param snap:
704 Whether to put the new sampling instants closest to multiples of
705 the sampling rate (according to absolute time).
706 :type snap:
707 bool
709 :param demean:
710 Whether to demean the signal before filtering.
711 :type demean:
712 bool
714 :param ftype:
715 Which FIR filter to use, choose from ``'iir'``, ``'fir'``,
716 ``'fir-remez'``. Default is ``'fir-remez'``.
718 :param cut:
719 Whether to cut off samples in the beginning of the trace which
720 are polluted by artifacts of the anti-aliasing filter.
721 :type cut:
722 bool
723 '''
724 newdeltat = self.deltat*ndecimate
725 b, a, n = util.decimate_coeffs(ndecimate, None, ftype)
726 if snap:
727 ilag = int(round((math.ceil(
728 (self.tmin+(n//2 if cut else 0)*self.deltat) /
729 newdeltat) * newdeltat - self.tmin) / self.deltat))
730 else:
731 ilag = (n//2 if cut else 0)
733 data = self.ydata.astype(num.float64)
734 if data.size != 0:
735 if demean:
736 data -= num.mean(data)
737 y = signal.lfilter(b, a, data)
738 self.ydata = y[n//2+ilag::ndecimate].copy()
739 else:
740 self.ydata = data
742 self.tmin += ilag * self.deltat
743 self.deltat = reuse(self.deltat*ndecimate)
744 self.tmax = self.tmin+(len(self.ydata)-1)*self.deltat
745 self._update_ids()
747 def downsample_to(
748 self, deltat, snap=False, allow_upsample_max=1, demean=False,
749 ftype='fir-remez', cut=False):
751 '''
752 Downsample to given sampling rate.
754 Tries to downsample the trace to a target sampling interval of
755 ``deltat``. This runs :py:meth:`downsample` one or several times. If
756 ``allow_upsample_max`` is set to a value larger than 1, intermediate
757 upsampling steps are allowed, in order to increase the number of
758 possible downsampling ratios.
760 If the requested ratio is not supported, an exception of type
761 :py:exc:`pyrocko.util.UnavailableDecimation` is raised.
763 The downsampled trace will be shorter than the input trace because the
764 anti-aliasing filter introduces edge effects. If `cut` is selected,
765 additionally, polluted samples in the beginning of the trace are
766 removed. The approximate amount of cutoff which will be produced by a
767 given downsampling configuration can be estimated with
768 :py:func:`downsample_tpad`.
770 See also: :meth:`Trace.downsample`.
772 :param deltat:
773 Desired sampling interval in [s].
774 :type deltat:
775 float
777 :param allow_upsample_max:
778 Maximum allowed upsampling factor of the signal to achieve the
779 desired sampling interval. Default is ``1``.
780 :type allow_upsample_max:
781 int
783 :param snap:
784 Whether to put the new sampling instants closest to multiples of
785 the sampling rate (according to absolute time).
786 :type snap:
787 bool
789 :param demean:
790 Whether to demean the signal before filtering.
791 :type demean:
792 bool
794 :param ftype:
795 Which FIR filter to use, choose from ``'iir'``, ``'fir'``,
796 ``'fir-remez'``. Default is ``'fir-remez'``.
798 :param cut:
799 Whether to cut off samples in the beginning of the trace which
800 are polluted by artifacts of the anti-aliasing filter.
801 :type demean:
802 bool
803 '''
805 upsratio, deci_seq = _configure_downsampling(
806 self.deltat, deltat, allow_upsample_max)
808 if demean:
809 self.drop_growbuffer()
810 self.ydata = self.ydata.astype(num.float64)
811 self.ydata -= num.mean(self.ydata)
813 if upsratio > 1:
814 self.drop_growbuffer()
815 ydata = self.ydata
816 self.ydata = num.zeros(
817 ydata.size*upsratio-(upsratio-1), ydata.dtype)
818 self.ydata[::upsratio] = ydata
819 for i in range(1, upsratio):
820 self.ydata[i::upsratio] = \
821 float(i)/upsratio * ydata[:-1] \
822 + float(upsratio-i)/upsratio * ydata[1:]
823 self.deltat = self.deltat/upsratio
825 for i, ndecimate in enumerate(deci_seq):
826 self.downsample(
827 ndecimate, snap=snap, demean=False, ftype=ftype, cut=cut)
829 def resample(self, deltat):
830 '''
831 Resample to given sampling rate ``deltat``.
833 Resampling is performed in the frequency domain.
834 '''
836 ndata = self.ydata.size
837 ntrans = nextpow2(ndata)
838 fntrans2 = ntrans * self.deltat/deltat
839 ntrans2 = int(round(fntrans2))
840 deltat2 = self.deltat * float(ntrans)/float(ntrans2)
841 ndata2 = int(round(ndata*self.deltat/deltat2))
842 if abs(fntrans2 - ntrans2) > 1e-7:
843 logger.warning(
844 'resample: requested deltat %g could not be matched exactly: '
845 '%g' % (deltat, deltat2))
847 data = self.ydata
848 data_pad = num.zeros(ntrans, dtype=float)
849 data_pad[:ndata] = data
850 fdata = num.fft.rfft(data_pad)
851 fdata2 = num.zeros((ntrans2+1)//2, dtype=fdata.dtype)
852 n = min(fdata.size, fdata2.size)
853 fdata2[:n] = fdata[:n]
854 data2 = num.fft.irfft(fdata2)
855 data2 = data2[:ndata2]
856 data2 *= float(ntrans2) / float(ntrans)
857 self.deltat = deltat2
858 self.set_ydata(data2)
860 def resample_simple(self, deltat):
861 tyear = 3600*24*365.
863 if deltat == self.deltat:
864 return
866 if abs(self.deltat - deltat) * tyear/deltat < deltat:
867 logger.warning(
868 'resample_simple: less than one sample would have to be '
869 'inserted/deleted per year. Doing nothing.')
870 return
872 ninterval = int(round(deltat / (self.deltat - deltat)))
873 if abs(ninterval) < 20:
874 logger.error(
875 'resample_simple: sample insertion/deletion interval less '
876 'than 20. results would be erroneous.')
877 raise ResamplingFailed()
879 delete = False
880 if ninterval < 0:
881 ninterval = - ninterval
882 delete = True
884 tyearbegin = util.year_start(self.tmin)
886 nmin = int(round((self.tmin - tyearbegin)/deltat))
888 ibegin = (((nmin-1)//ninterval)+1) * ninterval - nmin
889 nindices = (len(self.ydata) - ibegin - 1) / ninterval + 1
890 if nindices > 0:
891 indices = ibegin + num.arange(nindices) * ninterval
892 data_split = num.split(self.ydata, indices)
893 data = []
894 for ln, h in zip(data_split[:-1], data_split[1:]):
895 if delete:
896 ln = ln[:-1]
898 data.append(ln)
899 if not delete:
900 if ln.size == 0:
901 v = h[0]
902 else:
903 v = 0.5*(ln[-1] + h[0])
904 data.append(num.array([v], dtype=ln.dtype))
906 data.append(data_split[-1])
908 ydata_new = num.concatenate(data)
910 self.tmin = tyearbegin + nmin * deltat
911 self.deltat = deltat
912 self.set_ydata(ydata_new)
914 def stretch(self, tmin_new, tmax_new):
915 '''
916 Stretch signal while preserving sample rate using sinc interpolation.
918 :param tmin_new: new time of first sample
919 :param tmax_new: new time of last sample
921 This method can be used to correct for a small linear time drift or to
922 introduce sub-sample time shifts. The amount of stretching is limited
923 to 10% by the implementation and is expected to be much smaller than
924 that by the approximations used.
925 '''
927 i_control = num.array([0, self.ydata.size-1], dtype=num.int64)
928 t_control = num.array([tmin_new, tmax_new], dtype=float)
930 r = (tmax_new - tmin_new) / self.deltat + 1.0
931 n_new = int(round(r))
932 if abs(n_new - r) > 0.001:
933 n_new = int(math.floor(r))
935 assert n_new >= 2
937 tmax_new = tmin_new + (n_new-1) * self.deltat
939 ydata_new = num.empty(n_new, dtype=float)
940 signal_ext.antidrift(i_control, t_control,
941 self.ydata.astype(float),
942 tmin_new, self.deltat, ydata_new)
944 self.tmin = tmin_new
945 self.set_ydata(ydata_new)
946 self._update_ids()
948 def nyquist_check(self, frequency, intro='Corner frequency', warn=True,
949 raise_exception=False):
951 '''
952 Check if a given frequency is above the Nyquist frequency of the trace.
954 :param intro: string used to introduce the warning/error message
955 :param warn: whether to emit a warning
956 :param raise_exception: whether to raise an :py:exc:`AboveNyquist`
957 exception.
958 '''
960 if frequency >= 0.5/self.deltat:
961 message = '%s (%g Hz) is equal to or higher than nyquist ' \
962 'frequency (%g Hz). (Trace %s)' \
963 % (intro, frequency, 0.5/self.deltat, self.name())
964 if warn:
965 logger.warning(message)
966 if raise_exception:
967 raise AboveNyquist(message)
969 def lowpass(self, order, corner, nyquist_warn=True,
970 nyquist_exception=False, demean=True):
972 '''
973 Apply Butterworth lowpass to the trace.
975 :param order: order of the filter
976 :param corner: corner frequency of the filter
978 Mean is removed before filtering.
979 '''
981 self.nyquist_check(
982 corner, 'Corner frequency of lowpass', nyquist_warn,
983 nyquist_exception)
985 (b, a) = _get_cached_filter_coeffs(
986 order, [corner*2.0*self.deltat], btype='low')
988 if len(a) != order+1 or len(b) != order+1:
989 logger.warning(
990 'Erroneous filter coefficients returned by '
991 'scipy.signal.butter(). You may need to downsample the '
992 'signal before filtering.')
994 data = self.ydata.astype(num.float64)
995 if demean:
996 data -= num.mean(data)
997 self.drop_growbuffer()
998 self.ydata = signal.lfilter(b, a, data)
1000 def highpass(self, order, corner, nyquist_warn=True,
1001 nyquist_exception=False, demean=True):
1003 '''
1004 Apply butterworth highpass to the trace.
1006 :param order: order of the filter
1007 :param corner: corner frequency of the filter
1009 Mean is removed before filtering.
1010 '''
1012 self.nyquist_check(
1013 corner, 'Corner frequency of highpass', nyquist_warn,
1014 nyquist_exception)
1016 (b, a) = _get_cached_filter_coeffs(
1017 order, [corner*2.0*self.deltat], btype='high')
1019 data = self.ydata.astype(num.float64)
1020 if len(a) != order+1 or len(b) != order+1:
1021 logger.warning(
1022 'Erroneous filter coefficients returned by '
1023 'scipy.signal.butter(). You may need to downsample the '
1024 'signal before filtering.')
1025 if demean:
1026 data -= num.mean(data)
1027 self.drop_growbuffer()
1028 self.ydata = signal.lfilter(b, a, data)
1030 def bandpass(self, order, corner_hp, corner_lp, demean=True):
1031 '''
1032 Apply butterworth bandpass to the trace.
1034 :param order: order of the filter
1035 :param corner_hp: lower corner frequency of the filter
1036 :param corner_lp: upper corner frequency of the filter
1038 Mean is removed before filtering.
1039 '''
1041 self.nyquist_check(corner_hp, 'Lower corner frequency of bandpass')
1042 self.nyquist_check(corner_lp, 'Higher corner frequency of bandpass')
1043 (b, a) = _get_cached_filter_coeffs(
1044 order,
1045 [corner*2.0*self.deltat for corner in (corner_hp, corner_lp)],
1046 btype='band')
1047 data = self.ydata.astype(num.float64)
1048 if demean:
1049 data -= num.mean(data)
1050 self.drop_growbuffer()
1051 self.ydata = signal.lfilter(b, a, data)
1053 def bandstop(self, order, corner_hp, corner_lp, demean=True):
1054 '''
1055 Apply bandstop (attenuates frequencies in band) to the trace.
1057 :param order: order of the filter
1058 :param corner_hp: lower corner frequency of the filter
1059 :param corner_lp: upper corner frequency of the filter
1061 Mean is removed before filtering.
1062 '''
1064 self.nyquist_check(corner_hp, 'Lower corner frequency of bandstop')
1065 self.nyquist_check(corner_lp, 'Higher corner frequency of bandstop')
1066 (b, a) = _get_cached_filter_coeffs(
1067 order,
1068 [corner*2.0*self.deltat for corner in (corner_hp, corner_lp)],
1069 btype='bandstop')
1070 data = self.ydata.astype(num.float64)
1071 if demean:
1072 data -= num.mean(data)
1073 self.drop_growbuffer()
1074 self.ydata = signal.lfilter(b, a, data)
1076 def envelope(self, inplace=True):
1077 '''
1078 Calculate the envelope of the trace.
1080 :param inplace: calculate envelope in place
1082 The calculation follows:
1084 .. math::
1086 Y' = \\sqrt{Y^2+H(Y)^2}
1088 where H is the Hilbert-Transform of the signal Y.
1089 '''
1091 y = self.ydata.astype(float)
1092 env = num.abs(hilbert(y))
1093 if inplace:
1094 self.drop_growbuffer()
1095 self.ydata = env
1096 else:
1097 tr = self.copy(data=False)
1098 tr.ydata = env
1099 return tr
1101 def taper(self, taperer, inplace=True, chop=False):
1102 '''
1103 Apply a :py:class:`Taper` to the trace.
1105 :param taperer: instance of :py:class:`Taper` subclass
1106 :param inplace: apply taper inplace
1107 :param chop: if ``True``: exclude tapered parts from the resulting
1108 trace
1109 '''
1111 if not inplace:
1112 tr = self.copy()
1113 else:
1114 tr = self
1116 if chop:
1117 i, n = taperer.span(tr.ydata, tr.tmin, tr.deltat)
1118 tr.shift(i*tr.deltat)
1119 tr.set_ydata(tr.ydata[i:i+n])
1121 taperer(tr.ydata, tr.tmin, tr.deltat)
1123 if not inplace:
1124 return tr
1126 def whiten(self, order=6):
1127 '''
1128 Whiten signal in time domain using autoregression and recursive filter.
1130 :param order: order of the autoregression process
1131 '''
1133 b, a = self.whitening_coefficients(order)
1134 self.drop_growbuffer()
1135 self.ydata = signal.lfilter(b, a, self.ydata)
1137 def whitening_coefficients(self, order=6):
1138 ar = yulewalker(self.ydata, order)
1139 b, a = [1.] + ar.tolist(), [1.]
1140 return b, a
1142 def ampspec_whiten(
1143 self,
1144 width,
1145 td_taper='auto',
1146 fd_taper='auto',
1147 pad_to_pow2=True,
1148 demean=True):
1150 '''
1151 Whiten signal via frequency domain using moving average on amplitude
1152 spectra.
1154 :param width: width of smoothing kernel [Hz]
1155 :param td_taper: time domain taper, object of type :py:class:`Taper` or
1156 ``None`` or ``'auto'``.
1157 :param fd_taper: frequency domain taper, object of type
1158 :py:class:`Taper` or ``None`` or ``'auto'``.
1159 :param pad_to_pow2: whether to pad the signal with zeros up to a length
1160 of 2^n
1161 :param demean: whether to demean the signal before tapering
1163 The signal is first demeaned and then tapered using ``td_taper``. Then,
1164 the spectrum is calculated and inversely weighted with a smoothed
1165 version of its amplitude spectrum. A moving average is used for the
1166 smoothing. The smoothed spectrum is then tapered using ``fd_taper``.
1167 Finally, the smoothed and tapered spectrum is back-transformed into the
1168 time domain.
1170 If ``td_taper`` is set to ``'auto'``, ``CosFader(1.0/width)`` is used.
1171 If ``fd_taper`` is set to ``'auto'``, ``CosFader(width)`` is used.
1172 '''
1174 ndata = self.data_len()
1176 if pad_to_pow2:
1177 ntrans = nextpow2(ndata)
1178 else:
1179 ntrans = ndata
1181 df = 1./(ntrans*self.deltat)
1182 nw = int(round(width/df))
1183 if ndata//2+1 <= nw:
1184 raise TraceTooShort(
1185 'Samples in trace: %s, samples needed: %s' % (ndata, nw))
1187 if td_taper == 'auto':
1188 td_taper = CosFader(1./width)
1190 if fd_taper == 'auto':
1191 fd_taper = CosFader(width)
1193 if td_taper:
1194 self.taper(td_taper)
1196 ydata = self.get_ydata().astype(float)
1197 if demean:
1198 ydata -= ydata.mean()
1200 spec = num.fft.rfft(ydata, ntrans)
1202 amp = num.abs(spec)
1203 nspec = amp.size
1204 csamp = num.cumsum(amp)
1205 amp_smoothed = num.empty(nspec, dtype=csamp.dtype)
1206 n1, n2 = nw//2, nw//2 + nspec - nw
1207 amp_smoothed[n1:n2] = (csamp[nw:] - csamp[:-nw]) / nw
1208 amp_smoothed[:n1] = amp_smoothed[n1]
1209 amp_smoothed[n2:] = amp_smoothed[n2-1]
1211 denom = amp_smoothed * amp
1212 numer = amp
1213 eps = num.mean(denom) * 1e-9
1214 if eps == 0.0:
1215 eps = 1e-9
1217 numer += eps
1218 denom += eps
1219 spec *= numer/denom
1221 if fd_taper:
1222 fd_taper(spec, 0., df)
1224 ydata = num.fft.irfft(spec)
1225 self.set_ydata(ydata[:ndata])
1227 def _get_cached_freqs(self, nf, deltaf):
1228 ck = (nf, deltaf)
1229 if ck not in Trace.cached_frequencies:
1230 Trace.cached_frequencies[ck] = deltaf * num.arange(
1231 nf, dtype=float)
1233 return Trace.cached_frequencies[ck]
1235 def bandpass_fft(self, corner_hp, corner_lp):
1236 '''
1237 Apply boxcar bandbpass to trace (in spectral domain).
1238 '''
1240 n = len(self.ydata)
1241 n2 = nextpow2(n)
1242 data = num.zeros(n2, dtype=num.float64)
1243 data[:n] = self.ydata
1244 fdata = num.fft.rfft(data)
1245 freqs = self._get_cached_freqs(len(fdata), 1./(self.deltat*n2))
1246 fdata[0] = 0.0
1247 fdata *= num.logical_and(corner_hp < freqs, freqs < corner_lp)
1248 data = num.fft.irfft(fdata)
1249 self.drop_growbuffer()
1250 self.ydata = data[:n]
1252 def shift(self, tshift):
1253 '''
1254 Time shift the trace.
1255 '''
1257 self.tmin += tshift
1258 self.tmax += tshift
1259 self._update_ids()
1261 def snap(self, inplace=True, interpolate=False):
1262 '''
1263 Shift trace samples to nearest even multiples of the sampling rate.
1265 :param inplace: (boolean) snap traces inplace
1267 If ``inplace`` is ``False`` and the difference of tmin and tmax of
1268 both, the snapped and the original trace is smaller than 0.01 x deltat,
1269 :py:func:`snap` returns the unsnapped instance of the original trace.
1270 '''
1272 tmin = round(self.tmin/self.deltat)*self.deltat
1273 tmax = tmin + (self.ydata.size-1)*self.deltat
1275 if inplace:
1276 xself = self
1277 else:
1278 if abs(self.tmin - tmin) < 1e-2*self.deltat and \
1279 abs(self.tmax - tmax) < 1e-2*self.deltat:
1280 return self
1282 xself = self.copy()
1284 if interpolate:
1285 n = xself.data_len()
1286 ydata_new = num.empty(n, dtype=float)
1287 i_control = num.array([0, n-1], dtype=num.int64)
1288 tref = tmin
1289 t_control = num.array(
1290 [float(xself.tmin-tref), float(xself.tmax-tref)],
1291 dtype=float)
1293 signal_ext.antidrift(i_control, t_control,
1294 xself.ydata.astype(float),
1295 float(tmin-tref), xself.deltat, ydata_new)
1297 xself.ydata = ydata_new
1299 xself.tmin = tmin
1300 xself.tmax = tmax
1301 xself._update_ids()
1303 return xself
1305 def fix_deltat_rounding_errors(self):
1306 '''
1307 Try to undo sampling rate rounding errors.
1309 See :py:func:`fix_deltat_rounding_errors`.
1310 '''
1312 self.deltat = fix_deltat_rounding_errors(self.deltat)
1313 self.tmax = self.tmin + (self.data_len() - 1) * self.deltat
1315 def sta_lta_centered(self, tshort, tlong, quad=True, scalingmethod=1):
1316 '''
1317 Run special STA/LTA filter where the short time window is centered on
1318 the long time window.
1320 :param tshort: length of short time window in [s]
1321 :param tlong: length of long time window in [s]
1322 :param quad: whether to square the data prior to applying the STA/LTA
1323 filter
1324 :param scalingmethod: integer key to select how output values are
1325 scaled / normalized (``1``, ``2``, or ``3``)
1327 ============= ====================================== ===========
1328 Scalingmethod Implementation Range
1329 ============= ====================================== ===========
1330 ``1`` As/Al* Ts/Tl [0,1]
1331 ``2`` (As/Al - 1) / (Tl/Ts - 1) [-Ts/Tl,1]
1332 ``3`` Like ``2`` but clipping range at zero [0,1]
1333 ============= ====================================== ===========
1335 '''
1337 nshort = int(round(tshort/self.deltat))
1338 nlong = int(round(tlong/self.deltat))
1340 assert nshort < nlong
1341 if nlong > len(self.ydata):
1342 raise TraceTooShort(
1343 'Samples in trace: %s, samples needed: %s'
1344 % (len(self.ydata), nlong))
1346 if quad:
1347 sqrdata = self.ydata**2
1348 else:
1349 sqrdata = self.ydata
1351 mavg_short = moving_avg(sqrdata, nshort)
1352 mavg_long = moving_avg(sqrdata, nlong)
1354 self.drop_growbuffer()
1356 if scalingmethod not in (1, 2, 3):
1357 raise Exception('Invalid argument to scalingrange argument.')
1359 if scalingmethod == 1:
1360 self.ydata = mavg_short/mavg_long * float(nshort)/float(nlong)
1361 elif scalingmethod in (2, 3):
1362 self.ydata = (mavg_short/mavg_long - 1.) \
1363 / ((float(nlong)/float(nshort)) - 1)
1365 if scalingmethod == 3:
1366 self.ydata = num.maximum(self.ydata, 0.)
1368 def sta_lta_right(self, tshort, tlong, quad=True, scalingmethod=1):
1369 '''
1370 Run special STA/LTA filter where the short time window is overlapping
1371 with the last part of the long time window.
1373 :param tshort: length of short time window in [s]
1374 :param tlong: length of long time window in [s]
1375 :param quad: whether to square the data prior to applying the STA/LTA
1376 filter
1377 :param scalingmethod: integer key to select how output values are
1378 scaled / normalized (``1``, ``2``, or ``3``)
1380 ============= ====================================== ===========
1381 Scalingmethod Implementation Range
1382 ============= ====================================== ===========
1383 ``1`` As/Al* Ts/Tl [0,1]
1384 ``2`` (As/Al - 1) / (Tl/Ts - 1) [-Ts/Tl,1]
1385 ``3`` Like ``2`` but clipping range at zero [0,1]
1386 ============= ====================================== ===========
1388 With ``scalingmethod=1``, the values produced by this variant of the
1389 STA/LTA are equivalent to
1391 .. math::
1392 s_i = \\frac{s}{l} \\frac{\\frac{1}{s}\\sum_{j=i}^{i+s-1} f_j}
1393 {\\frac{1}{l}\\sum_{j=i+s-l}^{i+s-1} f_j}
1395 where :math:`f_j` are the input samples, :math:`s` are the number of
1396 samples in the short time window and :math:`l` are the number of
1397 samples in the long time window.
1398 '''
1400 n = self.data_len()
1401 tmin = self.tmin
1403 nshort = max(1, int(round(tshort/self.deltat)))
1404 nlong = max(1, int(round(tlong/self.deltat)))
1406 assert nshort < nlong
1408 if nlong > len(self.ydata):
1409 raise TraceTooShort(
1410 'Samples in trace: %s, samples needed: %s'
1411 % (len(self.ydata), nlong))
1413 if quad:
1414 sqrdata = self.ydata**2
1415 else:
1416 sqrdata = self.ydata
1418 nshift = int(math.floor(0.5 * (nlong - nshort)))
1419 if nlong % 2 != 0 and nshort % 2 == 0:
1420 nshift += 1
1422 mavg_short = moving_avg(sqrdata, nshort)[nshift:]
1423 mavg_long = moving_avg(sqrdata, nlong)[:sqrdata.size-nshift]
1425 self.drop_growbuffer()
1427 if scalingmethod not in (1, 2, 3):
1428 raise Exception('Invalid argument to scalingrange argument.')
1430 if scalingmethod == 1:
1431 ydata = mavg_short/mavg_long * nshort/nlong
1432 elif scalingmethod in (2, 3):
1433 ydata = (mavg_short/mavg_long - 1.) \
1434 / ((float(nlong)/float(nshort)) - 1)
1436 if scalingmethod == 3:
1437 ydata = num.maximum(self.ydata, 0.)
1439 self.set_ydata(ydata)
1441 self.shift((math.ceil(0.5*nlong) - nshort + 1) * self.deltat)
1443 self.chop(
1444 tmin + (nlong - nshort) * self.deltat,
1445 tmin + (n - nshort) * self.deltat)
1447 def peaks(self, threshold, tsearch,
1448 deadtime=False,
1449 nblock_duration_detection=100):
1451 '''
1452 Detect peaks above a given threshold (method 1).
1454 From every instant, where the signal rises above ``threshold``, a time
1455 length of ``tsearch`` seconds is searched for a maximum. A list with
1456 tuples (time, value) for each detected peak is returned. The
1457 ``deadtime`` argument turns on a special deadtime duration detection
1458 algorithm useful in combination with recursive STA/LTA filters.
1459 '''
1461 y = self.ydata
1462 above = num.where(y > threshold, 1, 0)
1463 deriv = num.zeros(y.size, dtype=num.int8)
1464 deriv[1:] = above[1:]-above[:-1]
1465 itrig_positions = num.nonzero(deriv > 0)[0]
1466 tpeaks = []
1467 apeaks = []
1468 tzeros = []
1469 tzero = self.tmin
1471 for itrig_pos in itrig_positions:
1472 ibeg = itrig_pos
1473 iend = min(
1474 len(self.ydata),
1475 itrig_pos + int(math.ceil(tsearch/self.deltat)))
1476 ipeak = num.argmax(y[ibeg:iend])
1477 tpeak = self.tmin + (ipeak+ibeg)*self.deltat
1478 apeak = y[ibeg+ipeak]
1480 if tpeak < tzero:
1481 continue
1483 if deadtime:
1484 ibeg = itrig_pos
1485 iblock = 0
1486 nblock = nblock_duration_detection
1487 totalsum = 0.
1488 while True:
1489 if ibeg+iblock*nblock >= len(y):
1490 tzero = self.tmin + (len(y)-1) * self.deltat
1491 break
1493 logy = num.log(
1494 y[ibeg+iblock*nblock:ibeg+(iblock+1)*nblock])
1495 logy[0] += totalsum
1496 ysum = num.cumsum(logy)
1497 totalsum = ysum[-1]
1498 below = num.where(ysum <= 0., 1, 0)
1499 deriv = num.zeros(ysum.size, dtype=num.int8)
1500 deriv[1:] = below[1:]-below[:-1]
1501 izero_positions = num.nonzero(deriv > 0)[0] + iblock*nblock
1502 if len(izero_positions) > 0:
1503 tzero = self.tmin + self.deltat * (
1504 ibeg + izero_positions[0])
1505 break
1506 iblock += 1
1507 else:
1508 tzero = ibeg*self.deltat + self.tmin + tsearch
1510 tpeaks.append(tpeak)
1511 apeaks.append(apeak)
1512 tzeros.append(tzero)
1514 if deadtime:
1515 return tpeaks, apeaks, tzeros
1516 else:
1517 return tpeaks, apeaks
1519 def peaks2(self, threshold, tsearch):
1521 '''
1522 Detect peaks above a given threshold (method 2).
1524 This variant of peak detection is a bit more robust (and slower) than
1525 the one implemented in :py:meth:`Trace.peaks`. First all samples with
1526 ``a[i-1] < a[i] > a[i+1]`` are masked as potential peaks. From these,
1527 iteratively the one with the maximum amplitude ``a[j]`` and time
1528 ``t[j]`` is choosen and potential peaks within
1529 ``t[j] - tsearch, t[j] + tsearch``
1530 are discarded. The algorithm stops, when ``a[j] < threshold`` or when
1531 no more potential peaks are left.
1532 '''
1534 a = num.copy(self.ydata)
1536 amin = num.min(a)
1538 a[0] = amin
1539 a[1: -1][num.diff(a, 2) <= 0.] = amin
1540 a[-1] = amin
1542 data = []
1543 while True:
1544 imax = num.argmax(a)
1545 amax = a[imax]
1547 if amax < threshold or amax == amin:
1548 break
1550 data.append((self.tmin + imax * self.deltat, amax))
1552 ntsearch = int(round(tsearch / self.deltat))
1553 a[max(imax-ntsearch//2, 0):min(imax+ntsearch//2, a.size)] = amin
1555 if data:
1556 data.sort()
1557 tpeaks, apeaks = list(zip(*data))
1558 else:
1559 tpeaks, apeaks = [], []
1561 return tpeaks, apeaks
1563 def extend(self, tmin=None, tmax=None, fillmethod='zeros'):
1564 '''
1565 Extend trace to given span.
1567 :param tmin: begin time of new span
1568 :param tmax: end time of new span
1569 :param fillmethod: ``'zeros'``, ``'repeat'``, ``'mean'``, or
1570 ``'median'``
1571 '''
1573 nold = self.ydata.size
1575 if tmin is not None:
1576 nl = min(0, int(round((tmin-self.tmin)/self.deltat)))
1577 else:
1578 nl = 0
1580 if tmax is not None:
1581 nh = max(nold - 1, int(round((tmax-self.tmin)/self.deltat)))
1582 else:
1583 nh = nold - 1
1585 n = nh - nl + 1
1586 data = num.zeros(n, dtype=self.ydata.dtype)
1587 data[-nl:-nl + nold] = self.ydata
1588 if self.ydata.size >= 1:
1589 if fillmethod == 'repeat':
1590 data[:-nl] = self.ydata[0]
1591 data[-nl + nold:] = self.ydata[-1]
1592 elif fillmethod == 'median':
1593 v = num.median(self.ydata)
1594 data[:-nl] = v
1595 data[-nl + nold:] = v
1596 elif fillmethod == 'mean':
1597 v = num.mean(self.ydata)
1598 data[:-nl] = v
1599 data[-nl + nold:] = v
1601 self.drop_growbuffer()
1602 self.ydata = data
1604 self.tmin += nl * self.deltat
1605 self.tmax = self.tmin + (self.ydata.size - 1) * self.deltat
1607 self._update_ids()
1609 def transfer(self,
1610 tfade=0.,
1611 freqlimits=None,
1612 transfer_function=None,
1613 cut_off_fading=True,
1614 demean=True,
1615 invert=False):
1617 '''
1618 Return new trace with transfer function applied (convolution).
1620 :param tfade: rise/fall time in seconds of taper applied in timedomain
1621 at both ends of trace.
1622 :param freqlimits: 4-tuple with corner frequencies in Hz.
1623 :param transfer_function: FrequencyResponse object; must provide a
1624 method 'evaluate(freqs)', which returns the transfer function
1625 coefficients at the frequencies 'freqs'.
1626 :param cut_off_fading: whether to cut off rise/fall interval in output
1627 trace.
1628 :param demean: remove mean before applying transfer function
1629 :param invert: set to True to do a deconvolution
1630 '''
1632 if transfer_function is None:
1633 transfer_function = g_one_response
1635 if self.tmax - self.tmin <= tfade*2.:
1636 raise TraceTooShort(
1637 'Trace %s.%s.%s.%s too short for fading length setting. '
1638 'trace length = %g, fading length = %g'
1639 % (self.nslc_id + (self.tmax-self.tmin, tfade)))
1641 if freqlimits is None and (
1642 transfer_function is None or transfer_function.is_scalar()):
1644 # special case for flat responses
1646 output = self.copy()
1647 data = self.ydata
1648 ndata = data.size
1650 if transfer_function is not None:
1651 c = num.abs(transfer_function.evaluate(num.ones(1))[0])
1653 if invert:
1654 c = 1.0/c
1656 data *= c
1658 if tfade != 0.0:
1659 data *= costaper(
1660 0., tfade, self.deltat*(ndata-1)-tfade, self.deltat*ndata,
1661 ndata, self.deltat)
1663 output.ydata = data
1665 else:
1666 ndata = self.ydata.size
1667 ntrans = nextpow2(ndata*1.2)
1668 coeffs = self._get_tapered_coeffs(
1669 ntrans, freqlimits, transfer_function, invert=invert,
1670 demean=demean)
1672 data = self.ydata
1674 data_pad = num.zeros(ntrans, dtype=float)
1675 data_pad[:ndata] = data
1676 if demean:
1677 data_pad[:ndata] -= data.mean()
1679 if tfade != 0.0:
1680 data_pad[:ndata] *= costaper(
1681 0., tfade, self.deltat*(ndata-1)-tfade, self.deltat*ndata,
1682 ndata, self.deltat)
1684 fdata = num.fft.rfft(data_pad)
1685 fdata *= coeffs
1686 ddata = num.fft.irfft(fdata)
1687 output = self.copy()
1688 output.ydata = ddata[:ndata]
1690 if cut_off_fading and tfade != 0.0:
1691 try:
1692 output.chop(output.tmin+tfade, output.tmax-tfade, inplace=True)
1693 except NoData:
1694 raise TraceTooShort(
1695 'Trace %s.%s.%s.%s too short for fading length setting. '
1696 'trace length = %g, fading length = %g'
1697 % (self.nslc_id + (self.tmax-self.tmin, tfade)))
1698 else:
1699 output.ydata = output.ydata.copy()
1701 return output
1703 def differentiate(self, n=1, order=4, inplace=True):
1704 '''
1705 Approximate first or second derivative of the trace.
1707 :param n: 1 for first derivative, 2 for second
1708 :param order: order of the approximation 2 and 4 are supported
1709 :param inplace: if ``True`` the trace is differentiated in place,
1710 otherwise a new trace object with the derivative is returned.
1712 Raises :py:exc:`ValueError` for unsupported `n` or `order`.
1714 See :py:func:`~pyrocko.util.diff_fd` for implementation details.
1715 '''
1717 ddata = util.diff_fd(n, order, self.deltat, self.ydata)
1719 if inplace:
1720 self.ydata = ddata
1721 else:
1722 output = self.copy(data=False)
1723 output.set_ydata(ddata)
1724 return output
1726 def drop_chain_cache(self):
1727 if self._pchain:
1728 self._pchain.clear()
1730 def init_chain(self):
1731 self._pchain = pchain.Chain(
1732 do_downsample,
1733 do_extend,
1734 do_pre_taper,
1735 do_fft,
1736 do_filter,
1737 do_ifft)
1739 def run_chain(self, tmin, tmax, deltat, setup, nocache):
1740 if setup.domain == 'frequency_domain':
1741 _, _, data = self._pchain(
1742 (self, deltat),
1743 (tmin, tmax),
1744 (setup.taper,),
1745 (setup.filter,),
1746 (setup.filter,),
1747 nocache=nocache)
1749 return num.abs(data), num.abs(data)
1751 else:
1752 processed = self._pchain(
1753 (self, deltat),
1754 (tmin, tmax),
1755 (setup.taper,),
1756 (setup.filter,),
1757 (setup.filter,),
1758 (),
1759 nocache=nocache)
1761 if setup.domain == 'time_domain':
1762 data = processed.get_ydata()
1764 elif setup.domain == 'envelope':
1765 processed = processed.envelope(inplace=False)
1767 elif setup.domain == 'absolute':
1768 processed.set_ydata(num.abs(processed.get_ydata()))
1770 return processed.get_ydata(), processed
1772 def misfit(self, candidate, setup, nocache=False, debug=False):
1773 '''
1774 Calculate misfit and normalization factor against candidate trace.
1776 :param candidate: :py:class:`Trace` object
1777 :param setup: :py:class:`MisfitSetup` object
1778 :returns: tuple ``(m, n)``, where m is the misfit value and n is the
1779 normalization divisor
1781 If the sampling rates of ``self`` and ``candidate`` differ, the trace
1782 with the higher sampling rate will be downsampled.
1783 '''
1785 a = self
1786 b = candidate
1788 for tr in (a, b):
1789 if not tr._pchain:
1790 tr.init_chain()
1792 deltat = max(a.deltat, b.deltat)
1793 tmin = min(a.tmin, b.tmin) - deltat
1794 tmax = max(a.tmax, b.tmax) + deltat
1796 adata, aproc = a.run_chain(tmin, tmax, deltat, setup, nocache)
1797 bdata, bproc = b.run_chain(tmin, tmax, deltat, setup, nocache)
1799 if setup.domain != 'cc_max_norm':
1800 m, n = Lx_norm(bdata, adata, norm=setup.norm)
1801 else:
1802 ctr = correlate(aproc, bproc, mode='full', normalization='normal')
1803 ccmax = ctr.max()[1]
1804 m = 0.5 - 0.5 * ccmax
1805 n = 0.5
1807 if debug:
1808 return m, n, aproc, bproc
1809 else:
1810 return m, n
1812 def spectrum(self, pad_to_pow2=False, tfade=None):
1813 '''
1814 Get FFT spectrum of trace.
1816 :param pad_to_pow2: whether to zero-pad the data to next larger
1817 power-of-two length
1818 :param tfade: ``None`` or a time length in seconds, to apply cosine
1819 shaped tapers to both
1821 :returns: a tuple with (frequencies, values)
1822 '''
1824 ndata = self.ydata.size
1826 if pad_to_pow2:
1827 ntrans = nextpow2(ndata)
1828 else:
1829 ntrans = ndata
1831 if tfade is None:
1832 ydata = self.ydata
1833 else:
1834 ydata = self.ydata * costaper(
1835 0., tfade, self.deltat*(ndata-1)-tfade, self.deltat*ndata,
1836 ndata, self.deltat)
1838 fydata = num.fft.rfft(ydata, ntrans)
1839 df = 1./(ntrans*self.deltat)
1840 fxdata = num.arange(len(fydata))*df
1841 return fxdata, fydata
1843 def multi_filter(self, filter_freqs, bandwidth):
1845 class Gauss(FrequencyResponse):
1846 f0 = Float.T()
1847 a = Float.T(default=1.0)
1849 def __init__(self, f0, a=1.0, **kwargs):
1850 FrequencyResponse.__init__(self, f0=f0, a=a, **kwargs)
1852 def evaluate(self, freqs):
1853 omega0 = 2.*math.pi*self.f0
1854 omega = 2.*math.pi*freqs
1855 return num.exp(-((omega-omega0)
1856 / (self.a*omega0))**2)
1858 freqs, coeffs = self.spectrum()
1859 n = self.data_len()
1860 nfilt = len(filter_freqs)
1861 signal_tf = num.zeros((nfilt, n))
1862 centroid_freqs = num.zeros(nfilt)
1863 for ifilt, f0 in enumerate(filter_freqs):
1864 taper = Gauss(f0, a=bandwidth)
1865 weights = taper.evaluate(freqs)
1866 nhalf = freqs.size
1867 analytic_spec = num.zeros(n, dtype=complex)
1868 analytic_spec[:nhalf] = coeffs*weights
1870 enorm = num.abs(analytic_spec[:nhalf])**2
1871 enorm /= num.sum(enorm)
1873 if n % 2 == 0:
1874 analytic_spec[1:nhalf-1] *= 2.
1875 else:
1876 analytic_spec[1:nhalf] *= 2.
1878 analytic = num.fft.ifft(analytic_spec)
1879 signal_tf[ifilt, :] = num.abs(analytic)
1881 enorm = num.abs(analytic_spec[:nhalf])**2
1882 enorm /= num.sum(enorm)
1883 centroid_freqs[ifilt] = num.sum(freqs*enorm)
1885 return centroid_freqs, signal_tf
1887 def _get_tapered_coeffs(
1888 self, ntrans, freqlimits, transfer_function, invert=False,
1889 demean=True):
1891 cache_key = (
1892 ntrans, self.deltat, freqlimits, transfer_function.uuid, invert,
1893 demean)
1895 if cache_key in g_tapered_coeffs_cache:
1896 return g_tapered_coeffs_cache[cache_key]
1898 deltaf = 1./(self.deltat*ntrans)
1899 nfreqs = ntrans//2 + 1
1900 transfer = num.ones(nfreqs, dtype=complex)
1901 hi = snapper(nfreqs, deltaf)
1902 if freqlimits is not None:
1903 kmin, kmax = hi(freqlimits[0]), hi(freqlimits[3])
1904 freqs = num.arange(kmin, kmax)*deltaf
1905 coeffs = transfer_function.evaluate(freqs)
1906 if invert:
1907 if num.any(coeffs == 0.0):
1908 raise InfiniteResponse('%s.%s.%s.%s' % self.nslc_id)
1910 transfer[kmin:kmax] = 1.0 / coeffs
1911 else:
1912 transfer[kmin:kmax] = coeffs
1914 tapered_transfer = costaper(*freqlimits, nfreqs, deltaf) * transfer
1915 else:
1916 if invert:
1917 raise Exception(
1918 'transfer: `freqlimits` must be given when `invert` is '
1919 'set to `True`')
1921 freqs = num.arange(nfreqs) * deltaf
1922 tapered_transfer = transfer_function.evaluate(freqs)
1924 g_tapered_coeffs_cache[cache_key] = tapered_transfer
1926 if demean:
1927 tapered_transfer[0] = 0.0 # don't introduce static offsets
1929 return tapered_transfer
1931 def fill_template(self, template, **additional):
1932 '''
1933 Fill string template with trace metadata.
1935 Uses normal python '%(placeholder)s' string templates. The following
1936 placeholders are considered: ``network``, ``station``, ``location``,
1937 ``channel``, ``tmin`` (time of first sample), ``tmax`` (time of last
1938 sample), ``tmin_ms``, ``tmax_ms``, ``tmin_us``, ``tmax_us``,
1939 ``tmin_year``, ``tmax_year``, ``tmin_month``, ``tmax_month``,
1940 ``tmin_day``, ``tmax_day``, ``julianday``. The variants ending with
1941 ``'_ms'`` include milliseconds, those with ``'_us'`` include
1942 microseconds, those with ``'_year'`` contain only the year.
1943 '''
1945 template = template.replace('%n', '%(network)s')\
1946 .replace('%s', '%(station)s')\
1947 .replace('%l', '%(location)s')\
1948 .replace('%c', '%(channel)s')\
1949 .replace('%b', '%(tmin)s')\
1950 .replace('%e', '%(tmax)s')\
1951 .replace('%j', '%(julianday)s')
1953 params = dict(
1954 zip(('network', 'station', 'location', 'channel'), self.nslc_id))
1955 params['tmin'] = util.time_to_str(
1956 self.tmin, format='%Y-%m-%d_%H-%M-%S')
1957 params['tmax'] = util.time_to_str(
1958 self.tmax, format='%Y-%m-%d_%H-%M-%S')
1959 params['tmin_ms'] = util.time_to_str(
1960 self.tmin, format='%Y-%m-%d_%H-%M-%S.3FRAC')
1961 params['tmax_ms'] = util.time_to_str(
1962 self.tmax, format='%Y-%m-%d_%H-%M-%S.3FRAC')
1963 params['tmin_us'] = util.time_to_str(
1964 self.tmin, format='%Y-%m-%d_%H-%M-%S.6FRAC')
1965 params['tmax_us'] = util.time_to_str(
1966 self.tmax, format='%Y-%m-%d_%H-%M-%S.6FRAC')
1967 params['tmin_year'], params['tmin_month'], params['tmin_day'] \
1968 = util.time_to_str(self.tmin, format='%Y-%m-%d').split('-')
1969 params['tmax_year'], params['tmax_month'], params['tmax_day'] \
1970 = util.time_to_str(self.tmax, format='%Y-%m-%d').split('-')
1971 params['julianday'] = util.julian_day_of_year(self.tmin)
1972 params.update(additional)
1973 return template % params
1975 def plot(self):
1976 '''
1977 Show trace with matplotlib.
1979 See also: :py:meth:`Trace.snuffle`.
1980 '''
1982 import pylab
1983 pylab.plot(self.get_xdata(), self.get_ydata())
1984 name = '%s %s %s - %s' % (
1985 self.channel,
1986 self.station,
1987 time.strftime('%d-%m-%y %H:%M:%S', time.gmtime(self.tmin)),
1988 time.strftime('%d-%m-%y %H:%M:%S', time.gmtime(self.tmax)))
1990 pylab.title(name)
1991 pylab.show()
1993 def snuffle(self, **kwargs):
1994 '''
1995 Show trace in a snuffler window.
1997 :param stations: list of :py:class:`pyrocko.model.station.Station`
1998 objects or ``None``
1999 :param events: list of :py:class:`pyrocko.model.event.Event` objects or
2000 ``None``
2001 :param markers: list of :py:class:`pyrocko.gui.snuffler.marker.Marker`
2002 objects or ``None``
2003 :param ntracks: float, number of tracks to be shown initially (default:
2004 12)
2005 :param follow: time interval (in seconds) for real time follow mode or
2006 ``None``
2007 :param controls: bool, whether to show the main controls (default:
2008 ``True``)
2009 :param opengl: bool, whether to use opengl (default: ``False``)
2010 '''
2012 return snuffle([self], **kwargs)
2015def snuffle(traces, **kwargs):
2016 '''
2017 Show traces in a snuffler window.
2019 :param stations: list of :py:class:`pyrocko.model.station.Station` objects
2020 or ``None``
2021 :param events: list of :py:class:`pyrocko.model.event.Event` objects or
2022 ``None``
2023 :param markers: list of :py:class:`pyrocko.gui.snuffler.marker.Marker`
2024 objects or ``None``
2025 :param ntracks: int, number of tracks to be shown initially (default: 12)
2026 :param follow: time interval (in seconds) for real time follow mode or
2027 ``None``
2028 :param controls: bool, whether to show the main controls (default:
2029 ``True``)
2030 :param opengl: bool, whether to use opengl (default: ``False``)
2031 '''
2033 from pyrocko import pile
2034 from pyrocko.gui.snuffler import snuffler
2035 p = pile.Pile()
2036 if traces:
2037 trf = pile.MemTracesFile(None, traces)
2038 p.add_file(trf)
2039 return snuffler.snuffle(p, **kwargs)
2042def downsample_tpad(
2043 deltat_in, deltat_out, allow_upsample_max=1, ftype='fir-remez'):
2044 '''
2045 Get approximate amount of cutoff which will be produced by downsampling.
2047 The :py:meth:`Trace.downsample_to` method removes some samples at the
2048 beginning and end of the trace which is downsampled. This function
2049 estimates the approximate length [s] which will be cut off for a given set
2050 of parameters supplied to :py:meth:`Trace.downsample_to`.
2052 :param deltat_in:
2053 Input sampling interval [s].
2054 :type deltat_in:
2055 float
2057 :param deltat_out:
2058 Output samling interval [s].
2059 :type deltat_out:
2060 float
2062 :returns:
2063 Approximate length [s] which will be cut off.
2065 See :py:meth:`Trace.downsample_to` for details.
2066 '''
2068 upsratio, deci_seq = _configure_downsampling(
2069 deltat_in, deltat_out, allow_upsample_max)
2071 tpad = 0.0
2072 deltat = deltat_in / upsratio
2073 for deci in deci_seq:
2074 b, a, n = util.decimate_coeffs(deci, None, ftype)
2075 # n//2 for the antialiasing
2076 # +deci for possible snap to multiples
2077 # +1 for rounding errors
2078 tpad += (n//2 + deci + 1) * deltat
2079 deltat = deltat * deci
2081 return tpad
2084def _configure_downsampling(deltat_in, deltat_out, allow_upsample_max):
2085 for upsratio in range(1, allow_upsample_max+1):
2086 dratio = (upsratio/deltat_in) / (1./deltat_out)
2087 deci_seq = util.decitab(int(round(dratio)))
2088 if abs(dratio - round(dratio)) / dratio < 0.0001 and deci_seq:
2089 return upsratio, [deci for deci in deci_seq if deci != 1]
2091 raise util.UnavailableDecimation('ratio = %g' % (deltat_out / deltat_in))
2094def _all_same(xs):
2095 return all(x == xs[0] for x in xs)
2098def _incompatibilities(traces):
2099 if not traces:
2100 return None
2102 params = [
2103 (tr.ydata.size, tr.ydata.dtype, tr.deltat, tr.tmin)
2104 for tr in traces]
2106 if not _all_same(params):
2107 return params
2108 else:
2109 return None
2112def _raise_incompatible_traces(params):
2113 raise IncompatibleTraces(
2114 'Given traces are incompatible. Sampling rate, start time, '
2115 'number of samples and data type must match.\n%s\n%s' % (
2116 ' %10s %-10s %12s %22s' % (
2117 'nsamples', 'dtype', 'deltat', 'tmin'),
2118 '\n'.join(
2119 ' %10i %-10s %12.5e %22s' % (
2120 nsamples, dtype, deltat, util.time_to_str(tmin))
2121 for (nsamples, dtype, deltat, tmin) in params)))
2124def _ensure_compatible(traces):
2125 params = _incompatibilities(traces)
2126 if params:
2127 _raise_incompatible_traces(params)
2130def _almost_equal(a, b, atol):
2131 return abs(a-b) < atol
2134def get_traces_data_as_array(traces):
2135 '''
2136 Merge data samples from multiple traces into a 2D array.
2138 :param traces:
2139 Input waveforms.
2140 :type traces:
2141 list of :py:class:`pyrocko.Trace <pyrocko.trace.Trace>` objects
2143 :raises:
2144 :py:class:`IncompatibleTraces` if traces have different time
2145 span, sample rate or data type, or if traces is an empty list.
2147 :returns:
2148 2D array as ``data[itrace, isample]``.
2149 :rtype:
2150 :py:class:`numpy.ndarray`
2151 '''
2153 if not traces:
2154 raise IncompatibleTraces('Need at least one trace.')
2156 _ensure_compatible(traces)
2158 return num.vstack([tr.ydata for tr in traces])
2161def make_traces_compatible(
2162 traces,
2163 dtype=None,
2164 deltat=None,
2165 enforce_global_snap=True,
2166 warn_snap=False):
2168 '''
2169 Homogenize sampling rate, time span, sampling instants, and data type.
2171 This function takes a group of traces and tries to make them compatible in
2172 terms of data type and sampling rate, time span, and sampling instants of
2173 time.
2175 If necessary, traces are (in order):
2177 - casted to the same data type.
2178 - downsampled to a common sampling rate, using decimation cascades.
2179 - resampled to common sampling instants in time, using Sinc interpolation.
2180 - cut to the same time span. The longest time span covered by all traces is
2181 used.
2183 :param traces:
2184 Input waveforms.
2185 :type traces:
2186 :py:class:`list` of :py:class:`Trace`
2188 :param dtype:
2189 Force traces to be casted to the given data type. If not specified, the
2190 traces are cast to :py:class:`float`.
2191 :type dtype:
2192 :py:class:`numpy.dtype`
2194 :param deltat:
2195 Sampling interval [s]. If not specified, the longest sampling interval
2196 among the input traces is chosen.
2197 :type deltat:
2198 float
2200 :param enforce_global_snap:
2201 If ``True``, choose sampling instants to be even multiples of the
2202 sampling rate in system time. When set to ``False`` traces are still
2203 resampled to common time instants (if necessary), but they may be
2204 offset to the system time sampling rate multiples.
2205 :type enforce_global_snap:
2206 bool
2208 :param warn_snap:
2209 If set to ``True`` warn, when resampling has to be performed.
2210 :type warn_snap:
2211 bool
2212 '''
2214 eps_snap = 1e-3
2216 if not traces:
2217 return []
2219 traces = list(traces)
2221 dtypes = [tr.ydata.dtype for tr in traces]
2222 if not _all_same(dtypes) or dtype is not None:
2224 if dtype is None:
2225 dtype = float
2226 logger.warning(
2227 'make_traces_compatible: Inconsistent data types - converting '
2228 'sample datatype to %s.' % str(dtype))
2230 for itr, tr in enumerate(traces):
2231 tr_copy = tr.copy(data=False)
2232 tr_copy.set_ydata(tr.ydata.astype(dtype))
2233 traces[itr] = tr_copy
2235 deltats = [tr.deltat for tr in traces]
2236 if not _all_same(deltats) or deltat is not None:
2237 if deltat is None:
2238 deltat = max(deltats)
2239 logger.warning(
2240 'make_traces_compatible: Inconsistent sampling rates - '
2241 'downsampling to lowest rate among input traces: %g Hz.'
2242 % (1.0 / deltat))
2244 for itr, tr in enumerate(traces):
2245 if tr.deltat != deltat:
2246 tr_copy = tr.copy()
2247 tr_copy.downsample_to(deltat, snap=True, cut=True)
2248 traces[itr] = tr_copy
2250 tmins = num.array([tr.tmin for tr in traces])
2251 is_aligned = num.abs(num.round(tmins / deltat) * deltat - tmins) \
2252 > deltat * eps_snap
2254 if enforce_global_snap or any(is_aligned):
2255 tref = util.to_time_float(0.0)
2256 else:
2257 # to keep a common subsample shift
2258 tref = num.max(tmins)
2260 tmins_snap = num.round((tmins - tref) / deltat) * deltat + tref
2261 need_snap = num.abs(tmins_snap - tmins) > deltat * eps_snap
2262 if num.any(need_snap):
2263 if warn_snap:
2264 logger.warning(
2265 'make_traces_compatible: Misaligned sampling - introducing '
2266 'subsample shifts for proper alignment.')
2268 for itr, tr in enumerate(traces):
2269 if need_snap[itr]:
2270 tr_copy = tr.copy()
2271 if tref != 0.0:
2272 tr_copy.shift(-tref)
2274 tr_copy.snap(interpolate=True)
2275 if tref != 0.0:
2276 tr_copy.shift(tref)
2278 traces[itr] = tr_copy
2280 tmins = num.array([tr.tmin for tr in traces])
2281 nsamples = num.array([tr.ydata.size for tr in traces])
2282 tmaxs = tmins + (nsamples - 1) * deltat
2284 tmin = num.max(tmins)
2285 tmax = num.min(tmaxs)
2287 if tmin > tmax:
2288 raise IncompatibleTraces('Traces do not overlap.')
2290 nsamples_must = int(round((tmax - tmin) / deltat)) + 1
2291 for itr, tr in enumerate(traces):
2292 if not (_almost_equal(tr.tmin, tmin, deltat*eps_snap)
2293 and _almost_equal(tr.tmax, tmax, deltat*eps_snap)):
2295 traces[itr] = tr.chop(
2296 tmin, tmax,
2297 inplace=False,
2298 want_incomplete=False,
2299 include_last=True)
2301 xtr = traces[itr]
2302 assert _almost_equal(xtr.tmin, tmin, deltat*eps_snap)
2303 assert int(round((xtr.tmax - xtr.tmin) / deltat)) + 1 == nsamples_must
2304 xtr.tmin = tmin
2305 xtr.tmax = tmax
2306 xtr.deltat = deltat
2307 xtr._update_ids()
2309 return traces
2312class IncompatibleTraces(Exception):
2313 '''
2314 Raised when traces have incompatible sampling rate, time span or data type.
2315 '''
2318class InfiniteResponse(Exception):
2319 '''
2320 This exception is raised by :py:class:`Trace` operations when deconvolution
2321 of a frequency response (instrument response transfer function) would
2322 result in a division by zero.
2323 '''
2326class MisalignedTraces(Exception):
2327 '''
2328 This exception is raised by some :py:class:`Trace` operations when tmin,
2329 tmax or number of samples do not match.
2330 '''
2332 pass
2335class NoData(Exception):
2336 '''
2337 This exception is raised by some :py:class:`Trace` operations when no or
2338 not enough data is available.
2339 '''
2341 pass
2344class AboveNyquist(Exception):
2345 '''
2346 This exception is raised by some :py:class:`Trace` operations when given
2347 frequencies are above the Nyquist frequency.
2348 '''
2350 pass
2353class TraceTooShort(Exception):
2354 '''
2355 This exception is raised by some :py:class:`Trace` operations when the
2356 trace is too short.
2357 '''
2359 pass
2362class ResamplingFailed(Exception):
2363 pass
2366def minmax(traces, key=None, mode='minmax', outer_mode='minmax'):
2368 '''
2369 Get data range given traces grouped by selected pattern.
2371 :param key: a callable which takes as single argument a trace and returns a
2372 key for the grouping of the results. If this is ``None``, the default,
2373 ``lambda tr: (tr.network, tr.station, tr.location, tr.channel)`` is
2374 used.
2375 :param mode: ``'minmax'`` or floating point number. If this is
2376 ``'minmax'``, minimum and maximum of the traces are used, if it is a
2377 number, mean +- standard deviation times ``mode`` is used.
2379 param outer_mode: ``'minmax'`` to use mininum and maximum of the
2380 single-trace ranges, or ``'robust'`` to use the interval to discard 10%
2381 extreme values on either end.
2383 :returns: a dict with the combined data ranges.
2385 Examples::
2387 ranges = minmax(traces, lambda tr: tr.channel)
2388 print ranges['N'] # print min & max of all traces with channel == 'N'
2389 print ranges['E'] # print min & max of all traces with channel == 'E'
2391 ranges = minmax(traces, lambda tr: (tr.network, tr.station))
2392 print ranges['GR', 'HAM3'] # print min & max of all traces with
2393 # network == 'GR' and station == 'HAM3'
2395 ranges = minmax(traces, lambda tr: None)
2396 print ranges[None] # prints min & max of all traces
2397 '''
2399 if key is None:
2400 key = _default_key
2402 ranges = defaultdict(list)
2403 for trace in traces:
2404 if isinstance(mode, str) and mode == 'minmax':
2405 mi, ma = num.nanmin(trace.ydata), num.nanmax(trace.ydata)
2406 else:
2407 mean = trace.ydata.mean()
2408 std = trace.ydata.std()
2409 mi, ma = mean-std*mode, mean+std*mode
2411 k = key(trace)
2412 ranges[k].append((mi, ma))
2414 for k in ranges:
2415 mins, maxs = num.array(ranges[k]).T
2416 if outer_mode == 'minmax':
2417 ranges[k] = num.nanmin(mins), num.nanmax(maxs)
2418 elif outer_mode == 'robust':
2419 ranges[k] = num.percentile(mins, 10.), num.percentile(maxs, 90.)
2421 return ranges
2424def minmaxtime(traces, key=None):
2426 '''
2427 Get time range given traces grouped by selected pattern.
2429 :param key: a callable which takes as single argument a trace and returns a
2430 key for the grouping of the results. If this is ``None``, the default,
2431 ``lambda tr: (tr.network, tr.station, tr.location, tr.channel)`` is
2432 used.
2434 :returns: a dict with the combined data ranges.
2435 '''
2437 if key is None:
2438 key = _default_key
2440 ranges = {}
2441 for trace in traces:
2442 mi, ma = trace.tmin, trace.tmax
2443 k = key(trace)
2444 if k not in ranges:
2445 ranges[k] = mi, ma
2446 else:
2447 tmi, tma = ranges[k]
2448 ranges[k] = min(tmi, mi), max(tma, ma)
2450 return ranges
2453def degapper(
2454 traces,
2455 maxgap=5,
2456 fillmethod='interpolate',
2457 deoverlap='use_second',
2458 maxlap=None):
2460 '''
2461 Try to connect traces and remove gaps.
2463 This method will combine adjacent traces, which match in their network,
2464 station, location and channel attributes. Overlapping parts are handled
2465 according to the ``deoverlap`` argument.
2467 :param traces: input traces, must be sorted by their full_id attribute.
2468 :param maxgap: maximum number of samples to interpolate.
2469 :param fillmethod: what to put into the gaps: 'interpolate' or 'zeros'.
2470 :param deoverlap: how to handle overlaps: 'use_second' to use data from
2471 second trace (default), 'use_first' to use data from first trace,
2472 'crossfade_cos' to crossfade with cosine taper, 'add' to add amplitude
2473 values.
2474 :param maxlap: maximum number of samples of overlap which are removed
2476 :returns: list of traces
2477 '''
2479 in_traces = traces
2480 out_traces = []
2481 if not in_traces:
2482 return out_traces
2483 out_traces.append(in_traces.pop(0))
2484 while in_traces:
2486 a = out_traces[-1]
2487 b = in_traces.pop(0)
2489 avirt, bvirt = a.ydata is None, b.ydata is None
2490 assert avirt == bvirt, \
2491 'traces given to degapper() must either all have data or have ' \
2492 'no data.'
2494 virtual = avirt and bvirt
2496 if (a.nslc_id == b.nslc_id and a.deltat == b.deltat
2497 and a.data_len() >= 1 and b.data_len() >= 1
2498 and (virtual or a.ydata.dtype == b.ydata.dtype)):
2500 dist = (b.tmin-(a.tmin+(a.data_len()-1)*a.deltat))/a.deltat
2501 idist = int(round(dist))
2502 if abs(dist - idist) > 0.05 and idist <= maxgap:
2503 # logger.warning('Cannot degap traces with displaced sampling '
2504 # '(%s, %s, %s, %s)' % a.nslc_id)
2505 pass
2506 else:
2507 if 1 < idist <= maxgap:
2508 if not virtual:
2509 if fillmethod == 'interpolate':
2510 filler = a.ydata[-1] + (
2511 ((1.0 + num.arange(idist-1, dtype=float))
2512 / idist) * (b.ydata[0]-a.ydata[-1])
2513 ).astype(a.ydata.dtype)
2514 elif fillmethod == 'zeros':
2515 filler = num.zeros(idist-1, dtype=a.ydata.dtype)
2516 a.ydata = num.concatenate((a.ydata, filler, b.ydata))
2517 a.tmax = b.tmax
2518 if a.mtime and b.mtime:
2519 a.mtime = max(a.mtime, b.mtime)
2520 continue
2522 elif idist == 1:
2523 if not virtual:
2524 a.ydata = num.concatenate((a.ydata, b.ydata))
2525 a.tmax = b.tmax
2526 if a.mtime and b.mtime:
2527 a.mtime = max(a.mtime, b.mtime)
2528 continue
2530 elif idist <= 0 and (maxlap is None or -maxlap < idist):
2531 if b.tmax > a.tmax:
2532 if not virtual:
2533 na = a.ydata.size
2534 n = -idist+1
2535 if deoverlap == 'use_second':
2536 a.ydata = num.concatenate(
2537 (a.ydata[:-n], b.ydata))
2538 elif deoverlap in ('use_first', 'crossfade_cos'):
2539 a.ydata = num.concatenate(
2540 (a.ydata, b.ydata[n:]))
2541 elif deoverlap == 'add':
2542 a.ydata[-n:] += b.ydata[:n]
2543 a.ydata = num.concatenate(
2544 (a.ydata, b.ydata[n:]))
2545 else:
2546 assert False, 'unknown deoverlap method'
2548 if deoverlap == 'crossfade_cos':
2549 n = -idist+1
2550 taper = 0.5-0.5*num.cos(
2551 (1.+num.arange(n))/(1.+n)*num.pi)
2552 a.ydata[na-n:na] *= 1.-taper
2553 a.ydata[na-n:na] += b.ydata[:n] * taper
2555 a.tmax = b.tmax
2556 if a.mtime and b.mtime:
2557 a.mtime = max(a.mtime, b.mtime)
2558 continue
2559 else:
2560 # make short second trace vanish
2561 continue
2563 if b.data_len() >= 1:
2564 out_traces.append(b)
2566 for tr in out_traces:
2567 tr._update_ids()
2569 return out_traces
2572def rotate(traces, azimuth, in_channels, out_channels):
2573 '''
2574 2D rotation of traces.
2576 :param traces: list of input traces
2577 :param azimuth: difference of the azimuths of the component directions
2578 (azimuth of out_channels[0]) - (azimuth of in_channels[0])
2579 :param in_channels: names of the input channels (e.g. 'N', 'E')
2580 :param out_channels: names of the output channels (e.g. 'R', 'T')
2581 :returns: list of rotated traces
2582 '''
2584 phi = azimuth/180.*math.pi
2585 cphi = math.cos(phi)
2586 sphi = math.sin(phi)
2587 rotated = []
2588 in_channels = tuple(_channels_to_names(in_channels))
2589 out_channels = tuple(_channels_to_names(out_channels))
2590 for a in traces:
2591 for b in traces:
2592 if ((a.channel, b.channel) == in_channels and
2593 a.nslc_id[:3] == b.nslc_id[:3] and
2594 abs(a.deltat-b.deltat) < a.deltat*0.001):
2595 tmin = max(a.tmin, b.tmin)
2596 tmax = min(a.tmax, b.tmax)
2598 if tmin < tmax:
2599 ac = a.chop(tmin, tmax, inplace=False, include_last=True)
2600 bc = b.chop(tmin, tmax, inplace=False, include_last=True)
2601 if abs(ac.tmin - bc.tmin) > ac.deltat*0.01:
2602 logger.warning(
2603 'Cannot rotate traces with displaced sampling '
2604 '(%s, %s, %s, %s)' % a.nslc_id)
2605 continue
2607 acydata = ac.get_ydata()*cphi+bc.get_ydata()*sphi
2608 bcydata = -ac.get_ydata()*sphi+bc.get_ydata()*cphi
2609 ac.set_ydata(acydata)
2610 bc.set_ydata(bcydata)
2612 ac.set_codes(channel=out_channels[0])
2613 bc.set_codes(channel=out_channels[1])
2614 rotated.append(ac)
2615 rotated.append(bc)
2617 return rotated
2620def rotate_to_rt(n, e, source, receiver, out_channels=('R', 'T')):
2621 '''
2622 Rotate traces from NE to RT system.
2624 :param n:
2625 North trace.
2626 :type n:
2627 :py:class:`~pyrocko.trace.Trace`
2629 :param e:
2630 East trace.
2631 :type e:
2632 :py:class:`~pyrocko.trace.Trace`
2634 :param source:
2635 Source of the recorded signal.
2636 :type source:
2637 :py:class:`pyrocko.gf.seismosizer.Source`
2639 :param receiver:
2640 Receiver of the recorded signal.
2641 :type receiver:
2642 :py:class:`pyrocko.model.location.Location`
2644 :param out_channels:
2645 Channel codes of the output channels (radial, transversal).
2646 Default is ('R', 'T').
2648 :type out_channels
2649 optional, tuple[str, str]
2651 :returns:
2652 Rotated traces (radial, transversal).
2653 :rtype:
2654 tuple[
2655 :py:class:`~pyrocko.trace.Trace`,
2656 :py:class:`~pyrocko.trace.Trace`]
2657 '''
2658 azimuth = orthodrome.azimuth(receiver, source) + 180.
2659 in_channels = n.channel, e.channel
2660 out = rotate(
2661 [n, e], azimuth,
2662 in_channels=in_channels,
2663 out_channels=out_channels)
2665 assert len(out) == 2
2666 for tr in out:
2667 if tr.channel == out_channels[0]:
2668 r = tr
2669 elif tr.channel == out_channels[1]:
2670 t = tr
2671 else:
2672 assert False
2674 return r, t
2677def rotate_to_lqt(traces, backazimuth, incidence, in_channels,
2678 out_channels=('L', 'Q', 'T')):
2679 '''
2680 Rotate traces from ZNE to LQT system.
2682 :param traces: list of traces in arbitrary order
2683 :param backazimuth: backazimuth in degrees clockwise from north
2684 :param incidence: incidence angle in degrees from vertical
2685 :param in_channels: input channel names
2686 :param out_channels: output channel names (default: ('L', 'Q', 'T'))
2687 :returns: list of transformed traces
2688 '''
2689 i = incidence/180.*num.pi
2690 b = backazimuth/180.*num.pi
2692 ci = num.cos(i)
2693 cb = num.cos(b)
2694 si = num.sin(i)
2695 sb = num.sin(b)
2697 rotmat = num.array(
2698 [[ci, -cb*si, -sb*si], [si, cb*ci, sb*ci], [0., sb, -cb]])
2699 return project(traces, rotmat, in_channels, out_channels)
2702def _decompose(a):
2703 '''
2704 Decompose matrix into independent submatrices.
2705 '''
2707 def depends(iout, a):
2708 row = a[iout, :]
2709 return set(num.arange(row.size).compress(row != 0.0))
2711 def provides(iin, a):
2712 col = a[:, iin]
2713 return set(num.arange(col.size).compress(col != 0.0))
2715 a = num.asarray(a)
2716 outs = set(range(a.shape[0]))
2717 systems = []
2718 while outs:
2719 iout = outs.pop()
2721 gout = set()
2722 for iin in depends(iout, a):
2723 gout.update(provides(iin, a))
2725 if not gout:
2726 continue
2728 gin = set()
2729 for iout2 in gout:
2730 gin.update(depends(iout2, a))
2732 if not gin:
2733 continue
2735 for iout2 in gout:
2736 if iout2 in outs:
2737 outs.remove(iout2)
2739 gin = list(gin)
2740 gin.sort()
2741 gout = list(gout)
2742 gout.sort()
2744 systems.append((gin, gout, a[gout, :][:, gin]))
2746 return systems
2749def _channels_to_names(channels):
2750 from pyrocko import squirrel
2751 names = []
2752 for ch in channels:
2753 if isinstance(ch, model.Channel):
2754 names.append(ch.name)
2755 elif isinstance(ch, squirrel.Channel):
2756 names.append(ch.codes.channel)
2757 else:
2758 names.append(ch)
2760 return names
2763def project(traces, matrix, in_channels, out_channels):
2764 '''
2765 Affine transform of three-component traces.
2767 Compute matrix-vector product of three-component traces, to e.g. rotate
2768 traces into a different basis. The traces are distinguished and ordered by
2769 their channel attribute. The tranform is applied to overlapping parts of
2770 any appropriate combinations of the input traces. This should allow this
2771 function to be robust with data gaps. It also tries to apply the
2772 tranformation to subsets of the channels, if this is possible, so that, if
2773 for example a vertical compontent is missing, horizontal components can
2774 still be rotated.
2776 :param traces: list of traces in arbitrary order
2777 :param matrix: tranformation matrix
2778 :param in_channels: input channel names
2779 :param out_channels: output channel names
2780 :returns: list of transformed traces
2781 '''
2783 in_channels = tuple(_channels_to_names(in_channels))
2784 out_channels = tuple(_channels_to_names(out_channels))
2785 systems = _decompose(matrix)
2787 # fallback to full matrix if some are not quadratic
2788 for iins, iouts, submatrix in systems:
2789 if submatrix.shape[0] != submatrix.shape[1]:
2790 if len(in_channels) != 3 or len(out_channels) != 3:
2791 return []
2792 else:
2793 return _project3(traces, matrix, in_channels, out_channels)
2795 projected = []
2796 for iins, iouts, submatrix in systems:
2797 in_cha = tuple([in_channels[iin] for iin in iins])
2798 out_cha = tuple([out_channels[iout] for iout in iouts])
2799 if submatrix.shape[0] == 1:
2800 projected.extend(_project1(traces, submatrix, in_cha, out_cha))
2801 elif submatrix.shape[1] == 2:
2802 projected.extend(_project2(traces, submatrix, in_cha, out_cha))
2803 else:
2804 projected.extend(_project3(traces, submatrix, in_cha, out_cha))
2806 return projected
2809def project_dependencies(matrix, in_channels, out_channels):
2810 '''
2811 Figure out what dependencies project() would produce.
2812 '''
2814 in_channels = tuple(_channels_to_names(in_channels))
2815 out_channels = tuple(_channels_to_names(out_channels))
2816 systems = _decompose(matrix)
2818 subpro = []
2819 for iins, iouts, submatrix in systems:
2820 if submatrix.shape[0] != submatrix.shape[1]:
2821 subpro.append((matrix, in_channels, out_channels))
2823 if not subpro:
2824 for iins, iouts, submatrix in systems:
2825 in_cha = tuple([in_channels[iin] for iin in iins])
2826 out_cha = tuple([out_channels[iout] for iout in iouts])
2827 subpro.append((submatrix, in_cha, out_cha))
2829 deps = {}
2830 for mat, in_cha, out_cha in subpro:
2831 for oc in out_cha:
2832 if oc not in deps:
2833 deps[oc] = []
2835 for ic in in_cha:
2836 deps[oc].append(ic)
2838 return deps
2841def _project1(traces, matrix, in_channels, out_channels):
2842 assert len(in_channels) == 1
2843 assert len(out_channels) == 1
2844 assert matrix.shape == (1, 1)
2846 projected = []
2847 for a in traces:
2848 if not (a.channel,) == in_channels:
2849 continue
2851 ac = a.copy()
2852 ac.set_ydata(matrix[0, 0]*a.get_ydata())
2853 ac.set_codes(channel=out_channels[0])
2854 projected.append(ac)
2856 return projected
2859def _project2(traces, matrix, in_channels, out_channels):
2860 assert len(in_channels) == 2
2861 assert len(out_channels) == 2
2862 assert matrix.shape == (2, 2)
2863 projected = []
2864 for a in traces:
2865 for b in traces:
2866 if not ((a.channel, b.channel) == in_channels and
2867 a.nslc_id[:3] == b.nslc_id[:3] and
2868 abs(a.deltat-b.deltat) < a.deltat*0.001):
2869 continue
2871 tmin = max(a.tmin, b.tmin)
2872 tmax = min(a.tmax, b.tmax)
2874 if tmin > tmax:
2875 continue
2877 ac = a.chop(tmin, tmax, inplace=False, include_last=True)
2878 bc = b.chop(tmin, tmax, inplace=False, include_last=True)
2879 if abs(ac.tmin - bc.tmin) > ac.deltat*0.01:
2880 logger.warning(
2881 'Cannot project traces with displaced sampling '
2882 '(%s, %s, %s, %s)' % a.nslc_id)
2883 continue
2885 acydata = num.dot(matrix[0], (ac.get_ydata(), bc.get_ydata()))
2886 bcydata = num.dot(matrix[1], (ac.get_ydata(), bc.get_ydata()))
2888 ac.set_ydata(acydata)
2889 bc.set_ydata(bcydata)
2891 ac.set_codes(channel=out_channels[0])
2892 bc.set_codes(channel=out_channels[1])
2894 projected.append(ac)
2895 projected.append(bc)
2897 return projected
2900def _project3(traces, matrix, in_channels, out_channels):
2901 assert len(in_channels) == 3
2902 assert len(out_channels) == 3
2903 assert matrix.shape == (3, 3)
2904 projected = []
2905 for a in traces:
2906 for b in traces:
2907 for c in traces:
2908 if not ((a.channel, b.channel, c.channel) == in_channels
2909 and a.nslc_id[:3] == b.nslc_id[:3]
2910 and b.nslc_id[:3] == c.nslc_id[:3]
2911 and abs(a.deltat-b.deltat) < a.deltat*0.001
2912 and abs(b.deltat-c.deltat) < b.deltat*0.001):
2914 continue
2916 tmin = max(a.tmin, b.tmin, c.tmin)
2917 tmax = min(a.tmax, b.tmax, c.tmax)
2919 if tmin >= tmax:
2920 continue
2922 ac = a.chop(tmin, tmax, inplace=False, include_last=True)
2923 bc = b.chop(tmin, tmax, inplace=False, include_last=True)
2924 cc = c.chop(tmin, tmax, inplace=False, include_last=True)
2925 if (abs(ac.tmin - bc.tmin) > ac.deltat*0.01
2926 or abs(bc.tmin - cc.tmin) > bc.deltat*0.01):
2928 logger.warning(
2929 'Cannot project traces with displaced sampling '
2930 '(%s, %s, %s, %s)' % a.nslc_id)
2931 continue
2933 acydata = num.dot(
2934 matrix[0],
2935 (ac.get_ydata(), bc.get_ydata(), cc.get_ydata()))
2936 bcydata = num.dot(
2937 matrix[1],
2938 (ac.get_ydata(), bc.get_ydata(), cc.get_ydata()))
2939 ccydata = num.dot(
2940 matrix[2],
2941 (ac.get_ydata(), bc.get_ydata(), cc.get_ydata()))
2943 ac.set_ydata(acydata)
2944 bc.set_ydata(bcydata)
2945 cc.set_ydata(ccydata)
2947 ac.set_codes(channel=out_channels[0])
2948 bc.set_codes(channel=out_channels[1])
2949 cc.set_codes(channel=out_channels[2])
2951 projected.append(ac)
2952 projected.append(bc)
2953 projected.append(cc)
2955 return projected
2958def correlate(a, b, mode='valid', normalization=None, use_fft=False):
2959 '''
2960 Cross correlation of two traces.
2962 :param a,b: input traces
2963 :param mode: ``'valid'``, ``'full'``, or ``'same'``
2964 :param normalization: ``'normal'``, ``'gliding'``, or ``None``
2965 :param use_fft: bool, whether to do cross correlation in spectral domain
2967 :returns: trace containing cross correlation coefficients
2969 This function computes the cross correlation between two traces. It
2970 evaluates the discrete equivalent of
2972 .. math::
2974 c(t) = \\int_{-\\infty}^{\\infty} a^{\\ast}(\\tau) b(t+\\tau) d\\tau
2976 where the star denotes complex conjugate. Note, that the arguments here are
2977 swapped when compared with the :py:func:`numpy.correlate` function,
2978 which is internally called. This function should be safe even with older
2979 versions of NumPy, where the correlate function has some problems.
2981 A trace containing the cross correlation coefficients is returned. The time
2982 information of the output trace is set so that the returned cross
2983 correlation can be viewed directly as a function of time lag.
2985 Example::
2987 # align two traces a and b containing a time shifted similar signal:
2988 c = pyrocko.trace.correlate(a,b)
2989 t, coef = c.max() # get time and value of maximum
2990 b.shift(-t) # align b with a
2992 '''
2994 assert_same_sampling_rate(a, b)
2996 ya, yb = a.ydata, b.ydata
2998 # need reversed order here:
2999 yc = numpy_correlate_fixed(yb, ya, mode=mode, use_fft=use_fft)
3000 kmin, kmax = numpy_correlate_lag_range(yb, ya, mode=mode, use_fft=use_fft)
3002 if normalization == 'normal':
3003 normfac = num.sqrt(num.sum(ya**2))*num.sqrt(num.sum(yb**2))
3004 yc = yc/normfac
3006 elif normalization == 'gliding':
3007 if mode != 'valid':
3008 assert False, 'gliding normalization currently only available ' \
3009 'with "valid" mode.'
3011 if ya.size < yb.size:
3012 yshort, ylong = ya, yb
3013 else:
3014 yshort, ylong = yb, ya
3016 epsilon = 0.00001
3017 normfac_short = num.sqrt(num.sum(yshort**2))
3018 normfac = normfac_short * num.sqrt(
3019 moving_sum(ylong**2, yshort.size, mode='valid')) \
3020 + normfac_short*epsilon
3022 if yb.size <= ya.size:
3023 normfac = normfac[::-1]
3025 yc /= normfac
3027 c = a.copy()
3028 c.set_ydata(yc)
3029 c.set_codes(*merge_codes(a, b, '~'))
3030 c.shift(-c.tmin + b.tmin-a.tmin + kmin * c.deltat)
3032 return c
3035def deconvolve(
3036 a, b, waterlevel,
3037 tshift=0.,
3038 pad=0.5,
3039 fd_taper=None,
3040 pad_to_pow2=True):
3042 same_sampling_rate(a, b)
3043 assert abs(a.tmin - b.tmin) < a.deltat * 0.001
3044 deltat = a.deltat
3045 npad = int(round(a.data_len()*pad + tshift / deltat))
3047 ndata = max(a.data_len(), b.data_len())
3048 ndata_pad = ndata + npad
3050 if pad_to_pow2:
3051 ntrans = nextpow2(ndata_pad)
3052 else:
3053 ntrans = ndata
3055 aspec = num.fft.rfft(a.ydata, ntrans)
3056 bspec = num.fft.rfft(b.ydata, ntrans)
3058 out = aspec * num.conj(bspec)
3060 bautocorr = bspec*num.conj(bspec)
3061 denom = num.maximum(bautocorr, waterlevel * bautocorr.max())
3063 out /= denom
3064 df = 1/(ntrans*deltat)
3066 if fd_taper is not None:
3067 fd_taper(out, 0.0, df)
3069 ydata = num.roll(num.fft.irfft(out), int(round(tshift/deltat)))
3070 c = a.copy(data=False)
3071 c.set_ydata(ydata[:ndata])
3072 c.set_codes(*merge_codes(a, b, '/'))
3073 return c
3076def assert_same_sampling_rate(a, b, eps=1.0e-6):
3077 assert same_sampling_rate(a, b, eps), \
3078 'Sampling rates differ: %g != %g' % (a.deltat, b.deltat)
3081def same_sampling_rate(a, b, eps=1.0e-6):
3082 '''
3083 Check if two traces have the same sampling rate.
3085 :param a,b: input traces
3086 :param eps: relative tolerance
3087 '''
3089 return abs(a.deltat - b.deltat) < (a.deltat + b.deltat)*eps
3092def fix_deltat_rounding_errors(deltat):
3093 '''
3094 Try to undo sampling rate rounding errors.
3096 Fix rounding errors of sampling intervals when these are read from single
3097 precision floating point values.
3099 Assumes that the true sampling rate or sampling interval was an integer
3100 value. No correction will be applied if this would change the sampling
3101 rate by more than 0.001%.
3102 '''
3104 if deltat <= 1.0:
3105 deltat_new = 1.0 / round(1.0 / deltat)
3106 else:
3107 deltat_new = round(deltat)
3109 if abs(deltat_new - deltat) / deltat > 1e-5:
3110 deltat_new = deltat
3112 return deltat_new
3115def merge_codes(a, b, sep='-'):
3116 '''
3117 Merge network-station-location-channel codes of a pair of traces.
3118 '''
3120 o = []
3121 for xa, xb in zip(a.nslc_id, b.nslc_id):
3122 if xa == xb:
3123 o.append(xa)
3124 else:
3125 o.append(sep.join((xa, xb)))
3126 return o
3129class Taper(Object):
3130 '''
3131 Base class for tapers.
3133 Does nothing by default.
3134 '''
3136 def __call__(self, y, x0, dx):
3137 pass
3140class CosTaper(Taper):
3141 '''
3142 Cosine Taper.
3144 :param a: start of fading in
3145 :param b: end of fading in
3146 :param c: start of fading out
3147 :param d: end of fading out
3148 '''
3150 a = Float.T()
3151 b = Float.T()
3152 c = Float.T()
3153 d = Float.T()
3155 def __init__(self, a, b, c, d):
3156 Taper.__init__(self, a=a, b=b, c=c, d=d)
3158 def __call__(self, y, x0, dx):
3160 if y.dtype == num.dtype(float):
3161 _apply_costaper = signal_ext.apply_costaper
3162 else:
3163 _apply_costaper = apply_costaper
3165 _apply_costaper(self.a, self.b, self.c, self.d, y, x0, dx)
3167 def span(self, y, x0, dx):
3168 return span_costaper(self.a, self.b, self.c, self.d, y, x0, dx)
3170 def time_span(self):
3171 return self.a, self.d
3174class CosFader(Taper):
3175 '''
3176 Cosine Fader.
3178 :param xfade: fade in and fade out time in seconds (optional)
3179 :param xfrac: fade in and fade out as fraction between 0. and 1. (optional)
3181 Only one argument can be set. The other should to be ``None``.
3182 '''
3184 xfade = Float.T(optional=True)
3185 xfrac = Float.T(optional=True)
3187 def __init__(self, xfade=None, xfrac=None):
3188 Taper.__init__(self, xfade=xfade, xfrac=xfrac)
3189 assert (xfade is None) != (xfrac is None)
3190 self._xfade = xfade
3191 self._xfrac = xfrac
3193 def __call__(self, y, x0, dx):
3195 xfade = self._xfade
3197 xlen = (y.size - 1)*dx
3198 if xfade is None:
3199 xfade = xlen * self._xfrac
3201 a = x0
3202 b = x0 + xfade
3203 c = x0 + xlen - xfade
3204 d = x0 + xlen
3206 apply_costaper(a, b, c, d, y, x0, dx)
3208 def span(self, y, x0, dx):
3209 return 0, y.size
3211 def time_span(self):
3212 return None, None
3215def none_min(li):
3216 if None in li:
3217 return None
3218 else:
3219 return min(x for x in li if x is not None)
3222def none_max(li):
3223 if None in li:
3224 return None
3225 else:
3226 return max(x for x in li if x is not None)
3229class MultiplyTaper(Taper):
3230 '''
3231 Multiplication of several tapers.
3232 '''
3234 tapers = List.T(Taper.T())
3236 def __init__(self, tapers=None):
3237 if tapers is None:
3238 tapers = []
3240 Taper.__init__(self, tapers=tapers)
3242 def __call__(self, y, x0, dx):
3243 for taper in self.tapers:
3244 taper(y, x0, dx)
3246 def span(self, y, x0, dx):
3247 spans = []
3248 for taper in self.tapers:
3249 spans.append(taper.span(y, x0, dx))
3251 mins, maxs = list(zip(*spans))
3252 return min(mins), max(maxs)
3254 def time_span(self):
3255 spans = []
3256 for taper in self.tapers:
3257 spans.append(taper.time_span())
3259 mins, maxs = list(zip(*spans))
3260 return none_min(mins), none_max(maxs)
3263class GaussTaper(Taper):
3264 '''
3265 Frequency domain Gaussian filter.
3266 '''
3268 alpha = Float.T()
3270 def __init__(self, alpha):
3271 Taper.__init__(self, alpha=alpha)
3272 self._alpha = alpha
3274 def __call__(self, y, x0, dx):
3275 f = x0 + num.arange(y.size)*dx
3276 y *= num.exp(-num.pi**2 / (self._alpha**2) * f**2)
3279cached_coefficients = {}
3282def _get_cached_filter_coeffs(order, corners, btype):
3283 ck = (order, tuple(corners), btype)
3284 if ck not in cached_coefficients:
3285 if len(corners) == 1:
3286 corners = corners[0]
3288 cached_coefficients[ck] = signal.butter(
3289 order, corners, btype=btype)
3291 return cached_coefficients[ck]
3294class _globals(object):
3295 _numpy_has_correlate_flip_bug = None
3298def _default_key(tr):
3299 return (tr.network, tr.station, tr.location, tr.channel)
3302def numpy_has_correlate_flip_bug():
3303 '''
3304 Check if NumPy's correlate function reveals old behaviour.
3305 '''
3307 if _globals._numpy_has_correlate_flip_bug is None:
3308 a = num.array([0, 0, 1, 0, 0, 0, 0])
3309 b = num.array([0, 0, 0, 0, 1, 0, 0, 0])
3310 ab = num.correlate(a, b, mode='same')
3311 ba = num.correlate(b, a, mode='same')
3312 _globals._numpy_has_correlate_flip_bug = num.all(ab == ba)
3314 return _globals._numpy_has_correlate_flip_bug
3317def numpy_correlate_fixed(a, b, mode='valid', use_fft=False):
3318 '''
3319 Call :py:func:`numpy.correlate` with fixes.
3321 c[k] = sum_i a[i+k] * conj(b[i])
3323 Note that the result produced by newer numpy.correlate is always flipped
3324 with respect to the formula given in its documentation (if ascending k
3325 assumed for the output).
3326 '''
3328 if use_fft:
3329 if a.size < b.size:
3330 c = signal.fftconvolve(b[::-1], a, mode=mode)
3331 else:
3332 c = signal.fftconvolve(a, b[::-1], mode=mode)
3333 return c
3335 else:
3336 buggy = numpy_has_correlate_flip_bug()
3338 a = num.asarray(a)
3339 b = num.asarray(b)
3341 if buggy:
3342 b = num.conj(b)
3344 c = num.correlate(a, b, mode=mode)
3346 if buggy and a.size < b.size:
3347 return c[::-1]
3348 else:
3349 return c
3352def numpy_correlate_emulate(a, b, mode='valid'):
3353 '''
3354 Slow version of :py:func:`numpy.correlate` for comparison.
3355 '''
3357 a = num.asarray(a)
3358 b = num.asarray(b)
3359 kmin = -(b.size-1)
3360 klen = a.size-kmin
3361 kmin, kmax = numpy_correlate_lag_range(a, b, mode=mode)
3362 kmin = int(kmin)
3363 kmax = int(kmax)
3364 klen = kmax - kmin + 1
3365 c = num.zeros(klen, dtype=num.promote_types(b.dtype, a.dtype))
3366 for k in range(kmin, kmin+klen):
3367 imin = max(0, -k)
3368 ilen = min(b.size, a.size-k) - imin
3369 c[k-kmin] = num.sum(
3370 a[imin+k:imin+ilen+k] * num.conj(b[imin:imin+ilen]))
3372 return c
3375def numpy_correlate_lag_range(a, b, mode='valid', use_fft=False):
3376 '''
3377 Get range of lags for which :py:func:`numpy.correlate` produces values.
3378 '''
3380 a = num.asarray(a)
3381 b = num.asarray(b)
3383 kmin = -(b.size-1)
3384 if mode == 'full':
3385 klen = a.size-kmin
3386 elif mode == 'same':
3387 klen = max(a.size, b.size)
3388 kmin += (a.size+b.size-1 - max(a.size, b.size)) // 2 + \
3389 int(not use_fft and a.size % 2 == 0 and b.size > a.size)
3390 elif mode == 'valid':
3391 klen = abs(a.size - b.size) + 1
3392 kmin += min(a.size, b.size) - 1
3394 return kmin, kmin + klen - 1
3397def autocorr(x, nshifts):
3398 '''
3399 Compute biased estimate of the first autocorrelation coefficients.
3401 :param x: input array
3402 :param nshifts: number of coefficients to calculate
3403 '''
3405 mean = num.mean(x)
3406 std = num.std(x)
3407 n = x.size
3408 xdm = x - mean
3409 r = num.zeros(nshifts)
3410 for k in range(nshifts):
3411 r[k] = 1./((n-num.abs(k))*std) * num.sum(xdm[:n-k] * xdm[k:])
3413 return r
3416def yulewalker(x, order):
3417 '''
3418 Compute autoregression coefficients using Yule-Walker method.
3420 :param x: input array
3421 :param order: number of coefficients to produce
3423 A biased estimate of the autocorrelation is used. The Yule-Walker equations
3424 are solved by :py:func:`numpy.linalg.inv` instead of Levinson-Durbin
3425 recursion which is normally used.
3426 '''
3428 gamma = autocorr(x, order+1)
3429 d = gamma[1:1+order]
3430 a = num.zeros((order, order))
3431 gamma2 = num.concatenate((gamma[::-1], gamma[1:order]))
3432 for i in range(order):
3433 ioff = order-i
3434 a[i, :] = gamma2[ioff:ioff+order]
3436 return num.dot(num.linalg.inv(a), -d)
3439def moving_avg(x, n):
3440 n = int(n)
3441 cx = x.cumsum()
3442 nn = len(x)
3443 y = num.zeros(nn, dtype=cx.dtype)
3444 y[n//2:n//2+(nn-n)] = (cx[n:]-cx[:-n])/n
3445 y[:n//2] = y[n//2]
3446 y[n//2+(nn-n):] = y[n//2+(nn-n)-1]
3447 return y
3450def moving_sum(x, n, mode='valid'):
3451 n = int(n)
3452 cx = x.cumsum()
3453 nn = len(x)
3455 if mode == 'valid':
3456 if nn-n+1 <= 0:
3457 return num.zeros(0, dtype=cx.dtype)
3458 y = num.zeros(nn-n+1, dtype=cx.dtype)
3459 y[0] = cx[n-1]
3460 y[1:nn-n+1] = cx[n:nn]-cx[0:nn-n]
3462 if mode == 'full':
3463 y = num.zeros(nn+n-1, dtype=cx.dtype)
3464 if n <= nn:
3465 y[0:n] = cx[0:n]
3466 y[n:nn] = cx[n:nn]-cx[0:nn-n]
3467 y[nn:nn+n-1] = cx[-1]-cx[nn-n:nn-1]
3468 else:
3469 y[0:nn] = cx[0:nn]
3470 y[nn:n] = cx[nn-1]
3471 y[n:nn+n-1] = cx[nn-1] - cx[0:nn-1]
3473 if mode == 'same':
3474 n1 = (n-1)//2
3475 y = num.zeros(nn, dtype=cx.dtype)
3476 if n <= nn:
3477 y[0:n-n1] = cx[n1:n]
3478 y[n-n1:nn-n1] = cx[n:nn]-cx[0:nn-n]
3479 y[nn-n1:nn] = cx[nn-1] - cx[nn-n:nn-n+n1]
3480 else:
3481 y[0:max(0, nn-n1)] = cx[min(n1, nn):nn]
3482 y[max(nn-n1, 0):min(n-n1, nn)] = cx[nn-1]
3483 y[min(n-n1, nn):nn] = cx[nn-1] - cx[0:max(0, nn-(n-n1))]
3485 return y
3488def nextpow2(i):
3489 return 2**int(math.ceil(math.log(i)/math.log(2.)))
3492def snapper_w_offset(nmax, offset, delta, snapfun=math.ceil):
3493 def snap(x):
3494 return max(0, min(int(snapfun((x-offset)/delta)), nmax))
3495 return snap
3498def snapper(nmax, delta, snapfun=math.ceil):
3499 def snap(x):
3500 return max(0, min(int(snapfun(x/delta)), nmax))
3501 return snap
3504def apply_costaper(a, b, c, d, y, x0, dx):
3505 abcd = num.array((a, b, c, d), dtype=float)
3506 ja, jb, jc, jd = num.clip(num.ceil((abcd-x0)/dx).astype(int), 0, y.size)
3507 y[:ja] = 0.
3508 y[ja:jb] *= 0.5 \
3509 - 0.5*num.cos((dx*num.arange(ja, jb)-(a-x0))/(b-a)*num.pi)
3510 y[jc:jd] *= 0.5 \
3511 + 0.5*num.cos((dx*num.arange(jc, jd)-(c-x0))/(d-c)*num.pi)
3512 y[jd:] = 0.
3515def span_costaper(a, b, c, d, y, x0, dx):
3516 hi = snapper_w_offset(y.size, x0, dx)
3517 return hi(a), hi(d) - hi(a)
3520def costaper(a, b, c, d, nfreqs, deltaf):
3521 hi = snapper(nfreqs, deltaf)
3522 tap = num.zeros(nfreqs)
3523 tap[hi(a):hi(b)] = 0.5 \
3524 - 0.5*num.cos((deltaf*num.arange(hi(a), hi(b))-a)/(b-a)*num.pi)
3525 tap[hi(b):hi(c)] = 1.
3526 tap[hi(c):hi(d)] = 0.5 \
3527 + 0.5*num.cos((deltaf*num.arange(hi(c), hi(d))-c)/(d-c)*num.pi)
3529 return tap
3532def t2ind(t, tdelta, snap=round):
3533 return int(snap(t/tdelta))
3536def hilbert(x, N=None):
3537 '''
3538 Return the hilbert transform of x of length N.
3540 (from scipy.signal, but changed to use fft and ifft from numpy.fft)
3541 '''
3543 x = num.asarray(x)
3544 if N is None:
3545 N = len(x)
3546 if N <= 0:
3547 raise ValueError('N must be positive.')
3548 if num.iscomplexobj(x):
3549 logger.warning('imaginary part of x ignored.')
3550 x = num.real(x)
3552 Xf = num.fft.fft(x, N, axis=0)
3553 h = num.zeros(N)
3554 if N % 2 == 0:
3555 h[0] = h[N//2] = 1
3556 h[1:N//2] = 2
3557 else:
3558 h[0] = 1
3559 h[1:(N+1)//2] = 2
3561 if len(x.shape) > 1:
3562 h = h[:, num.newaxis]
3563 x = num.fft.ifft(Xf*h)
3564 return x
3567def near(a, b, eps):
3568 return abs(a-b) < eps
3571def coroutine(func):
3572 def wrapper(*args, **kwargs):
3573 gen = func(*args, **kwargs)
3574 next(gen)
3575 return gen
3577 wrapper.__name__ = func.__name__
3578 wrapper.__dict__ = func.__dict__
3579 wrapper.__doc__ = func.__doc__
3580 return wrapper
3583class States(object):
3584 '''
3585 Utility to store channel-specific state in coroutines.
3586 '''
3588 def __init__(self):
3589 self._states = {}
3591 def get(self, tr):
3592 k = tr.nslc_id
3593 if k in self._states:
3594 tmin, deltat, dtype, value = self._states[k]
3595 if (near(tmin, tr.tmin, deltat/100.)
3596 and near(deltat, tr.deltat, deltat/10000.)
3597 and dtype == tr.ydata.dtype):
3599 return value
3601 return None
3603 def set(self, tr, value):
3604 k = tr.nslc_id
3605 if k in self._states and self._states[k][-1] is not value:
3606 self.free(self._states[k][-1])
3608 self._states[k] = (tr.tmax+tr.deltat, tr.deltat, tr.ydata.dtype, value)
3610 def free(self, value):
3611 pass
3614@coroutine
3615def co_list_append(list):
3616 while True:
3617 list.append((yield))
3620class ScipyBug(Exception):
3621 pass
3624@coroutine
3625def co_lfilter(target, b, a):
3626 '''
3627 Successively filter broken continuous trace data (coroutine).
3629 Create coroutine which takes :py:class:`Trace` objects, filters their data
3630 through :py:func:`scipy.signal.lfilter` and sends new :py:class:`Trace`
3631 objects containing the filtered data to target. This is useful, if one
3632 wants to filter a long continuous time series, which is split into many
3633 successive traces without producing filter artifacts at trace boundaries.
3635 Filter states are kept *per channel*, specifically, for each (network,
3636 station, location, channel) combination occuring in the input traces, a
3637 separate state is created and maintained. This makes it possible to filter
3638 multichannel or multistation data with only one :py:func:`co_lfilter`
3639 instance.
3641 Filter state is reset, when gaps occur.
3643 Use it like this::
3645 from pyrocko.trace import co_lfilter, co_list_append
3647 filtered_traces = []
3648 pipe = co_lfilter(co_list_append(filtered_traces), a, b)
3649 for trace in traces:
3650 pipe.send(trace)
3652 pipe.close()
3654 '''
3656 try:
3657 states = States()
3658 output = None
3659 while True:
3660 input = (yield)
3662 zi = states.get(input)
3663 if zi is None:
3664 zi = num.zeros(max(len(a), len(b))-1, dtype=float)
3666 output = input.copy(data=False)
3667 try:
3668 ydata, zf = signal.lfilter(b, a, input.get_ydata(), zi=zi)
3669 except ValueError:
3670 raise ScipyBug(
3671 'signal.lfilter failed: could be related to a bug '
3672 'in some older scipy versions, e.g. on opensuse42.1')
3674 output.set_ydata(ydata)
3675 states.set(input, zf)
3676 target.send(output)
3678 except GeneratorExit:
3679 target.close()
3682def co_antialias(target, q, n=None, ftype='fir'):
3683 b, a, n = util.decimate_coeffs(q, n, ftype)
3684 anti = co_lfilter(target, b, a)
3685 return anti
3688@coroutine
3689def co_dropsamples(target, q, nfir):
3690 try:
3691 states = States()
3692 while True:
3693 tr = (yield)
3694 newdeltat = q * tr.deltat
3695 ioffset = states.get(tr)
3696 if ioffset is None:
3697 # for fir filter, the first nfir samples are pulluted by
3698 # boundary effects; cut it off.
3699 # for iir this may be (much) more, we do not correct for that.
3700 # put sample instances to a time which is a multiple of the
3701 # new sampling interval.
3702 newtmin_want = math.ceil(
3703 (tr.tmin+(nfir+1)*tr.deltat) / newdeltat) * newdeltat \
3704 - (nfir/2*tr.deltat)
3705 ioffset = int(round((newtmin_want - tr.tmin)/tr.deltat))
3706 if ioffset < 0:
3707 ioffset = ioffset % q
3709 newtmin_have = tr.tmin + ioffset * tr.deltat
3710 newtr = tr.copy(data=False)
3711 newtr.deltat = newdeltat
3712 # because the fir kernel shifts data by nfir/2 samples:
3713 newtr.tmin = newtmin_have - (nfir/2*tr.deltat)
3714 newtr.set_ydata(tr.get_ydata()[ioffset::q].copy())
3715 states.set(tr, (ioffset % q - tr.data_len() % q) % q)
3716 target.send(newtr)
3718 except GeneratorExit:
3719 target.close()
3722def co_downsample(target, q, n=None, ftype='fir'):
3723 '''
3724 Successively downsample broken continuous trace data (coroutine).
3726 Create coroutine which takes :py:class:`Trace` objects, downsamples their
3727 data and sends new :py:class:`Trace` objects containing the downsampled
3728 data to target. This is useful, if one wants to downsample a long
3729 continuous time series, which is split into many successive traces without
3730 producing filter artifacts and gaps at trace boundaries.
3732 Filter states are kept *per channel*, specifically, for each (network,
3733 station, location, channel) combination occuring in the input traces, a
3734 separate state is created and maintained. This makes it possible to filter
3735 multichannel or multistation data with only one :py:func:`co_lfilter`
3736 instance.
3738 Filter state is reset, when gaps occur. The sampling instances are choosen
3739 so that they occur at (or as close as possible) to even multiples of the
3740 sampling interval of the downsampled trace (based on system time).
3741 '''
3743 b, a, n = util.decimate_coeffs(q, n, ftype)
3744 return co_antialias(co_dropsamples(target, q, n), q, n, ftype)
3747@coroutine
3748def co_downsample_to(target, deltat):
3750 decimators = {}
3751 try:
3752 while True:
3753 tr = (yield)
3754 ratio = deltat / tr.deltat
3755 rratio = round(ratio)
3756 if abs(rratio - ratio)/ratio > 0.0001:
3757 raise util.UnavailableDecimation('ratio = %g' % ratio)
3759 deci_seq = tuple(x for x in util.decitab(int(rratio)) if x != 1)
3760 if deci_seq not in decimators:
3761 pipe = target
3762 for q in deci_seq[::-1]:
3763 pipe = co_downsample(pipe, q)
3765 decimators[deci_seq] = pipe
3767 decimators[deci_seq].send(tr)
3769 except GeneratorExit:
3770 for g in decimators.values():
3771 g.close()
3774class DomainChoice(StringChoice):
3775 choices = [
3776 'time_domain',
3777 'frequency_domain',
3778 'envelope',
3779 'absolute',
3780 'cc_max_norm']
3783class MisfitSetup(Object):
3784 '''
3785 Contains misfit setup to be used in :py:meth:`Trace.misfit`
3787 :param description: Description of the setup
3788 :param norm: L-norm classifier
3789 :param taper: Object of :py:class:`Taper`
3790 :param filter: Object of :py:class:`~pyrocko.response.FrequencyResponse`
3791 :param domain: ['time_domain', 'frequency_domain', 'envelope', 'absolute',
3792 'cc_max_norm']
3794 Can be dumped to a yaml file.
3795 '''
3797 xmltagname = 'misfitsetup'
3798 description = String.T(optional=True)
3799 norm = Int.T(optional=False)
3800 taper = Taper.T(optional=False)
3801 filter = FrequencyResponse.T(optional=True)
3802 domain = DomainChoice.T(default='time_domain')
3805def equalize_sampling_rates(trace_1, trace_2):
3806 '''
3807 Equalize sampling rates of two traces (reduce higher sampling rate to
3808 lower).
3810 :param trace_1: :py:class:`Trace` object
3811 :param trace_2: :py:class:`Trace` object
3813 Returns a copy of the resampled trace if resampling is needed.
3814 '''
3816 if same_sampling_rate(trace_1, trace_2):
3817 return trace_1, trace_2
3819 if trace_1.deltat < trace_2.deltat:
3820 t1_out = trace_1.copy()
3821 t1_out.downsample_to(deltat=trace_2.deltat, snap=True)
3822 logger.debug('Trace downsampled (return copy of trace): %s'
3823 % '.'.join(t1_out.nslc_id))
3824 return t1_out, trace_2
3826 elif trace_1.deltat > trace_2.deltat:
3827 t2_out = trace_2.copy()
3828 t2_out.downsample_to(deltat=trace_1.deltat, snap=True)
3829 logger.debug('Trace downsampled (return copy of trace): %s'
3830 % '.'.join(t2_out.nslc_id))
3831 return trace_1, t2_out
3834def Lx_norm(u, v, norm=2):
3835 '''
3836 Calculate the misfit denominator *m* and the normalization divisor *n*
3837 according to norm.
3839 The normalization divisor *n* is calculated from ``v``.
3841 :param u: :py:class:`numpy.ndarray`
3842 :param v: :py:class:`numpy.ndarray`
3843 :param norm: (default = 2)
3845 ``u`` and ``v`` must be of same size.
3846 '''
3848 if norm == 1:
3849 return (
3850 num.sum(num.abs(v-u)),
3851 num.sum(num.abs(v)))
3853 elif norm == 2:
3854 return (
3855 num.sqrt(num.sum((v-u)**2)),
3856 num.sqrt(num.sum(v**2)))
3858 else:
3859 return (
3860 num.power(num.sum(num.abs(num.power(v - u, norm))), 1./norm),
3861 num.power(num.sum(num.abs(num.power(v, norm))), 1./norm))
3864def do_downsample(tr, deltat):
3865 if abs(tr.deltat - deltat) / tr.deltat > 1e-6:
3866 tr = tr.copy()
3867 tr.downsample_to(deltat, snap=True, demean=False)
3868 else:
3869 if tr.tmin/tr.deltat > 1e-6 or tr.tmax/tr.deltat > 1e-6:
3870 tr = tr.copy()
3871 tr.snap()
3872 return tr
3875def do_extend(tr, tmin, tmax):
3876 if tmin < tr.tmin or tmax > tr.tmax:
3877 tr = tr.copy()
3878 tr.extend(tmin=tmin, tmax=tmax, fillmethod='repeat')
3880 return tr
3883def do_pre_taper(tr, taper):
3884 return tr.taper(taper, inplace=False, chop=True)
3887def do_fft(tr, filter):
3888 if filter is None:
3889 return tr
3890 else:
3891 ndata = tr.ydata.size
3892 nfft = nextpow2(ndata)
3893 padded = num.zeros(nfft, dtype=float)
3894 padded[:ndata] = tr.ydata
3895 spectrum = num.fft.rfft(padded)
3896 df = 1.0 / (tr.deltat * nfft)
3897 frequencies = num.arange(spectrum.size)*df
3898 return [tr, frequencies, spectrum]
3901def do_filter(inp, filter):
3902 if filter is None:
3903 return inp
3904 else:
3905 tr, frequencies, spectrum = inp
3906 spectrum *= filter.evaluate(frequencies)
3907 return [tr, frequencies, spectrum]
3910def do_ifft(inp):
3911 if isinstance(inp, Trace):
3912 return inp
3913 else:
3914 tr, _, spectrum = inp
3915 ndata = tr.ydata.size
3916 tr = tr.copy(data=False)
3917 tr.set_ydata(num.fft.irfft(spectrum)[:ndata])
3918 return tr
3921def check_alignment(t1, t2):
3922 if abs(t1.tmin-t2.tmin) > t1.deltat * 1e-4 or \
3923 abs(t1.tmax - t2.tmax) > t1.deltat * 1e-4 or \
3924 t1.ydata.shape != t2.ydata.shape:
3925 raise MisalignedTraces(
3926 'Cannot calculate misfit of %s and %s due to misaligned '
3927 'traces.' % ('.'.join(t1.nslc_id), '.'.join(t2.nslc_id)))