function of dascore.proc.correlate source

    patch: Patch ,
    samples = False,
    lag = None,
    **kwargs ,
)-> ‘PatchType’

Correlate source row/columns in a 2D patch with all other row/columns.

Correlations are done in the frequency domain. This function can accept a patch whose target dimension has already been transformed with the Patch.dft method, otherwise the dft will be performed. If the input has already been transformed, Patch.correlation_shift is useful to undo dft artefacts after the idft is applied.

While a 2D patch is required for input, a 3D patch is returned where the 3rd dimesion cooresponds to the source rows/columns. For the case of a single source, the Patch.squeeze method can be helpful to remove length 1 dimensions.


Parameter Description
patch : PatchType The input data patch to be cross-correlated. Must be 2-dimensional.
The patch can be in time or frequency domains.
samples : bool, optional (default = False) If True, the argument specified in kwargs refers to the sample not
value along that axis. See examples for details.
lag Deprecated, just use select on the output patch instead.
**kwargs Specifies correlation dimension and the master source(s), to which
we want to cross-correlate all other channels/time samples.If the
master source is an array, the function will compute correlations for
all the posible pairs.


import dascore as dc
from dascore.units import m, s
# Get a patch composed of sin waves whose correlation results
# can easily be checked.
patch = dc.get_example_patch(
    frequency=range(10, 20),

# Example 1
# Calculate cc for all channels as receivers and
# the 10 m channel as the master channel. Squeeze the output
# so the returned patch is 2D.
cc_patch = patch.correlate(distance = 10 * m).squeeze()

# Example 2
# Get cc within (-2,2) sec of lag for all channels as receivers
# and the 10 m channel as the master channel. The new patch has dimensions
# (lag_time, distance, source_distance)
cc_patch = (
    patch.correlate(distance = 10 * m)
    .select(lag_time=(-2, 2))

# Example 3
# First remove every other distance channel (less memory usage)
# the use the new 2nd channel as the source.
cc_patch = (
    patch.decimate(distance=2, filter_type=None)
    .correlate(distance=1, samples=True)

# Example 4
# Correlate along time dimension (perhaps for template matching
# applications)
cc_patch = patch.correlate(time=100, samples=True)

# Example 5
# A pipeline of frequency domain correlation and an array of sources
padded_patch = patch.pad(time="correlate")  # pad to at least 2n + 1
dft_patch = patch.dft("time", real=True)
# Any other pre-processing steps go here...
# ...
# Perform the correlation with 3 source channels
cc_patch = dft_patch.correlate(distance=[1, 3, 7], samples=True)
# Perform any post-processing here
# ...
# Convert back to time domain, apply `correlate shift` to undo
# fft related shifting and scaling as well as create lag coordinate.
cc_out = cc_patch.idft().correlate_shift("time")

1 - The cross-correlation is performed in the frequency domain.

2 - The output dimension is opposite of the one specified in kwargs and shares a name with the original coord except the string “lag_” is prepended. For example, “lag_time”.