nenupy.io.tf_utils.flatten_subband

nenupy.io.tf_utils.flatten_subband(data, channels, smooth_frequency_profile=False)[source]

Correct each subband if their levels is distributed like a sawtooth. Sometimes, in particular for strong sources, the subbands adopt a strange behavior (an example is shown in flatten_subband()). This function aims at correcting this effect.

For a given subband:

  • The median frequency profile is computed over the whole data time selection

  • Affine function is computed based on two points taken as the medians of both half profiles of the subband

  • This function is normalized by the median of the subband total profile

  • The corrected subband is the original subband divided by this normalized linear profile

Parameters:

  • data (ndarray) – The data to correct, shaped like (time, frequency, (polarizations))

  • channels (int) – Number of channels per subband.

  • smooth_frequency_profile (bool, optional) – Not yet finalized…, by default False

Returns:

The corrected data, where the subbands are flattened.

Return type:

ndarray

Raises:

ValueError – Raised if the data dimension 1, assumed to be the frequency, does not match channels.

Examples

>>> from nenupy.io.tf_utils import flatten_subband
>>> import numpy as np
>>> import matplotlib.pyplot as plt

>>> n_channels = 32
>>> n_subbands = 10

>>> rng = np.random.default_rng(12345)
>>> coeffs = rng.uniform(-0.1, 0.1, size=n_subbands)

>>> x_data = np.linspace(0, 10, n_channels * n_subbands)
>>> y_data = np.ones(x_data.size) # Idel data are flat
>>> saw_pattern = (coeffs[:, None]*np.linspace(-3, 3, n_channels) )[None, :]
>>> noise = rng.uniform(-0.05, 0.05, x_data.size)
>>> y_data_altered = (y_data.reshape((n_subbands, n_channels)) + saw_pattern).ravel() + noise

>>> y_data_corrected = flatten_subband(
        data=y_data_altered.reshape(1, y_data_altered.size, 1, 1),
        channels=n_channels
    )

>>> fig = plt.figure(figsize=(10, 5))
>>> for sb in range(n_subbands):
>>>     plt.axvline(x_data[sb*n_channels], color="black", linestyle=":")
>>> plt.plot(x_data, y_data, label="Original", linewidth=3, linestyle="--")
>>> plt.plot(x_data, y_data_altered, label="Altered")
>>> plt.plot(x_data, y_data_corrected[0, :, 0, 0], label="Corrected")
>>> plt.legend()
../_images/flatten.png