epyr.signalprocessing.preprocessing

Preprocessing utilities for time-domain EPR signal analysis.

Standalone, chainable functions for ESEEM, HYSCORE, and Rabi pipelines: remove_baseline, apodize, zero_pad.

Functions

apodize(signal[, window, alpha, ...])

Apply an apodization window to reduce spectral leakage from signal truncation.

remove_baseline(time, signal[, method, ...])

Fit and subtract a slowly-varying background from a time-domain EPR signal.

zero_pad(signal[, factor, n_points, axis, plot])

Pad a signal with trailing zeros to increase FFT frequency resolution.
epyr.signalprocessing.preprocessing.remove_baseline(time, signal, method='polynomial', order=3, end_fraction=0.15, axis=1, plot=False)[source]

Fit and subtract a slowly-varying background from a time-domain EPR signal.

Parameters:

  • time (np.ndarray) – Time axis, 1D array, any unit. For 2D data, must match the length of the signal along axis.

  • signal (np.ndarray) – Signal array, 1D or 2D.

  • method (str) – 'polynomial' or 'exponential'.

  • order (int) – Polynomial degree; ignored when method='exponential'.

  • end_fraction (float) – Fraction of trailing points used to anchor the polynomial fit. For exponential fits, the full time axis is used.

  • axis (int) – For 2D data: axis along which baseline is fitted and removed (0 = column-wise, 1 = row-wise). Default 1.

  • plot (bool) – Show a before/after figure.

Returns:

  • signal_corrected (np.ndarray) – Signal with baseline subtracted, same shape as signal.

  • baseline (np.ndarray) – Fitted baseline array, same shape as signal.

Raises:

  • ValueError – If method is not recognised, if signal is not 1D or 2D, or if axis is invalid for 2D data.

  • RuntimeError – If the exponential fit fails to converge.

Return type:

Tuple[ndarray, ndarray]

Examples

>>> time = np.linspace(0, 500, 256)
>>> signal = np.exp(-time / 200) + 0.05 * np.random.randn(256)
>>> corrected, baseline = remove_baseline(time, signal, method='exponential')
epyr.signalprocessing.preprocessing.apodize(signal, window='hann', alpha=None, half_window=None, axis='both', plot=False)[source]

Apply an apodization window to reduce spectral leakage from signal truncation.

Parameters:

  • signal (np.ndarray) – Signal array, 1D or 2D.

  • window (str) – Window type: any key accepted by apowin(): 'hann', 'hamming', 'blackman', 'kaiser', 'gaussian', 'exponential', etc.

  • alpha (float, optional) – Shape parameter for kaiser and gaussian windows.

  • half_window (str, optional) – None (symmetric window), 'left' (first half only), or 'right' (second half only).

  • axis (int or str) – For 2D data: 0, 1, or 'both'. When 'both', a 2D window is built as the outer product of two 1D windows, one per axis. Ignored for 1D input.

  • plot (bool) – Show a figure with the original signal, window shape, and windowed signal.

Returns:

signal_windowed – Windowed signal, same shape as signal.

Return type:

np.ndarray

Raises:

ValueError – If signal is not 1D or 2D, or if axis is invalid for 2D.

Examples

>>> signal = np.exp(-t / 120) * np.sin(2 * np.pi * 8.5e-3 * t)
>>> windowed = apodize(signal, window='hann', half_window='right')
epyr.signalprocessing.preprocessing.zero_pad(signal, factor=None, n_points=None, axis=-1, plot=False)[source]

Pad a signal with trailing zeros to increase FFT frequency resolution.

Exactly one of factor or n_points must be supplied.

Parameters:

  • signal (np.ndarray) – Signal array, 1D or 2D.

  • factor (int, optional) – Multiplicative factor: output length = factor * N along axis.

  • n_points (int, optional) – Absolute output length along axis.

  • axis (int or str) – Axis to pad. For 1D input this parameter is ignored. For 2D input: 0, 1, -1 (equivalent to 1), or 'both' (pad each axis independently with the same factor or n_points). Default -1.

  • plot (bool) – Show a before/after figure.

Returns:

signal_padded – Zero-padded signal, same dtype as signal.

Return type:

np.ndarray

Raises:

ValueError – If both or neither of factor / n_points are given, if n_points is shorter than the signal length, if signal is not 1D or 2D, or if axis is invalid for 2D.

Examples

>>> signal = np.ones(256)
>>> padded = zero_pad(signal, factor=4)   # 1024 points
>>> padded = zero_pad(signal, n_points=1024)