If the series are not of equal length then the longer is truncated. The data are then extended by adding zeroes to obtain a suitable length dataset for the transformation.
The extended data array may be tapered at the ends to reduce 'leakage' in the power spectra. If tapering is requested then, as a default, a cosine bell is applied to 20% of the data points at each end of the data array. This may be altered using the appropriate key word. The resulting spectral estimates will fluctuate wildly unless smoothed. A Gaussian sigma of 2 or 3 may be suitable for light smoothing.
Alignment of the two input arrays is advisable if one lags the other. A shift of +n will shift the first series forward with respect to the second by n elements. This is what is required if their cross-correlation peaks at positive lags.
Subtopics: