Choosing parameters

Choosing parameters#

PyWavelet exposes a few key parameters that control the time–frequency grid and the window overlap.

The hard constraint: input length#

For the time-domain transform, your input must have length:

\[ND = Nt \times Nf\]

If your data does not naturally have that length, pad or crop it first (see below).

How Nt and Nf affect resolution#

For a signal of duration T:

\[\Delta T = T / Nt\]
\[\Delta F = \frac{1}{2\Delta T}\]

So increasing Nt gives finer time bins (smaller ΔT) but makes each frequency bin wider (larger ΔF). The grid is a deliberate time–frequency tradeoff.

How mult affects overlap#

mult controls the window support length K = 2 * mult * Nf (see Window function).

Rules of thumb:

  • start with mult = 16 or 32 for exploratory plots

  • if you see leakage, increase mult (up to roughly Nt/2)

  • if runtime is too slow, decrease mult before changing Nt/Nf

Padding/cropping recipes#

If you want to keep a fixed dt and choose an Nt × Nf grid:

ND = Nt * Nf
y = y[:ND]  # crop

Or pad:

import numpy as np
ND = Nt * Nf
if len(y) < ND:
    y = np.pad(y, (0, ND - len(y)))

If you just want a convenient length (power-of-two FFTs), consider the helper:

from pywavelet.types import TimeSeries
ts = TimeSeries(data=y, time=t).zero_pad_to_power_of_2(tukey_window_alpha=0.1)

Backend/precision#

If you’re tuning performance, also see Backends & Precision for switching NumPy/JAX/CuPy and float32/float64.

Contents