Multitaper Estimation

ffTRF includes an optional DPSS multi-taper estimator for more stable spectral estimates in noisy continuous-data settings.

Why Use Multi-Taper Estimation?

Compared with the standard single-window FFT estimator, multi-taper estimation can reduce variance in the spectral estimates by averaging across several orthogonal tapers.

This can help when:

• the data are noisy
• segments are relatively short
• you want smoother, more stable cross-spectral estimates

The tradeoff is stronger spectral smoothing.

Convenience API

scores = model.train_multitaper(
    stimulus=stimulus,
    response=response,
    fs=fs,
    tmin=0.0,
    tmax=0.120,
    regularization=np.logspace(-5, -1, 5),
    segment_duration=1.024,
    time_bandwidth=3.5,
    k="loo",
)

Representative lag-domain output from the bundled multi-taper example:

Equivalent Low-Level Form

scores = model.train(
    stimulus=stimulus,
    response=response,
    fs=fs,
    tmin=0.0,
    tmax=0.120,
    regularization=np.logspace(-5, -1, 5),
    spectral_method="multitaper",
    time_bandwidth=3.5,
    n_tapers=None,
    window=None,
)

train_multitaper(...) is just a convenience wrapper around this pattern.

Important Parameters

• time_bandwidth: controls the time-bandwidth product of the DPSS tapers
• n_tapers: number of tapers to average over
• segment_length or segment_duration: still controls the segment size used for the spectra

Larger time_bandwidth values:

• allow more tapers
• increase smoothing
• can improve stability in noisy settings

Important Restriction

window must stay None in multi-taper mode because the DPSS tapers already define the segment weighting.

The stored diagnostics from a multi-taper fit can also be inspected visually:

Practical Advice

• Start with time_bandwidth=3.5 unless you have a strong reason not to.
• If you mainly care about fine spectral detail, do not oversmooth.
• If ordinary fits are noisy or unstable, try multi-taper before making the model itself more complicated.

Downstream Analysis Is the Same

Once the model is fitted, the rest of the API stays the same:

• predict(...)
• score(...)
• plot(...)
• plot_transfer_function(...)
• cross_spectral_diagnostics(...)
• frequency_resolved_weights(...)

So multi-taper changes how the model is estimated, not how it is used later.