Sensor Noise Suppression (SNS)#
SNS suppresses sensor-specific noise by reconstructing each sensor from spatially redundant neighboring sensors. It assumes that signals of interest are represented across multiple sensors, whereas the targeted sensor noise is specific to individual sensors [1].
Usage#
from mne_denoise.sns import SNS
model = SNS(n_neighbors=4, n_iter=1)
clean = model.fit_transform(data) # data: (n_channels, n_times)
Key points#
n_neighbors, skip, and the neighbor-selection rule determine the local spatial predictors.
sample_weight, robust masking, and n_iter affect covariance estimation and repeated passes; inspect the fitted ranks and rejected-sample count.
NumPy input is channel-first. The estimator preserves supported MNE containers and uses the fitted channel layout for later transforms.
compute_sns_weights exposes the local weights; compute_sns provides the one-shot array operation.
SNS is not a generic spatial filter: it relies on sensor-specific noise and spatial redundancy being reasonable for the recording.