Implements simultaneous leave-one-sensor noise estimation followed by the source-space Wiener reconstruction of the SOUND algorithm.
Usage
soundClean(
x,
forward_model,
lambda = 0.1,
iterations = 10L,
tolerance = 0.01,
reference = "average",
analysis_window_ms = NULL,
epoch_start_ms = 0,
assay_name = NULL,
output_assay = "tms_sound",
overwrite = FALSE
)Arguments
- x
A pulse-cleaned
PhysioExperiment.- forward_model
A list containing a finite channel-matched
leadfield.- lambda
Positive source-reconstruction regularization.
- iterations
Maximum number of complete simultaneous updates.
- tolerance
Non-negative maximum relative noise change for convergence.
- reference
Currently exactly
"average".- analysis_window_ms
Optional closed noise-estimation interval.
NULLuses all samples except recorded pulse/recharge gaps.- epoch_start_ms
Time of the first epoch sample relative to the pulse.
- assay_name
Optional pulse-cleaned input assay.
- output_assay
Output assay name.
- overwrite
Whether to overwrite the selected input assay.
Value
A modified PhysioExperiment with the fixed SOUND cleaning matrix,
complete noise history, and convergence diagnostics.
Details
SOUND assumes diagonal sensor noise, an approximate channel-matched lead field, and i.i.d. source amplitudes for its regularized reconstruction. All channel noise estimates are updated simultaneously from the preceding round, and one final cleaning matrix is applied to every sample and trial. Pulse samples are excluded from noise estimation. Regularization is not a significance threshold and is not optimized against the reported TEP peaks.
References
Mutanen TP, Metsomaa J, Liljander S, Ilmoniemi RJ (2018). Automatic and robust noise suppression in EEG and MEG: The SOUND algorithm. NeuroImage, 166:135-151. doi:10.1016/j.neuroimage.2017.10.021