decoding_ssd.cpp File Reference

Implementation of the Spatio-Spectral Decomposition narrowband enhancer. More...

#include "decoding_ssd.h"
#include <Eigen/Eigenvalues>
#include <algorithm>
#include <cmath>
#include <stdexcept>

Include dependency graph for decoding_ssd.cpp:

Go to the source code of this file.

Detailed Description

Implementation of the Spatio-Spectral Decomposition narrowband enhancer.

SPDX-License-Identifier: BSD-3-Clause Copyright (c) 2026 MNE-CPP Authors

Author

Christoph Dinh chris.nosp@m.toph.nosp@m..dinh.nosp@m.@mne.nosp@m.-cpp..nosp@m.org

Since

2.2.1

Date

May 2026

Internally band-passes the input continuous data twice — once for the signal band and once for the flanking noise band — with a zero-phase windowed-sinc FIR (forward + reverse pass), computes the two covariance matrices, shrinks the noise covariance towards a scaled identity by regParam, and solves the generalised eigenproblem \(\Sigma_s w = \lambda \Sigma_n w\) via Eigen's GeneralizedSelfAdjointEigenSolver. The leading n_components eigenvectors are stored as filters; their pseudoinverse columns are stored as patterns and the eigenvalues themselves are kept verbatim as the per-component signal-band-to-noise-band power ratio that the caller uses to decide on the effective rank.

DecodingSsd::transform returns the projection of the input into the (lower-dimensional) signal subspace, while DecodingSsd::apply performs the low-rank back-projection \(X_{\text{clean}} = A_{:,1:k} W_{1:k,:} X\) to produce a sensor- space denoised reconstruction with the broadband background suppressed but the narrowband oscillation of interest preserved. The bandpass kernel is generated on the fly per call: this trades a negligible amount of compute for the freedom to reuse the same fitted model with different sampling rates downstream.

Definition in file decoding_ssd.cpp.