Skip to main content

InvBeamformerCompute

Namespace: INVLIB  ·  Library: Inverse Library

#include <inv/inv_beamformer_compute.h>

class INVLIB::InvBeamformerCompute

Shared computation routines used by both LCMV and DICS beamformers.

All methods are static — no state is kept between calls.

Core beamformer math.

Static Methods​

computeBeamformer(G, Cm, reg, nOrient, weightNorm, pickOri, reduceRank, invMethod, nn, W, maxPowerOri)​

Compute beamformer spatial filter weights from a leadfield and data covariance (or CSD).

This implements the core LCMV/DICS algorithm: Regularize and invert covariance

Reshape leadfield per source (n_channels x n_orient)

Optionally reduce leadfield rank

Select orientation (max-power, normal, or keep all)

Invert denominator G^T Cm^{-1} G

Apply weight normalization

Parameters:

  • G : const Eigen::MatrixXd & Leadfield matrix (n_channels, n_sources * n_orient).

  • Cm : const Eigen::MatrixXd & Data covariance or CSD matrix (n_channels, n_channels). Real-valued.

  • reg : double Regularization parameter (fraction of trace to add, e.g. 0.05).

  • nOrient : int Orientations per source: 1 (fixed) or 3 (free).

  • weightNorm : BeamformerWeightNorm Weight normalization strategy.

  • pickOri : BeamformerPickOri Orientation selection mode.

  • reduceRank : bool If true, reduce leadfield rank by 1 (for MEG sphere models).

  • invMethod : BeamformerInversion Denominator inversion strategy (matrix or scalar).

  • nn : const Eigen::MatrixX3d & Source normals (n_sources, 3). Used for max-power sign alignment.

  • W : Eigen::MatrixXd & Output spatial filter weights (n_sources * n_orient_out, n_channels).

  • maxPowerOri : Eigen::MatrixX3d & Output max-power orientations (n_sources, 3). Empty if not max-power.

Returns:

  • bool — True on success.

computePower(Cm, W, nOrient)​

Compute source power from spatial filter weights and a data covariance / CSD matrix.

power_i = trace(W_i Cm W_i^T)

Parameters:

  • Cm : const Eigen::MatrixXd & Data covariance or CSD (n_channels, n_channels).

  • W : const Eigen::MatrixXd & Spatial filter weights (n_sources * n_orient, n_channels).

  • nOrient : int Orientations per source.

Returns:

  • Eigen::VectorXd — Source power vector (n_sources).

symMatPow(X, p, reduceRank)​

Symmetric matrix power: X^p via eigendecomposition.

Parameters:

  • X : const Eigen::MatrixXd & Symmetric matrix.

  • p : double Power exponent (e.g. -1, -0.5).

  • reduceRank : bool Drop smallest eigenvalue before exponentiating.

Returns:

  • Eigen::MatrixXd — X^p.

Example​

No example of its own: Static numerical core (filter weights, power, regularised inverses) shared by InvLCMV and InvDICS; callers use those.

It is exercised by:

src/examples/ex_inv_beamformer/main.cpp
FiffCov dataCov = noiseCov; // noise plus the two sources
dataCov.data += g1 * g1.transpose() + g2 * g2.transpose();
const InvBeamformer lcmv = InvLCMV::makeLCMV(info, fwd, dataCov, 0.05, noiseCov,
                                             BeamformerPickOri::MaxPower, BeamformerWeightNorm::UnitNoiseGain);
const VectorXd lcmvPower = InvLCMV::applyLCMVCov(dataCov, lcmv).data.col(0); // one value per source

From ex_inv_beamformer.

src/examples/ex_inv_beamformer/main.cpp
// A single real cross-spectral density at 10 Hz equal to the data covariance gives the LCMV power.
const VectorXd frequencies = VectorXd::Constant(1, 10.0);
const InvBeamformer dics = InvDICS::makeDICS(info, fwd, {dataCov.data}, frequencies, 0.05, true, noiseCov,
                                             BeamformerPickOri::MaxPower, BeamformerWeightNorm::UnitNoiseGain);
const VectorXd dicsPower = InvDICS::applyDICSCsd({dataCov.data}, frequencies, dics).data.col(0);

From ex_inv_beamformer.

Authors of this file​