Diagonal Loading for Pattern Control
Diagonal Loaded Sample Covariance Matrix: C = XHX + kI = XLH XL
Adding a scaled identity matrix to the sample covariance can be accomplished by augmenting the training data matrix with the square root of the scaled identity matrix (which is also a scaled identity matrix).
Training data matrix: XLH =
The diagonal load matrix must be scaled to account for the number of training samples, the noise level, and the level of diagonal load power relative to the noise level.
k = (Noise amplitude) x sqrt(# of samples) x sqrt(diagonal load power factor)