deepmr.svd

Contents

deepmr.svd#

deepmr.svd(input, ncoeff, axis)[source]#

Perform SVD compression of a signal.

The routine returns the SVD subspace basis, the compressed signal and the explained variance of the subspace.

Parameters:
  • input (np.ndarray | torch.Tensor) – Input signal (real- or complex-valued).

  • ncoeff (int) – Number of subspace coefficients to be retained.

  • axis (int) – Compression axis.

Returns:

  • basis (np.ndarray | torch.Tensor) – Subspace basis of shape (input.shape[axis], ncoeff).

  • output (np.ndarray | torch.Tensor) – Compressed signal of shape (…, ncoeff, …).

  • explained_variance (float) – Explained variance of the subspace. Values close to 100% indicates that information content of the signal is preserved despite compression.