what algorithm does matlab ppca use? How does it resolve the rotational redundancy?

Background

I have read the original PPCA paper 1999 Bishop, which is MLE-PPCA.

I think original 1999 paper does not solve the problem of parameter redundancy for MLE-PPCA, i.e., the R matrix in original paper can cause W to be arbitrary in numerical process.

Question

How does matlab ppca solve this problem? I have tested the matlab PPCA, and I found the W is pretty aligned with the principle component computed from sample covariance just like standard PCA .

I know there are serveral algorithm can do this, just being curious to ask what magic does matlab do.