Abstract: Stationary subspace analysis is a recently developed technique for signal processing and data analysis, which is capable of separating stationary source signals from observed multidimensional nonstationary signals. The standard algorithm for stationary subspace analysis stems from the gradient descend over the Stiefel manifold, which converges slow and is time-consuming. In order to overcome those drawbacks, an iterative formula is constructed on the basis of the first-order optimality condition related to optimization problems over Stiefel manifolds, and thereby a novel fixed-point algorithm is proposed for stationary subspace analysis. Computer simulations show that the proposed algorithm is able to separate the stationary sources effectively with a better separation performance than that of the existing fixed-point algorithm, and in contrast to the standard algorithm that is based on the gradient descend over the Stiefel manifold, the proposed algorithm converges faster and is less time-consuming.

Key words: stationary subspace analysis, fixed-point iteration, Stiefel manifold optimization, stationarity, nonstationarity