Stationary subspace analysis based on second-order statistics

In stationary subspace analysis (SSA) one assumes that the observable p-variate time series is a linear mixture of k-variate nonstationary and (p−k)-variate series. The aim then to estimate unmixing matrix which transforms observed multivariate onto components. classical approach data are projected subspaces by minimizing Kullback–Leibler divergence between Gaussian distributions, method only detects nonstationarities in first two moments. this paper we consider SSA more general setting propose methods able detect mean, variance autocorrelation, or all them. Simulation studies illustrate performances proposed methods, it shown especially three types performs well various settings. concluded with an illustrative example.