Robust linear filtering for multivariable stationary time series

We consider the problem of asymptotic noncausal linear filtering for multivariable second-order stationary time series, under spectral uncertainty in both the signal and the noise processes. The spectral uncertainty is modeled by e-contaminated and p -point classes. For the case where both the signal and noise spectra are e-contaminated, we assume that the eigenvectors of the corresponding nominal spectral density matrices are identical. The problem is formulated as a game, whose saddle point solutions are found and analyzed.