Robust time-varying Wiener filters: Theory and time-frequency formulation

We propose a minimax robust time-varying Wiener filter that is based on a novel uncertainty model for nonstationary random processes. This filter maintains a certain performance for all second-order statistics within prescribed uncertainty classes. Furthermore, it requires less detailed prior knowledge than the ordinary Wiener filter. We also present an intuitively appealing time-frequency formulation of the robust time-varying Wiener filter in which signal subspaces are replaced with time-frequency regions.