Minimax robust nonstationary signal estimation based on a \(p\)-point uncertainty model (Q5942714)

The paper studies the problem of time-varying Wiener filtering for nonstationary processes. In particular it proposes a robust signal estimator which guarantees constant performance within prescribed uncertainty classes for the second order statistics of the model. A generalization of the so-called \(p\)-point uncertainty model is used, i.e. only mean subspace energies (relative to a given partition of the signal space) are prescribed. A time-frequency formulation and an efficient causal implementation are also discussed. Finally numerical simulations and a real data example are provided.