Improved order-statistics-based noise power estimator

Abstract This paper proposes an Order-Statistics (OS) based method for the robust estimation of the noise power. By deriving the analytic expression of the OS method, one deduces appropriate weighting coefficient which minimizes the Mean Squared Error (MSE) of the noise power estimate. The derived method, referred to as improved Order-Statistics (iOS), has an MSE close to the optimal one while preserving the robustness of the standard OS method against outliers or strong interferences. Simulation results are provided to illustrate the iOS performance in terms of MSE in both homogeneous and heterogeneous cases.