Unbiased parameter estimation of non-stationary signals on the block processing

Abstract

The authors present a nonlinear nonstationary (NN) model which represents time-varying characteristics of interest as the evolution over successive blocks in block processing. The NN model assumes that a nonstationary signal consists of a time-invariant component and a time-varying component over blocks. A set of parameters estimated up to the last block is used to model the time-varying parameters in the current block. Subtracting the time-varying component just modeled from the observed signal provides a transformed signal in the current block. The least-squares (LS) estimation with respect to the transformed signal again gives a new set of parameters. As a result less variance and unbiased estimation of time-varying parameters are achieved.<>