Robust AR analysis of nonstationary discrete-time signals

A new method for estimating time-varying parameters of nonstationary AR signal models, based on robust recursive least squares with variable forgetting factors (VFF), is described. The robust estimator differs from the conventional one by the insertion of nonlinear residual transformation, which has to cut off the spiky noise, while the VFF´s are adapted to the signal via the generalized likelihood ratio detection scheme.