Robust estimation of signal parameters with nonstationary and/or dependent data

A new approach toward robust parameter estimation is presented. As opposed to schemes based on classical saddlepoint techniques, the methods apply quite naturally to dependent and/or nonstationary data and also have the advantage of providing a quantitative measure of the degree of robustness offered by a particular estimator. It is shown why censoring observations of large magnitude does indeed impart robustness, and it also shown how nontraditional schemes have certain advantages, particularly when trading off performance and robustness. The application of the results is further demonstrated by including several illustrative examples. >