Signal model analysis via model-critical methods

It is shown that model-critical procedures provide a means to scrutinize an assumed parametric statistical model by varying the way the data are processed for repeated fits to the model. The criticism of the data is accomplished using the generalized likelihood function for the assumed probability density of the data. The degree of criticism is controlled by a user specified constant c. The model-critical parameter estimates are obtained by maximization of the generalized likelihood function. When c=0, no criticism is performed and maximum likelihood estimates are obtained. These procedures can be indicated if any model assumptions have been violated. Model critical estimation procedures are presented for autoregressive (AR) models. The analysis of an AR example is presented