Time series discriminant analysis of AR (p) plus noise processes: a time domain approach

The problem of discrimination between two autoregressive processes of order p is considered when the observed time series is contaminated with an extra noise and the main discriminatory information is in the covariance structure rather than mean. An analytic discrimination rule is given based on likelihood ratio and its performance is examined. It is well-know that the distribution of the discrimination can be expressed in terms of a weighted sum chi-square random variables of one degree of freedom. The weights in the sum have to be calculated numerically. The approximated weights are calculated. It is shown that they are very close to the true values.

کلیدواژه ها:

Quadratic Discrimination ، AR(p) plus noise processes ، Band matrices ، Cumulants