A General Prewhitening Procedure for Process and Measurement Noises

In data reconciliation and gross error detection as in quality control charts most methods assume that the data are serially independent. This assumption is convenient for mathematical treatment, but is often contradicted by experimental evidence. Recent work (Kao, et al., 1990, 1991) examines alternative procedures for mitigating such effects. One of these procedures is to prewhiten the process data before applying the usual data treatment methods. Prewhitening may be carried out using the Bryson and rfenrikson method (1968), which is applicable to first order autoregressive models. In this note we propose a prewhitening procedure which is applicable to any autoregressive moving average (ARMA) model, and which causes only a modest increase in the state space dimensions.