Abstract
When the Least Squares (LS) method is applied to parameter identification of general systems, it yields biased estimates. In order to remove this bias the Generalized Least Squares (GLS) method was proposed. But the conventional methods of the GLS have some defects. Namely, it is time consuming to transfer each input and output data into filter data every time the estimation algorithm is iterated, and it takes much time to calculate the coefficients of the error-whitening-filter.
In this paper we propose a new GLS estimation algorithm compensating for these defects. This new method has two advantages. The first is that it easily derives the coefficients of the error-whitening-filter using a simple recurrence formula. The second is an introduction of self and cross correlation functions of input and output data. If these correlation functions are calculated in advance, it is not necessary to transfer each input and output data into filter data using these correlation functions when the estimation algorithm is iterated. Because of these two advantages the calculation time for unbiased estimates can be reduced effectively.
