Least squares identification of non-stationary MA systems

The correlation analysis based methods are not suitable for identifying parameters of non-stationary MA systems, for which two algorithms are developed, an iterative and a recursive multi-innovation least squares ones. The basic idea is to replace immeasurable noise terms in the information vector by the estimation residuals, which are computed also according to the parameter estimates. This is a hierarchical computation process. Furthermore, the conditions of convergence of the parameter estimation by the recursive algorithm are derived. The simulation results validate the algorithms proposed.