An algorithm for RLS identification parameters that vary quickly with time

An alternative to the standard recursive least-squares (RLS) algorithm for fixed-order systems with exponential data weighting is presented. The approach uses Givens orthogonal transformations to update the Cholesky factor of the information matrix without ever needing to form it. The resulting algorithm gives higher-precision control and is less sensitive to ill-conditioning when compared to other reported approaches. It is demonstrated by an example that ill-conditioned problems with parameters that vary quickly can be modified to stabilize erratic parameter fluctuations