By switching between adaptive filter algorithms/structures one can have an adaptive filter system that produces fast convergence and low mean square error (MSE). An adaptive filter algorithm/structure for which convergence is not affected by the eigenvalue ratio of the input is used to produce fast convergence, while an adaptive filter that yields low MSE, low computational load, and good numerical stability is used otherwise. Switching is done when the fast convergence structure has converged, as defined by the average squared error being small and approximately constant for several iterations. By applying a computationally reduced order technique, one can also reduce the computational load and force the switching structure to reach the steady state faster.
展开▼