Transform domain LMS adaptation of split-path filter

Investigates the application of a transform domain adaptive technique to a linear predictor that is configured in a split-path structure. The split-path predictor is composed of two linear phase filters connected in parallel, and thus two matrices are required for LMS (least mean square) adaptation in transform domain. An effective way of selecting these matrices is introduced which can provide a significant improvement in the overall performance. Theoretical analysis of the adaptive process and its convergence behavior is given. Computer simulations have demonstrated that, in terms of convergence speed, the proposed system is almost twice as fast as a tapped delay line transform domain predictive filter