Skip to Main Content
A fast convergence adaptive filtering algorithm based on a split-path configuration and the discrete Walsh transform (DWT) is presented. An adaptive FIR predictor is first mapped to the DWT domain. It is then split into a pair of linear phase subfilters connected in parallel. The filter parameters are adapted using the LMS approach with different convergence step sizes to achieve better learning performance. Walsh-Hadamard ordering is applied to simplify and to speed up the algorithm.