A tree-structured piecewise linear filter with recursive least-squares adaptation

Tree-structured piecewise linear adaptive filters have several potential advantages over other nonlinear filtering structures. First, they can to a large extent exploit standard training algorithms at each node to identify the corresponding conditional linear models. Second, they allow for efficient selection of the subtree and the corresponding piecewise linear model of appropriate complexity. Overall, the approach is conceptually simple and computationally efficient. In this paper we present a recursive least squares adaptation algorithm for the class of tree-structured piecewise linear filters.