STAR recursive least square lattice adaptive filters

Abstract

The recursive least square lattice (LSL) algorithm based on the newly developed scaled tangent rotations (STAR) is derived. Similar to other recursive least square lattice algorithms for adaptive filtering, this algorithm requires only O(N) operations. This algorithm also preserves the desired properties of the STAR recursive least square (STAR-RLS) algorithm. Specifically, it can be pipelined at fine-grain level. To this end, a pipelined version of the STAR-LSL (referred to as PSTAR-LSL) is also developed. Computer simulations show that the performance of the STAR-LSL algorithm is as good as the QRD-LSL algorithm. The finite precision error properties of the STAR-LSL algorithm are also analyzed. The mean square error expressions show that the numerical error propagates from stage to stage in the lattice, and the numerical error of different quantities in the algorithm varies differently with /spl lambda/. This suggests that different word lengths need to be assigned to different variables in the algorithm for best performance. Finally, finite word length simulations are carried out to compare the performances of different topologies.