The normalized least-squares order-recursive lattice smoother

Abstract

This paper introduces a variance- and angle-normalized version of the least-squares order-recursive lattice (LSORL) smoother, say, the normalized LSORL smoother, via a geometric approach, The normalized LSORL smoother has excellent round-off noise properties and inherits all the other advantages of the normalized LSORL as well. Simulation results show that the normalized LSORL smoother outperforms the existing LSORL and QRD-based least-squares lattice smoother algorithms when finite-precision arithmetic is used. (C) 2002 Elsevier Science B.V. All rights reserved.