An adaptive normalized lattice algorithm for LS FIR multichannel filters with linear phase

A novel adaptive normalized lattice algorithm for the least-squares (LS) design of multichannel FIR (finite-impulse response) filters with symmetric impulse response is presented. A byproduct is an algorithm for the LS estimation of the unit step backward predictor, based on the simultaneous minimization of the forward and backward total squared error powers, which has successfully been used in MEM spectral analysis. An order-recursive scheme that computes the filters from the normalized lattice variables is also presented.<>