Weighted least squares algorithm for the design of lattice-type perfect reconstruction cosine modulated filter banks

Abstract

This paper presents a weighted least squares (WLS) algorithm for the design of lattice-type perfect reconstruction (PR) cosine modulated filter banks. The design of such a filter bank is formulated as a WLS problem with respect to the lattice coefficients, incorporating the condition that a number of the zeros of the prototype filter must be at specific locations in the stopband. Our iterative algorithm needs only a few iterations to derive a set of lattice coefficients which optimizes this WLS problem, while providing flexible control of the prototype filter's stopband ripples. By specifying appropriate zeros of the prototype filter in the stopband, good designs with no DC leakage can be easily obtained.