FIR digital filters using LMS-derived difference coefficients

FIR digital filters have the advantages of linear phase and guaranteed stability. However the large number of coefficient multiplications often result in conventional direct-form implementations which are expensive and computationally complex. One approach to overcome this is to use low-order recursive section B(z)^-1 which is chosen so that the modified FIR section H´(z)=H(z)·˜B(z) is realised as a single section with coefficients which are reduced to small integers while providing the same overall transfer function H(z). The aim of this contribution is to present a new approach for the choice of B(z) based on a least-mean-square (LMS) approach. A recursive section is `tailor-made´ for the given filter response and does not depend on the identification and selection from a limited set of pre-defined B(z) functions