Generalisation of linear prediction with autocorrelation method
Paavo Alku, Tom Bäckström, Carlo Magi · Electronics Letters · 2011
A generalisation of conventional linear prediction (LP) is proposed. The method implements the prediction using a parallel structure, where the input is first filtered with two FIRs having zeros at z=γ1 and z=γ2. Two symmetric LP polynomials are then defined from the pre-filter outputs. The generalised LP inverse filter is obtained by convolving the symmetric polynomials with fixed pre-filters with zeros at z=±1 and then summing the obtained polynomials. It is shown that by selecting −1≤γ2<γ1≤+1, the resulting all-pole filter is stable. Conventional LP is obtained by choosing γ1=1.0 and γ2=−1.0.