SIGNAL MODELING WITH NON-UNIFORM TOPOLOGY LATTICE FILTERS Sacha KRSTULOVIC´

Fr´ed´eric BIMBOT

IDIAP C.P. 592 - CH-1920 Martigny - Switzerland [email protected]

IRISA - Campus Beaulieu, 35042 Rennes - France [email protected]

ABSTRACT This article presents a new class of constrained and specialized Auto-Regressive (AR) processes. They are derived from lattice filters where some reflection coefficients are forced to zero at a priori locations. Optimizing the filter topology allows to build parametric spectral models that have a greater number of poles than the number of parameters needed to describe their location. These NUT (Non-Uniform Topology) models are assessed by evaluating the reduction of modeling error with respect to conventional AR models.

5 #&% 
3  is the transfer function of an 6-! order forward predictor, modeling the current sample as a linear combination , past samples; of 6 5 ) %
 is the3 transfer function of an 6-7 order backward predictor, modeling the 6 ! past sample as a linear combi, future samples; nation of 6 5 . %8' is the reflection ,  . to grow the pre
 coefficient
allowing dictors from order 6 to order 6:9 3 from step 6 of this recursion, a known number Suppose ,  the reflection coefficients following . %(' are fixed
!E ; %(' that E @ SZY?JL @to STzero L . Theofequivalent lattice flow chart looks like : CFEHG QR Q + G G G R @ SZY?JML @ STL Q RO