Recursive lattice forms for spectral estimation

A class of lattice prediction filters is proposed for high resolution spectral estimation. The square-root normalized lattice recursions are used to estimate a set of reflection coefficients from the data. The lattice variables determine the coefficients of a least-squares predictor, from which the spectrum can be evaluated. The prewindowed and sliding window (covariance) cases are considered for both AR and ARMA spectra. The behavior of the proposed spectral estimator is illustrated by simulation results.