Title :
Autocorrelation prediction
Author_Institution :
Bell Laboratories, Holmdel, New Jersey
Abstract :
Autocorrelation Prediction (AP) has been shown to be an effective technique for Pole-Zero modeling. This paper develops a new linear method for identifying a stable Pole-Zero model whose spectrum matches the envelope of a given spectrum. All the operations are performed in Autocorrelation domain, using no Fourier transformations. At one extreme, Autocorrelation Prediction reduces to a linear method for All-Zero modeling. At the other extreme, AP becomes the well-known Linear Prediction (LP). AP can also automatically determine the lowest denominator and numerator orders required to model efficiently the given spectral envelope. Spectra whose envelopes have deep valleys are shown to be matched more accurately at the valleys using AP rather than LP.
Keywords :
Autocorrelation; Computer science; Data compression; Fourier transforms; Nonlinear equations; Poles and zeros; Predictive models;
Conference_Titel :
Acoustics, Speech, and Signal Processing, IEEE International Conference on ICASSP '77.
DOI :
10.1109/ICASSP.1977.1170282