Title :
Principal components algorithms for ARMA spectrum estimation
Author_Institution :
Coord. Sci. Lab., Illinois Univ., Urbana, IL, USA
fDate :
4/1/1989 12:00:00 AM
Abstract :
Principal components algorithms are presented for the problem of fitting an ARMA model to a given segment of a sample sequence of a discrete-time stochastic process, and the model is used to estimate the process power spectrum. To reduce the effects of finite-wordlength errors, balanced state-pace parameterization of the ARMA model is used instead of the more popular difference equation parameterization. Model identification is formulated as a problem of selecting a partial state to span approximately an apparently large-dimensional information interface between the past and the future of the process. Different criteria are used to measure the quality of the approximation, which leads to principal-components algorithms for the problem that are based on singular value decomposition
Keywords :
estimation theory; identification; spectral analysis; state-space methods; statistical analysis; stochastic processes; time series; ARMA spectrum estimation; discrete-time stochastic process; finite-wordlength errors; information interface; model identification; parameterisation; power spectrum; singular value decomposition; Approximation algorithms; Autoregressive processes; Context modeling; Digital filters; Frequency estimation; Signal processing algorithms; Spectral analysis; Stochastic processes; Transfer functions; Working environment noise;
Journal_Title :
Acoustics, Speech and Signal Processing, IEEE Transactions on
