Time-Varying AR Spectral Estimation Using an Indefinite Matrix-Based Sliding Window Fast Linear Prediction


IEICE TRANSACTIONS on Fundamentals of Electronics, Communications and Computer Sciences   Vol.E97-A   No.2   pp.547-556
Publication Date: 2014/02/01
Online ISSN: 1745-1337
DOI: 10.1587/transfun.E97.A.547
Print ISSN: 0916-8508
Type of Manuscript: PAPER
Category: Digital Signal Processing
spectral estimation,  autoregressive model,  linear prediction,  fast algorithm,  sliding window,  indefinite matrix,  forgetting factor,  

Full Text: PDF>>
Buy this Article

A method for efficiently estimating the time-varying spectra of nonstationary autoregressive (AR) signals is derived using an indefinite matrix-based sliding window fast linear prediction (ISWFLP). In the linear prediction, the indefinite matrix plays a very important role in sliding an exponentially weighted finite-length window over the prediction error samples. The resulting ISWFLP algorithm successively estimates the time-varying AR parameters of order N at a computational complexity of O(N) per sample. The performance of the AR parameter estimation is superior to the performances of the conventional techniques, including the Yule-Walker, covariance, and Burg methods. Consequently, the ISWFLP-based AR spectral estimation method is able to rapidly track variations in the frequency components with a high resolution and at a low computational cost. The effectiveness of the proposed method is demonstrated by the spectral analysis results of a sinusoidal signal and a speech signal.