Abstract
This paper proposes a general and computationally efficient parametric model-based framework for recursive frequency/spectrum estimation and feature detection of nonstationary signals, which may contain different extents of nonstationarities and impulsive components. The estimation of time-varying frequency or spectrum is formulated as a time-varying linear model identification problem, where the spectral information is estimated from the model coefficients. We then employ a QR-decomposition-based recursive least M-estimate (QRRLM) algorithm for recursive estimation of the time-varying model coefficients in impulsive environment using M-estimation. New variable forgetting factor (VFF) schemes are developed to improve the tracking performance of the QRRLM method in nonstationary environment and we use theoretical derivation and simulations to prove that the proposed VFF schemes can approach the optimal VFF selection. The resultant VFF-QRRLM algorithm is able to restrain and isolate impulsive components whereas it is able to handle different extents of spectral variations. Simulation results show that the proposed VFF-QRRLM algorithm is more robust and accurate than conventional recursive least squares-based methods in estimating both time-varying narrowband frequency components and broadband spectral components with impulsive components. Potential applications of the proposed method can be found in power quality monitoring, online fault detection and speech analysis.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have