Detection and tracking of superimposed non-stationary harmonics

Abstract

Detection and estimation of multiple narrowband components in a time-series is a difficult signal processing problem that shows up in many applications. The authors propose a parametric approach to the problem and examine an algorithm based on singular value decomposition to estimate and track the parameters. The model suggested for each narrowband component is a sinusoid with slowly varying amplitude and frequency. The algorithm proposed provides the noise reduction associated with very long averaging intervals, and yet tolerates significant drift of parameters (instantaneous amplitudes and frequencies) over the averaging interval.<>