Multiple transient estimation using bootstrap and subspace methods

This paper presents a scheme for estimation of multiple overlapping transients in additive white noise using a single sensor. The scheme is based on three main ideas: a pole estimation method based on Kung´s algorithm; a model order estimation scheme by bootstrap/SVD; and an SVD-based change detector. By combining these schemes it is demonstrated that multiple transients with different starting points can be estimated. The method is tested on a number of simulations. The order estimation method is shown to be more accurate than AIC, for high SNR and short segments