Improved matrix pencil methods

We study the problem of estimating signal parameters from a noisy data sequence containing superimposed damped sinusoids. We propose three novel methods by combining the reduced-rank Hankel approximation and the matrix pencil method. We demonstrate that two of the proposed methods significantly outperform both the original matrix pencil method and the modified Kumaresan-Tufts (1982) method, especially at low signal-to-noise ratio.