Application of the structured total least norm technique in spectral estimation

In the problem of estimating parameters of exponentially damped sinusoids, an improved variant of Kung's method, called HTLS and based on the use of a Hankel data matrix, the singular value decomposition and the total least squares technique, has been proposed and proven to be accurate and efficient. In this paper, a more accurate estimator HTLN is proposed. It starts from the same Hankel data matrix, but uses a new technique, called structured total least norm, prior to the HTLS estimator. This technique computes the solution of a structured overdetermined linear system, AX ≈ B, with possible errors in both A and B. The better accuracy of the STLN and HTLN techniques is shown by means of computer simulations.