Recursive matrix pencil method

Abstract

The main concern for many applications (for example, for Coriolis Mass Flow Meter signal processing in two-phase flow conditions) is to track several domain poles of signals with minimum delay. The Matrix Pencil method (MPM) estimates the signal as a sum of complex exponentials. Creating a computationally efficient moving MPM implementation is of a high interest. The most computationally expensive step of the classical MPM is to calculate the singular value decomposition (SVD) of a matrix composed from the signal samples. When a new data point enters the data window, this matrix changes only slightly, and it is reasonable to find its SVD not directly but using the SVD of the old matrix and a low-rank SVD modification procedure. In this paper a well-known SVD modification procedure is adapted for MPM and a recursive version of MPM is proposed. The amended method is validated by numerical examples and is faster than the original suggesting it may be feasible to track signal parameters on-line. The errors accumulating over time due to the recursive calculation require the recursive MPM to be restarted periodically.