MATRIX PENCIL METHOD FOR ESTIMATION OF PARAMETERS OF VECTOR PROCESSES

Abstract

The paper considers one of the modern parametric methods of signal processing the matrix pencil method (MPM). The method makes it possible to eectively estimate signal parameters from its samples which is the sum of complex exponentials. The number of exponentials is not assumed to be known in advance and can also be estimated using a singular value decomposition of a matrix constructed from the signal samples. The object of this work is a vector process a set of signals having the same frequencies and damping factors (i.e., the same poles of the signal), but dierent complex amplitudes. Such signals occur, for example, when considering a phased array antenna, when it is necessary to evaluate the parameters of a signal generated by the same sources but coming from many antenna elements with their amplitudes and phases. A similar problem arises in the evaluation of signals parameters from two spatially-distributed motion sensors of a Coriolis owmeter. When processing several signals with classical MPM, we obtain several dierent sets of poles of these signals, which then, for example, need to be averaged to obtain the desired values of the poles assumed to be the same for all signals. The proposed modication of the MPM works with the entire set of signals at once, giving one set of signal poles, and at the same time it proves to be more eective both in terms of speed and in the accuracy of determining the signal parameters. The algorithms of the classical MPM and its modications for the vector process are presented, as well as numerical experiments with model and real signals taken from one of the commercially available Coriolis owmeters Du15. Experiments show that the proposed algorithm yields more accurate results in a shorter (approximately 1,5 times) time than the classical MPM