Application of ARMA modeling to multicomponent signals

Abstract

This paper investigates the problem of estimating the parameters of a multicomponent signal observed in noise. The process is modeled las a special nonstationary autoregressive moving average (ARMA) process. The parameters of the multicomponent signal are determined from the spectral estimate of the ARMA model The spectral lines are closely spaced and the ARMA model must be determined from very short data records. Two high-resolution ARMA algorithms are developed for determining the spectral estimates. The first ARMA algorithm modifies the extended Prony method to account for the nonstationary aspects of noise in the model.For comPonents signals with good signal to noise ratio (SNR) this algorithm provides excellent results, but for a lower SNR the performance degrades resulting in a loss in resolution. The second algorithm is based on the work of Cadzow. The algorithm presented overcomes the difficulties of Cadzow's and Kaye's algorithms and provides the coefficients for the complete model not just the spen ral estimate. This algorithm performs well in resolving multicomponent signals when the SNR is low.