No label defined (Q4383441)

The impulse response of a dynamic system is represented by the simple poles time series \(h_k=\sum_{j=1}^n r_je^{s_jk\Delta t},\quad k=0,1,\ldots,\) where \(r_j\) and \(s_j\), \(s_i\neq s_j\) for \(i\neq j\), are parameters that hypothetically govern the system, \(\Delta t\) is the sampling interval and \(n\) is the order of the system. A predictor polynomial \(P\) of degree \(N\geq n\) predicts \(h_{j+N}\) on the basis of known values \(\{h_j,h_{j+1},\ldots,h_{j+N-1}\}\). The parameters \(s_j\) are called poles and can be extracted from \(n\) zero points of \(P\). Since \(n\) is not a priori known, there are \(N-n\) extraneous zeros without physical meaning. The paper presents a method for separating the signal zeros from the extraneous ones. Defining a matrix \(H\) of samples \(h_k\) and introducing the companion matrix \(C\) to the polynomial \(P\), the authors show that rank\((H)=n\) and \(C\) has \(n\) of its eigenvalues located at \(z_i=e^{s_i\Delta t}\), \(i=1,\ldots,n\), the others fall inside the unit circle. Prediction in the reverse direction leads to eigenvalues \(z_i=e^{-s_i\Delta t}\). Finally, analogous results are derived for time series with multiple poles.