A periodically time-varying minimal partial realization algorithm based on twisting (Q1307123)

Given the impulse response \(h(i)\) of a linear system the minimal partial realization problem is to find a description of the system with minimal McMillan degree and that interpolates the given impulse response for \(i=0,\ldots,P\). A linear system with input \(u(t)\) and output \(y(t)\) can be described in the form of a difference equation \(y(t)+b_1y(t-1)+\cdots+b_Ly(t-L)=a_0 u(t)+\cdots+a_Lu(t-L)\). The problem in this form is to find \(b_k\) and \(a_k\) with \(L\) minimal. In the time-varying case, these coefficients depend on the time \(t\). In this paper, it is assumed that the system is \(P\)-periodic, hence one can assume that \(b_k(t)=b_j(t+P)\) and similarly for \(a_j\). One should also assume that the impulse response is given for an impulse at time \(i-1\) with \(i=0,\ldots,P\). It is shown how a scalar system of this type can be recast into a multivariate time invariant problem, by putting \(u\) and \(y\) into one vector. This is called twisting. The result is an algorithm that somehow resembles the algorithm for the time invariant case. Although developed for scalar systems, the results are easily transferred to the multivariate case.