Identifiability and persistent excitation in full matrix fraction parameter estimation (Q1361333)

The authors consider multivariable systems that are described by \[ y_k+ \sum^{n_a}_{i=1} A_iy_{k-i}= \sum^{n_b}_{j=1} B_ju_{k-j},\tag{1} \] where \(y_k\) and \(u_k\) are column vectors, and \(A_i\) and \(B_j\) are matrices. The transfer function of the model (1) is given in left matrix fraction description (MFD) form by \(H(q^{-1})= A^{-1}(q^{-1})B(q^{-1})\), where \(A(z)= I+ A_1z+\cdots+ A_{n_a}z^{n_a}\) and \(B(z)= B_1(z)+\cdots+ B_{n_b}z^{n_b}\). If the degrees of the matrices \(A(z)\) and \(B(z)\) are the maximum degrees, then MFD is called the full matrix fraction description (FMFD). They give a sufficient and necessary condition for the FMFD being identifiable and a bound relating the persistence of excitation of the input signal and the persistence of excitation of the input-output pair in fully parameterized matrix fraction of models, which is a measure of the robustness of identification algorithms against noise and parameter variations. Moreover, they analyze the relationship between the above persistent excitation conditions and the identifiability of FMFD models. Two simple examples are provided to illustrate the theoretical results.