Strongly consistent coefficient estimate for errors-in-variables models

For the single-input-single-output (SISO) errors-in-variables system it is assumed that the system input is an ARMA process and that the driven noise of the system input and the observation noise are jointly Gaussian. The two-dimensional observation made on system input and output is represented as a two-dimensional (2D) ARMA system of minimum phase driven by a sequence of 2D i.i.d. Gaussian random vectors (innovation representation). It is shown that the resulting ARMA system is identifiable, i.e., its coefficients are uniquely defined under reasonable conditions. Recursive algorithms are proposed for estimating coefficients of the ARMA representation including those contained in the original SISO system. The estimates are proved to be convergent to the true values with probability one and the convergence rate is derived as well. The theoretical results are justified by numerical simulation.