Identification of continuous-time AR processes from unevenly sampled data (Q1613177)

The objective of this paper is the identification of continuous-time AR systems from discrete-time data, where data are observed unevenly in time. The authors do not use consistent and statistically efficient maximum likelihood prediction error methods because they involve great computational complexity. A direct approach is chosen in which the derivatives are substituted by a discrete-time difference operator, for example by replacing differentiation by a delta operator. Two least-squares methods are introduced. They give consistent estimates for the posed problem. Consistency means that the estimated parameters tend to the true ones as the number of data goes to infinity and the sampling intervals approach zero.