System Identification Algorithm for Non-Uniformly Sampled Data.

Considerable effort has been devoted to the development of algorithms for identification of parsimonious discrete time models from noisy input/output data sets since this facilitates controller design. Several methods, such as nuclear norm minimization, have been used to provide approximate solutions to this non-convex problem. However, even though the field of continuous time system identification is now mature, results on parsimonious model identification of continuous time systems are still very limited. In this paper, an atomic norm minimization method is proposed for this purpose that can handle non-uniformly sampled data without preprocessing. The proposed approach provides an efficient way to use noisy, non-uniformly sampled data to determine a reliable, low-order continuous time model. Numerical performance is illustrated using academic examples and simulated behavioral data from a smoking cessation study.