Frequency-domain Approach to Continuous-time System Identification: Some Practical Aspects

Since the end of the 1950s - beginning of the 1960s - the control society developed for its control designs a technique to build descrete-time models of continuous-time processes. Due to its overwhelming success a classical time-domain school emerged, and its authority in the field of system identification was soon widely recognised. The continuous-time identification methods developed in the early days of system identification got into a tight corner, and were 'forgotten' for several decades. Nowadays many people select discrete-time models and classical time-domain identification methods to solve their particular moelling problems. If the input is zero-order-hold, then discrete-time models are the natural choice, however, in all other cases continuous-time models are the natural choice, however, in all other cases continuous-time models might be preferred. Also, if the final goal is physical interpretation, then continuous-time modelling is the prime choice. Since physical interpretation is mostly the main motivation for continuous-time modelling, special attention is paid in this chapter to some - often implicitly made - basic assumptions: the inter-sample behaviour of the excitation (zero-order-hold or band-limited), the measurement setup (zero-order-hold or band-limited, calibration of the systematic errors, open loop versus closed loop ...), the noise model (discrete-time or continuous-time, parametric or non-parametric), the stochastic framework (generalised output error or errors-in-variables), the sampling scheme (uniform or non-uniform), and the linearity (influence of non-linear distortions). Within this framework the advantages and drawbacks of the existing continuous-time identification methods are discussed. Several practical aspects are illustrated on two real measurements examples. The chapter concludes with some guidelines for the user.