Estimation of continuous-time models from sampled data via the bilinear transform

This paper presents a new technique for estimating Continuous-Time (CT) Linear Time-Invariant (LTI) models from discrete data. The method uses MOESP to estimate the order and parameters of a Discrete-Time (DT) system. The bilinear transform is then used to calculate an equivalent CT model. This gives rise to process zeros. Extensive simulation studies have demonstrated that there are few process zeros when no noise and quantization are present. However, when quantization and noise are present process zeros always lie above 0.5 times the Nyquist rate. Hence, in converting from DT to CT it is necessary to discard zeros above 0.5 times the Nyquist frequency, to yield an accurate CT model. With two experimental examples the authors demonstrate that method does indeed work