Data Processing for System Identification

The quality of black box models highly depends on the characteristics of the signals applied to the identification algorithms. In this chapter it is showed how measured data can and should be pre-processed before applying linear system identification techniques. We discuss how to remove disturbances from the data (e.g., spikes, drifts, offsets, differences in input and output power) and how to compensate for long time delays and non-linearities. We describe simple but effective techniques to remove these degrading effects from the measured signals. Thus the quality of the identified model can be significantly improved.