| Contents |
| The general framework of control theory is very dependent on the outcomes of previous and/or simultaneous system identification. Given a set of standard inputs, we can proceed by deriving a describing function for the input-output functional correlation. However, in an experimental setup this need is very difficult to be satisfied. First, from a general point of view we do not have total access to the inner state of the system to be controlled. Certainly, our data is mainly constructed by partial observations of the underlying system dynamics. Second, partial observations should be modeled in a fast and accurate way to achieve controlability and observability. In fact, if the considered systems are nonlinear and time-varying then the describing function must be adapted as the system evolves. Furthermore, the input-output relationship could be history dependent, which results in an adequate system identification in case we apply classical identification techniques. As a possible alternative we can construct data-driven control procedures guided by efficient and precise tools for detecting and characterizing events automatically. In this communication we discuss this possibility by means of our recent results on the application of symbolic dynamics and time-frequency signal processing to such a goal. |
|