Section: Research Program
Estimating the derivative of a (noisy) signal with a sufficient accuracy can be seen as a key problem in domains of control and diagnosis, as well as signal and image processing. At the present stage of our research, the estimation of the -th order time derivatives of noisy signals (including noise filtering for ) appears as a common area for the whole project, either as a research field, or as a tool that is used both for model-based and model-free techniques. One of the open questions is about the robustness issues (Indicator 3) with respect to the annihilator, the parameters and the numerical implementation choices.
Two classes of techniques are considered here (Model-based and Model-free), both of them aiming at non-asymptotic estimation.
In what we call model-based techniques, the derivative estimation is regarded as an observation problem, which means the software-based reconstruction of unmeasured variables and, more generally, a left inversion problem (Left invertibility deals with the question of recovering the full state of a system (“observation”) together with some of its inputs (“unknown input observers”), and also refers to algebraic structural conditions.). This involves linear/homogeneous/nonlinear state models, including ordinary equations, systems with delays, hybrid systems with impulses or switches (Note that hybrid dynamical systems (HDS) constitute an important field of investigation since, in this case, the discrete state can be considered as an unknown input.), which still has to be exploited in the finite-time and fixed-time context. Power electronics is already one of the possible applications.
Model-free techniques concern the works initiated by ALIEN, which rely on the only information contained in the output signal and its derivatives. The corresponding algorithms rely on our algebraic annihilation viewpoint. One open question is: How to provide an objective comparison analysis between Model-based and Model-free estimation techniques? For this, we will only concentrate on Non-Asymptotic ones. This comparison will have to be based on the three Indicators 1 (time), 2 (complexity) and 3 (robustness).