Time scaling for observer design with linearizable error dynamics. (Q1428699)

The authors consider the nonlinear observed dynamics \[ \Sigma:\dot x={dx\over dt}= f(x),\qquad y= h(x), \] where \(x(\cdot)\in \mathbb{R}^n\), and \(y(\cdot)\in \mathbb{R}\) is the measurement. They solve the problem of transforming, via coordinate change and time scaling, the dynamics \(\Sigma\) to a form which admits an observer with a linear error dynamics. The class of admissible time scalings used in this paper is given by the equation \[ d\tau= s(y(t))\,dt, \] where \(s\) is a real-valued positive function which depends only on the output of the observed dynamical system. Necessary and sufficient conditions are given which state that \(\Sigma\) is (locally) equivalent under a diffeomorphism \(z= \varphi(x)\) to \[ {dz\over dt}= s(y)(Az+ Ky),\qquad y= Cz= z_1,\tag{1} \] where \[ A= \begin{pmatrix} 0 & 1 &\cdots & 0\\ \vdots & \ddots & \ddots & \vdots\\ \vdots &&\ddots & 1\\ 0 &\cdots & \cdots & 0\end{pmatrix},\qquad k= \begin{pmatrix} k_1\\ \vdots\\ \vdots\\ k_n\end{pmatrix}, \] \(C= (1,0,\dots, 0)\) and \(s(y)\) is a nonvanishing real-valued function. Necessary and sufficient conditions are given also which state that \(\Sigma\) is (locally) equivalent under a diffeomorphism \(z= \varphi(x)\) to \[ {dz\over dt}= s(y)(Az+ \gamma(y)),\qquad y= Cz= z_1,\tag{2} \] where \[ \gamma(z_1)= \begin{pmatrix} \gamma(z_1)\\ \vdots\\ \gamma_n(z_1)\end{pmatrix}. \] For the two normal forms (1) and (2) the observer design problem can be easily solved. The authors compare the class of systems that can be put into observer form under a state-space diffeomorphism and time scaling with the one for which a state-space and an output-space diffeomorphism are used. Starting from \(n= 3\) those classes are different.