Suppose that you have a system: \begin{cases}\dot{x}(t)=A(u(t))x(t)+d_1(t)\\y(t)=f(x(t))+d_2(t),\end{cases} where $A:\mathbb{R}\longrightarrow\mathbb{R}^{n\times n}$ is a matrix function, $u:\mathbb{R}\longrightarrow\mathbb{R}$ is a known input and $d_1:\mathbb{R}\longrightarrow\mathbb{R}^n$, $d_2:\mathbb{R}\longrightarrow\mathbb{R}$ are unknown disturbance functions. The function $f:\mathbb{R}^n\longrightarrow\mathbb{R}$ is not linear and $y$ is the output. There is no more information.
If your goal is to approximate the state $x$ of the system with an observer, which observer would you use? Can you use the Extended Kalman Filter? how would you choose the matrices usually denoted by $Q$ and $R$?