I'm trying to understand concepts on Kalman Filters. Consider the overdetermined system $Ax=y$; $$\begin{bmatrix}1 \\ 1 \\ 1 \\1 \end{bmatrix} x = \begin{bmatrix} 3 \\ 5 \\ 4 \\ 8 \end{bmatrix}$$

Let $R$ be the errors in measurements $y$ and $P$ be the errors in the estimate of $\hat{x}$. Assume that the uncertainty in measurements, $\sigma^{2}$. Then, $R=\begin{bmatrix}\sigma^{2} & 0 & 0 & 0\\ 0 & \sigma^{2}& 0 & 0 \\ 0 & 0 &\sigma^{2}& 0 \\ 0 & 0 & 0 & \sigma^{2}\end{bmatrix}$.

Now consider Linear Kalman filter equations

\begin{aligned} x_{k} &=F_{k-1} x_{k-1}+G_{k-1} u_{k-1}+q_{k-1} \\ y_{k} &=A_{k} x_{k}+r_{k} \\ q_{k} & \sim N\left(0, Q_{k}\right) \\ r_{k} & \sim N\left(0, R_{k}\right) \\ \end{aligned}

Assume a static model, with trivial dynamics $F = I$, so that $F_{k+1} = F_k$ and let’s assume again that the variable $x$ is directly observable, so that $A = I$, and $A_{k+1} = A_k$. Also, assume the uncertainty in both model, $F$, and observation, $A$, are both equal to $\sigma^{2}$.

I'm trying to identify the parameters of Kalman filter equations through this system. Since the model is static with trivial dynamics, I guess $q_{k-1} = 0$. I'm trying to identify $G_{k-1}$ and $u_{k-1}$ for this system. Can you please help me to identify other parameters?

How can we use the matrix $\begin{bmatrix} 3 \\ 5 \\ 4 \\ 8 \end{bmatrix}$ to calculate $x_2^{+}, x_{2}^{-1}$ and other parameters?

There are similar questions in Chapter 2 of "Neural Control Engineering: The Emerging Intersection between Control Theory and Neuroscience". Thank you very much!

My working so far: Correct me if I'm wrong! Given that the model dynamics are trivial, we can assume that $$q_{k-1} = G_{k-1} =u_{k-1} = 0$$. Also, given that $F=A=I$. Then, Kalman filter equations become, $$x_k = x_{k-1}$$ $$y_k = x_k+r_k$$

Given the initial conditions are $y_1 = 3$ and $x_{1}^+ = y_1$. How can I find other parameters?

Since the initial condition $x_1^{+} = 3$, I can compute $$x_2^{-} = F_1x_1^{+}+G_1U_1 = I[3]+0 $$. I'm having a trouble finding $P_0^{+}$.

  • 1
    $\begingroup$ Numerous statements about the Kalman filter in this question don't make sense. $\begin{bmatrix}1 & 1 & 1 & 1\end{bmatrix}^T x$, for instance, is an invalid multiplication because $x$ is a 4-element column matrix. Also, the statement "Also, assume the uncertainty in both model, $F$, and observation, $A$, are both equal to $\sigma^2$" makes no sense -- the state transition matrix ($F$) and output matrix ($A$) in your nomenclature are assumed to be known, not uncertain. I could go on, but until the question makes sense, no answer can be formulated. $\endgroup$
    – TimWescott
    Commented Nov 22, 2020 at 4:19
  • $\begingroup$ I suggest you hit the books again, or change the title of this question to "Why is this problem formulation not a valid Kalman filter formulation?" $\endgroup$
    – TimWescott
    Commented Nov 22, 2020 at 4:19
  • $\begingroup$ @TimWescott Why do you think $x$ is a column vector? The problem in this way can be used to find $\hat{x}$ using the recursive Least square method. $\endgroup$
    – ccc
    Commented Nov 22, 2020 at 4:30
  • $\begingroup$ Because in every Kalman filter formulation I've ever seen, $x$ is a column vector. Since you got other things considerably wrong, I have no reason to believe that you got that right. However, if you're working from a text where it isn't a column vector, I suggest you state what it is. And, of course, correct the other problems with your question. $\endgroup$
    – TimWescott
    Commented Nov 22, 2020 at 4:38


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.