Extended Kalman Filter Jacobians

In summary, the author is asking how to form the A and C matrices for the state space model. The author seems confused on how to form these matrices. The formulae given in the source the author is learning from apparently relate to these matrices.
  • #1
Master1022
611
117
Homework Statement
Calculate the Jacobean matrices for the extended Kalman filter
Relevant Equations
Partial derivatives
Hi,

I have a question about calculating the Jacobian matrices for the Extended Kalman filter.

Question: If we have a system of the form:
[tex] \begin{align*} x_{k+1} =f_k (x_k , u_k) + w_k \\
y_k = h_k (x_k , u_k ) +v_k \end{align*} [/tex]
where the state [itex] x_k [/itex] comprises of the three variables [itex] p_1 [/itex], [itex] p_2 [/itex], and [itex] p_3 [/itex]. The input [itex] u_k [/itex] comprises of the variables [itex] q_1 [/itex] and [itex] q_2 [/itex]. Form an extended Kalman filter for this system.

Method:
From what I understand, I need to linearize the system to form the A and C matrices for the state space model. I am, however, confused on how to form these Jacobian matrices. The source I am learning from presents the following formulae:
[tex] A_k = \frac{\partial f_k}{\partial x_k} |_{\hat x_{k|k} , u_k} [/tex]
and
[tex] C_k = \frac{\partial h_k}{\partial x_k} |_{\hat x_{k|k} , u_k} [/tex]

I don't understand how I ought to interpret these formulae. Should my matrices have the following form, or have I misunderstood something?
[tex] A_k =
\begin{pmatrix}
\frac{\partial f_1}{\partial p_1} & \frac{\partial f_1}{\partial p_2} & \frac{\partial f_1}{\partial p_3} \\
\frac{\partial f_2}{\partial p_1} & \frac{\partial f_2}{\partial p_2} & \frac{\partial f_2}{\partial p_3} \\
\frac{\partial f_3}{\partial p_1} & \frac{\partial f_3}{\partial p_2} & \frac{\partial f_3}{\partial p_3}
\end{pmatrix} [/tex]
where [itex] f_i [/itex] refers to the function in the [itex] i^{th}[/itex] row of the vector [itex] f [/itex] (which has three rows) and
[tex] C_k =
\begin{pmatrix}
\frac{\partial h_1}{\partial p_1} & \frac{\partial h_1}{\partial p_2} & \frac{\partial h_1}{\partial p_3} \\
\frac{\partial h_2}{\partial p_1} & \frac{\partial h_2}{\partial p_2} & \frac{\partial h_2}{\partial p_3} \\
\end{pmatrix} [/tex]
where [itex] h_i [/itex] refers to the function in the [itex] i^{th}[/itex] row of the vector [itex] h [/itex] (which has two rows)

Thank you in advance for any help and guidance.
 
Last edited:
Physics news on Phys.org
  • #2
spoto2 said:
Previously, we used a Kalman Filter which could only model linear ... It can be very difficult to implement an EKF correctly: you will most likely get the Jacobian.
Thanks for your reply. Do my expressions above seem correct though?
 

FAQ: Extended Kalman Filter Jacobians

What is the purpose of Extended Kalman Filter Jacobians?

The Extended Kalman Filter Jacobians are used in the Extended Kalman Filter (EKF) algorithm, which is a type of recursive Bayesian filter used for state estimation in nonlinear systems. The Jacobians are used to linearize the nonlinear system dynamics and measurement models, making it possible to apply the Kalman filter equations for state estimation.

How are the Jacobians calculated in the Extended Kalman Filter?

The Jacobians are calculated using partial derivatives of the nonlinear system dynamics and measurement models with respect to the state variables. These partial derivatives are then evaluated at the current state estimate, resulting in a linearized model that can be used in the Kalman filter equations.

What is the difference between the Extended Kalman Filter Jacobians and the regular Kalman filter matrices?

The regular Kalman filter matrices (i.e. state transition matrix and measurement matrix) are constant and do not change throughout the filtering process. On the other hand, the Jacobians in the Extended Kalman Filter are calculated at each time step based on the current state estimate, allowing for the estimation of nonlinear systems.

Can the Extended Kalman Filter Jacobians handle highly nonlinear systems?

The Extended Kalman Filter Jacobians are designed to handle moderately nonlinear systems, but may struggle with highly nonlinear systems. In these cases, the Extended Kalman Filter may diverge or produce inaccurate estimates. In such cases, other nonlinear filtering techniques such as the Unscented Kalman Filter may be more suitable.

Are there any limitations to using Extended Kalman Filter Jacobians?

One limitation of using Extended Kalman Filter Jacobians is that they are based on linearization, which may introduce errors in the estimation process. Additionally, the Extended Kalman Filter may struggle with systems that have large uncertainties or non-Gaussian noise. It is important to carefully consider the system and measurement models before using the Extended Kalman Filter and to validate the results through simulation or experimentation.

Similar threads

Replies
1
Views
971
Replies
12
Views
2K
Replies
2
Views
2K
Replies
1
Views
2K
Replies
1
Views
2K
Replies
1
Views
2K
Replies
0
Views
2K
Replies
17
Views
10K
Back
Top