Inner matrix dimensions must agree. error

Asked by Khoa on 4 Dec 2012 at 7:13
Latest activity Answered by Khoa about 11 hours ago

Hi everybody. I'm very news in matlab. I try to design kalman filter but the error occurred 'Inner matrix dimensions must agree' in matlab function

function [ x, P] = kalman( z, Q, R, x_old, P_old, A, H)

I = eye(3);

% Measurement update
K = (P_old * H') / (H*P_old*H'+R);
x = x_old + K * (z - H*x_old);
P = (I-K*H)*P_old;

% Time update
x = A * x ;
P = A * P * A' + B*Q*B';

Maybe initial estimates of P_old and x_old is not set I think. How to set that value? I want to set the first value for x and P are [0 ; 0 ;0 ] and [0.1 0 0; 0 0.1 0; 0 0 0.1] but dont know how to do.

5 Comments

Show 2 older comments
Walter Roberson on 4 Dec 2012 at 7:36

What size are P_old and H ?

Question: are you sure you want algebraic matrix division? Matrix inverse of one of the parts, matrix multiplied by the other part? Or are you wanting element-by-element division which uses the ./ operator instead of the / operator?

Khoa on 4 Dec 2012 at 7:49

P_old(3x3) and H(1x3). I upload my simulink below. Can u check it for me? http://www.mediafire.com/view/?7qla8km2q33jbs9 Thank you.

A = [1 5e-5 12.5e-10 ; 0 1 5e-5 ; 0 0 1];
H = [1 0 0];
Q = [0.1 0 0 ; 0 0.1 0 ; 0 0 0.1];
R = 0.1^2;
I = [1 0 0 ; 0 1 0 ; 0 0 1];
Walter Roberson on 4 Dec 2012 at 8:45

Please go in with the debugger, with dbstop if error, run until the problem occurs, and then try the subexpressions to see which part is generating the error

(P_old * H')
(H*P_old*H'+R)

and then if neither of those, finally

(P_old * H') / (H*P_old*H'+R)

With the array sizes you report, I do not reproduce the error.

I have to wonder about possibilities such as on the first iteration, something might be empty or an unexpected size.

Khoa

Tags

Products

2 Answers

Answer by Azzi Abdelmalek on 4 Dec 2012 at 8:37
Edited by Azzi Abdelmalek on 4 Dec 2012 at 8:41
Accepted answer

As you said in your question, in your delay block you must set initials conditions to x0 and P0, where x0=zeros(n,1); n is your system order, and P0=0.1*eye(n). in your case n=3

2 Comments

Khoa on 6 Dec 2012 at 0:44

Thank you for your answer. But it still not work when I set initials conditions to x0 and P0 as you said. I post my simulink below: http://www.mediafire.com/?9wcwl6h1cwl6jx2 http://www.mediafire.com/?ycebwm7zly1tq88 I think the size of matrix x is not correct but not sure.

Azzi Abdelmalek on 6 Dec 2012 at 4:05

What is B in your code?, it should be H.

Corrected function

function [ x, P] = kalman( z, Q, R, x_old, P_old, A, H)
I = eye(3);
% Measurement update
K = P_old * H'/(H*P_old*H'+R);
x = x_old + K * (z - H*x_old);
P = (I-K*H)*P_old;
% Time update
x = A * x;
P = A * P * A' + H*Q*H';

Also, In the constant block H uncheck Interpret Vector as 1-D

Azzi Abdelmalek
Answer by Khoa about 11 hours ago
Edited by Khoa about 11 hours ago

Sorry for reply late. I followed you and it worked correctly. But the output of kalman filter is not correct. My input of kalman filter is measurement position and I want to output position, velocity and acceleration. You can see it from scope1 and scope2(position input and velocity). I dont know how to fix kalman output. In other hand, My kalman block show bad result for output velocity and acceleration (scope5,6). I post my simulink below: http://www.mediafire.com/?l4mfgcupgmge4ij http://www.mediafire.com/?laa26ib5bgsco88

0 Comments

Khoa

Contact us at files@mathworks.com