1
$\begingroup$

I'm starting to brush up on the Kalman Filtering I learned a couple decades ago. From what I remember, if you have a measurement vector

$$ z=H x + v $$ and the $n$ components of the measurement noise $v$ are uncorrelated with each other, then you can process the measurements sequentially in the KF equations. Is there a good reference somewhere (preferably online) that discusses why this is so? Intuitively it is understandable if each measurement $z$ depends on only one state element in $x$, but it isn't obvious if each measurement of $z$ depends on several elements of $x$.

$\endgroup$
2
  • 1
    $\begingroup$ I'm not sure what you mean by process the measurements sequentially in the KF equations ? Can you expand or write it out ? In words of one syllable or fewer... It's been a long day. :-) $\endgroup$ – Peter K. Mar 20 '13 at 0:05
  • $\begingroup$ See this answer: dsp.stackexchange.com/questions/3538/… $\endgroup$ – Deniz May 25 '13 at 20:35
1
$\begingroup$

I think it is evident in both cases. It follows directly from $$ P(A|B,C,D) = P(((A|B)|C)|D)$$ .i.e. $$P(x[k]\!=\!\hat{x}[k]|\!\hat{z}_1[k]\!=\!z_1[k],\!\hat{z}_2\![k]\!=\!z_2[k])\!=\! P\Bigg( \Big( x[k]\!=\!\hat{x}[k]|\!\hat{z}_1[k]\!=\!z_1[k] \Big)|\!\hat{z}_2[k]\!=\!z_2[k] \Bigg)$$ As long as you don't do a prediction update it should be fine. For nonlinear system, you won't get the same exact estimate, because you will end up with a new dynamics, input, measurement and feedback matrices on every "single-measurement" update

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.