How is this composite Gaussian Distribution derivated?

111 Views Asked by At

enter image description here

This is the prediction step of Kalman filter. I'm not sure how (18.25) derives into (18.26).. Does anyone have ideas about this?

1

There are 1 best solutions below

0
On

(Much) more simply, (18.25) is assuming that, conditionally on $z'$, the random vector $z$ is normal with mean $Az'+Bu$ and covariance matrix $Q$, and that the random vector $z'$ is normal with mean $\mu$ and covariance matrix $R$.

Thus there exists $x$ and $x'$ independent centered normal such that $$z=Az'+Bu+Lx$$ with $LL^t=Q$ and $$z'=\mu+Tx'$$ with $TT^t=R$. One sees that $$z=(ATx'+Lx)+(A\mu+Bu)$$ is normal with mean $A\mu+Bu$ and covariance matrix $$E((ATx'+Lx)(ATx'+Lx)^t)=ATT^tA^t+LL^t=ARA^t+Q$$ as desired. Elementary linear algebra all along...