Matthieu Bloch
Tuesday October 04, 2022
Let \(\dotp{\vecx_i}{\vecx_i}=\Pi_i\). Then \[ \forall i\geq 0\qquad \Pi_{i+1} = \matF_i\Pi_i\matF_i^T + \matG_i\matQ_i\matG_i^T \] Let \(\Phi(i,j)\eqdef \prod_{\ell=i-1}^j\matF_\ell\) for \(i>j\) and \(\Phi(i,i)=\matI\). Then \[ \dotp{\vecx_i}{\vecx_j} = \begin{cases}\Phi(i,j)\Pi_j\text{ for }i\geq j\\\Pi_i\Phi(j,i)^T\text{ for }i\leq j\end{cases} \] \[ \dotp{\vecy_i}{\vecy_j} = \begin{cases}\matH_i\Phi(i,j+1)\matN_j\text{ for }i> j\text{ with }\matN_i=\matF_i\Pi_i\matH_i^T+\matG_i\matS_i\\ \matR_i+\matH_i\Pi_i\matH_i^T\text{ for }i= j\\ \matN_i^T\Phi(j,i+1)^T\matH)j^T\text{ for }i< j\end{cases} \]
Objective: come up with a computationally efficient way of performing estimation
The innovation \(\vece_i\) is \[ \vece_i = \vecy_i - \matH_i\hat{\vecx}_{i|i-1} \]
The one step predition \(\hat{\vecx}_{i|i-1}\) can be computed as \[ \hat{\vecx}_{i+1|1} = \matF_i\hat{\vecx}_{i|i-1}+\matK_{p,i}\vece_i\quad\textsf{where } \matK_{p,i} = \dotp{\vecx_{i+1}}{\vece_i}\norm{\vece_i}^{-2} \]
Set \(\hat{\vecx}_i\eqdef\hat{\vecx}_{i|i-1}\), \(\tilde{\vecx}_i=\vecx_i-\hat{\vecx}_{i}\) and \(\matP_i \eqdef \dotp{\tilde{\vecx}_{i}}{\tilde{\vecx}_{i}}\) \[ \norm{\vece_i}^{2} = \matH_i\matP_i\matH_i^T+\matR_i \] \[ \matK_{p,i} = (\matF_i\matP_i\matH_i^T+\matG_i\matS_i)\norm{\vece_i}^{-2} \]
Recursion for \(\matP_i\): \[ \matP_{i+1} = \matF_i\matP_i\matF_i^T + \matG_i\matQ_i\matG_i^T - \matK_{p,i}\norm{\vece_i}^{2}\matK_{p,i}^T \]