experts on Estimation, on Kalman, I'm considering a temporal linear KF estimation problem, the whole system is as follows: $$Y_{2n*1}(t) = [Y1;Y2]$$ $$Y1 = Ax +\epsilon_{1}$$ $$Y2 = Ax +Bb+\epsilon_{2}$$ where the size of measurements $Y1$ and $Y2$ is the same $(n*1)$, the corresponding measurement noises $\epsilon_{1}$ and $\epsilon_{2}$, independent from each other, are both Gaussian White noise, $$\epsilon_1 \sim N(0, Q_1)$$ $$\epsilon_2 \sim N(0, Q_2)$$ and $Q_2$ is 100 times samller than $Q_1$,
$A$, of size $(n*m), m<n$, is the geometry matrix related to state vector $x$, of size $(m*1)$,
$B$, of size $(n*n)$, is the geometry matrix related to state vector $b$, of size $(n*1)$,
Is there any tuning rules, for example on the state process noise covariance $Q_x, Q_b$, that I should follow, to avoid the case that the state $x$ is biased-estimated.
However, according to the state nature, $b$ is a un-known constant vector to be estimated. The resonable process noise uncertainty $Q_b$ is of order,i.e. $1e-4$.
The biased estimation already happened, it seems that too much confidence has been put on the estimates $\hat b$.
As the measurement covariance matrix is unchangeable, either are the geometry matrices, so I'm wondering maybe I should start from the tuning of states process noises uncertainty.
Could any expert on estimation, KF tuning, help me, please?