Optimal filtering of discrete-time hybrid systems (Q1281967)

The author considers the hybrid system \[ x_{k+1}=F_{k}(\theta _{k})x_{k}+u_{k},\quad y_{k}=H_{k}(\theta _{k})x_{k}+v_{k},\quad k=0,1,\dots, \] where \(x_{k}\in \mathbb R ^{n_{1}}\) is a signal, \(y_{k}\in \mathbb R^{n_{2}} \) is an observation, \(\theta _{k}\) is a Markov chain with states \( i=1,\dots,m\), \(F_{k}(i)\), \(H_{k}(i)\) are matrices of appropriate dimensions, \( u_{k}\) and \(v_{k}\) are system and observation disturbances which are non-Gaussian. To construct a recursive filter \(\widehat{x}_{k}\), \(\widehat{\theta}_{k} \) the author studies an associated linear-quadratic optimal control problem. He also considers a generalization to nonlinear hybrid systems and gives numerical examples.