How to count and guess well: Discrete adaptive filters (Q1330925)

The problem of discrete estimation of the states of an unobserved process is considered. The process is not observed directly. On the basis of discrete observation process the best estimates are achieved. The model that is considered includes a hidden Markov model (HMM). A homogeneous, finite state, discrete-time Markov chain is used. The Markov chain is observed through a finite state function that is subject to random perturbations. This situation is modelled by hidden Markov model. A general filter is obtained. It provides recursive updates of estimates of processes related to Markov chain given the observations. In the normalized form particularly simple equations are achieved. The obtained result provides recursive updates of estimates and smoothers for the state of the process, for the number of jumps from one state to another, for the occupation time in any state and for a process related to the observations. Such a way ``self tuning'' to the data is achieved, based on re-estimation of the model. The authors state two main contributions of the paper: the introduction of an equivalent measure under which the observation values are independent and identically distributed, and the use of independent property when the state space of the Markov chain is identified with canonical unit vectors in a Euclidean space. The proposed model is of wide applicability and the achieved results introduce an adaptive discrete time filtering procedure that is ``self tuning'' to the observations.