On minimizing the divergence in discrete filters

The problem of divergence that often arises when the filter mechanization is based on lower order process dynamics is examined. This situation may be caused, for example, by the impossibility of completely identifying some of the process parameters. It is suggested that instead of totally neglecting the affected state variables a minimal sensitivity estimate of these be obtained on the basis of the nominal values of the imperfectly known parameters. The result of applying this scheme to a system, part of whose state vector is thus affected, is also examined.