On adaptive Kalman filtering for the Luenberger canonical form

This work reports on simulation results obtained from an adaptive filter for systems in Luenberger canonical form. In particular the effects of errors in estimation of innovations, autocorrelations and initial conditions are examined. The scheme proves to be more efficient than a similar one based on measurement autocorrelations. For a certain class of systems new formulae are given for calculating the structural indices required by the algorithm from the state and measurement dimensions. Extension is made to the more general problem of coloured measurement noise.