Identification by optimal smoothing using integrated random walks

In using optimal smoothing algorithms to identify time-varying linear systems, one must specify the covariance of the sequence generating the time-variation, usually by experiment, applying knowledge of what behaviour is likely. Alternatively, such knowledge may determine the model of the time-variation, reducing the need for experiment. An identification algorithm representing the time variation as integrated random walks is derived, and its performance compared with that of the original random-walk algorithm.