Some probabilistic and empirical aspects of chebyshev estimation and identification

Because the actual shape of a model´s response may be more important than the value of an index such as the integral squared error, a probabilistic and empirical study has been made of parameter and state estimates that minimize the maximum error between plant and model outputs. Preliminary results indicate that minimax estimation compares favorably with estimations based upon a minimum squared error criterion. It is further shown that if the measurement noise is characterized by a distribution function which is continuous, symmetric and differentiable, then the estimate of the measurement will be unbiased.