Performance of Kalman filter with missing measurements (Q1072516)

The purpose of this paper is to study the performance of a discrete-time Kalman filter when a non-zero probability exists that some of the measurements will not be available, i.e. missing, with the probability of occurrence of such cases being available to the estimator a priori. For the situation described, the time history of the error covariance matrix will be different for each possible measurement sequence. A useful measure for the filters performance is the expected value of the error covariance, which is found to be not self-propagating. However, equations for upper and lower bounds for the error covariance expected value, which are self-propagating, are developed.