Application of the SRIF to systems with time delay

Square-root information falter formulations of the discrete-time Kalman filter are sometimes thought to be restricted by a requirement that the state transition matrix be nonsingular. In a recent publication [1], Boncelet, Jr. and Dickinson have shown that this restriction is not necessary. For problems where both formulations can be applied, the generalized formulation is, however, more computation intensive than the direct formulation described by Bierman [2]. Also, it is shown here that transition singularities associated with time delays are removable and can be efficiently handled in the SRIF.