A computationally efficient fixed-lag smoother using recent finite measurements

This paper proposes a computationally efficient fixed-lag smoother with finite memory structure for discrete-time systems. The proposed fixed-lag smoother is developed under a weighted least square criterion using only the most recent finite measurements on the window. The proposed fixed-lag smoother is shown to have inherent properties such as unbiasedness, deadbeat, time-invariance, and robustness. The proposed fixed-lag smoother is shown to be much less computationally complex than the existing fixed-lag Kalman smoother with infinite memory structure. Since the window length can significantly affect the estimation performance as well as the on-line computational complexity, a guideline for choosing the appropriate window length is introduced. Using computer simulations, the proposed fixed-lag smoother is shown to be comparable to the fixed-lag Kalman smoother for the nominal system and superior to that for the temporarily uncertain system.