Selective Fusion of Delayed Measurements in Filtering

Delayed measurements can create difficulties for discrete time filtering. A number of procedures have been proposed to handle such delayed observations. A common characteristic of these methods is the policy of always incorporating, or fusing, delayed measurements at the time they are finally available. In this paper, we consider such methods, and develop thresholding techniques that result in delayed measurements being selectively fused. This selection is driven by an assessment of the utility of incorporating delayed measurements. With simulated and real data, we show that selective fusion can reduce computational costs while maintaining near optimal performance.