Skip to Main Content
Smoothers or nonlinear filters that make estimates purely on the basis of rank and temporal order determinations among members of the input data are inherently robust and compatible with VLSI implementations. The classic example is the moving median smoother, but it is possible to define others which have significantly different attributes both in relation to the statistical estimation problem and in relation to hardware implications. Attributes offered in the latter category, which are especially desirable in systolic VLSI, include essentially recursive operation and elimination of the interactions between rank and temporal ordering requirements which are a feature of median smoothers. Experimental results indicate significantly improved response to noisy steps for these smoothers, in comparison to median smoothers of comparable rectangular pulse detection capabilities.