A new VLSI architecture for rank order and stack filters

One-dimensional rank order filters are non-linear filters which choose an output based on its rank within a onedimensional window of sample inputs determined by sorting the inputs. Several extensions to rank order filtering include recursive rank order filtering, two-stage rank order filtering, stack filtering, and two-dimensional rank order filtering. Two classes of VLSI architectures are commonly used for rank order filtering. The first class stores the inputs within the sample window in a shift register and employs sorting networks to completely sort the window of inputs when a new sample arrives. The second class called running order sorters utilizes the overlapping windows between consecutive outputs to efficiently maintain a sorted list of the inputs within the window. Although the class two architectures are computationally more efficient than the class one architectures, they are difficult both to pipeline and to apply to the rank order filter extensions. A new VLSI architecture for rank order filtering is presented in this paper. This architecture is an addition to the class two rank order filter architectures and can be pipelined and applied to the extensions to rank order filtering.