Efficient stack filter implementations of rank order filters

This paper presents efficient stack filter implementations of rank order filters for bit-parallel, bit-serial, and nibble-serial data. The implementations support variations in word size, window size and rank order. All the implementations are sampled at a high rate by efficient incorporation of pipelining. While the bit-parallel realizations are pipelined by adding latches in the feed-forward paths, the recursive bit-serial and nibble-serial realizations are pipelined to p levels by increasing the number of PBF units by a factor of 2^p-1. The computational complexity of the bit-parallel realizations is reduced by processing a block of outputs, thereby exploiting the computational overlap between consecutive windows.