One- and two-dimensional systolic arrays for least-squares problems

In this paper homogeneous systolic solutions to overdetermined systems of linear equations are described. The least-squares problem (LSP) can be solved on two-dimensional mesh-connected or hexagonal arrays. Simple partitioning schemes can be applied to solve large-scale problems. A general systematic method is given to derive linear processor arrays from given two-dimensional ones. This hierarchical approach leads to one-dimensional systolic arrays for the LSP with advantageous properties. The data occur in a lexicographical input/ output scheme and pipelined arithmetic units can be used.