Filter approximations with constrained structure

Two techniques to constrain the structure of a linear transformation are presented. The objective is to generate an algorithm with improved computational speed. One type of constraint searches for the closest filter in a class known to be computationally efficient. The resulting representation is termed a triple matrix product. The second approach considers a given computing structure and generates a family of algorithms by modeling the partial transformations as a function of the compute cycle. The two approaches are combined to model the situation of a time-constrained triple matrix filter. The corresponding approximation problems are discussed