Skip to Main Content
A special purpose computer is described to evaluate the discrete convolution of two sequences of numbers. This computer abandons the traditional model of convolution as a series of inner products which, for input sequences of length n, requires n multipliers and (n − 1) adders to complete a convolution calculation in (2n − 1) time steps. Instead, it is shown that by reorganizing the algorithm, n interconnected processing units are able to evaluate a convolution in n time steps. Each processing unit consists of a multiplier, an adder, and the necessary buffers. In addition to providing increased throughput, the proposed organization results in a highly modular structure with a well defined interconnection pattern.