<title>Parallel Processing Algorithms And Architectures For Real-Time Signal Processing</title>

Modified engagement method for matrix operation

“…1 and 10-20. In this paper we use the engagement array as a PE array with local memory and a systolic array without local memory. We report that, based on two-dimensional ring interconnection topology, matrix-matrix multiplication can be efficiently performed with the engagement array, which involves fewer steps than those reported by Speiser and Whitehouse, 11 Bocker et al, 10 and Kung. 14 Figure 1 shows the Speiser-Whitehouse scheme for two 3 3 3 matrix-matrix multiplications.…”

“…1 Although the systolic and the engagement methods have been widely investigated, there are no formal and coherent definitions for the terms systolic array and engagement array. Speiser and Whitehouse, 11 Bocker et al, 10 and Whitehouse et al 12 coined the term engagement array to denote one specific type of matrix operation in which input matrices A, B enter an array in a specific engagement manner, and the processing elements 1PE's2 have local memory 1accumulation2, i.e., the calculated data 3c 1t2 5 c 1t212 1 ab4 are not moved to other cells. Fisher's definition of a systolic array requires that only boundary cells 13 can be input@output ports for the systems, i.e., the calculated data must be moved to the boundary cells.…”

“…In order to give clearly a comprehensive comparison of the performances of the systolic array and the engagement array, we now consider the matrix-vector multiplication case. Four systolic@engagement matrix-vector multipliers are considered here, i.e., the original systolic array, the Priester et al improved systolic array, 21 the Speiser-Whitehouse engagement method, 11 and the improved engagement version.…”