Fast convolution using modularized multiprocessor architectures
P. Alexander · 2005
This paper describes a recently-developed modular array processing system (MARS), capable of implementing fast convolutional algorithms for MHz rate sample streams. After a brief mathematical description of the problem, the system computational requirements are analyzed for a real data stream clocked at 2 MHz. For this example, a discussion is presented covering hardware modules, I/O overheads, and segmentation of the computational burden.