Abstract
Three pipelined multiprocessor implementations of adaptive lattice filters are examined. The three multiprocessor architectures, which can be characterized as a serial pipeline, a ladder-connected dual pipeline, and a ring pipeline, are derived directly from the computational and data-transfer requirements of adaptive lattice algorithms. The order-recursive nature of the adaptive lattice structure results in architectures which use pipelining extensively. A performance analysis is done for each multiprocessor system, with respect to two different adaptive lattice algorithms. Expressions for approximate computation time and speedup are derived for each combination of architecture and algorithm.<>