Optimal Module Distribution for Pipeline Digital Filter Analysis Algorithm

Abstract

The development of general techniques for the analyses of digital filters with arbitrary topology is an area of interest in the process of digital filter design, especially in the educational area. A few analyses with a large number of frequency points are required until the desired response is achieved. Therefore, it is necessary to provide fast analysis algorithms. The pipeline implementation of the Crout's algorithm enables parallelized execution of the digital filter analysis. The process distribution must be optimized in order to achieve faster analysis execution and balanced processor performance. This paper presents a program package with a generalized approach to optimize the process distribution, based on an algorithm for element combinations for a set of size L into all subsets of size m arranged in lexicographic order. Two optimization criteria are used: the number of processors and their utilization. To avoid variable execution times on different processors, the number of operations executed at each processor is taken as a measure of processor execution time. Obtained results indicate that optimal distribution is achieved using smaller number of well balanced processors. With a larger N processor efficiency rises to more than 95%.