CONVERGENCE BEHAVIOR OF PARTITIONED EVOLUTIONARY DIGITAL FILTERS

This paper proposes partitioned evolutionary digital filters (EDFs) in order to implement the EDF on parallel proces-sors. The execution time of the original EDF is longer than that of the LMS adaptive digital filter, since the EDF con-sists of many inner digital filters. Thus, we divide the orig-inal EDF into smaller sub-EDFs for parallel implementa-tions. In the partitioned EDF, a large population of the orig-inal EDF is divided into smaller subpopulations. Then, each sub-EDF has one subpopulation and executes the small-sized main loop of the original EDF. Numerical examples show that the partitioned EDF gives the same learning curves as the original EDF does in spite of dividing the original pop-ulation. 1