A novel particle swarm optimization for high-level synthesis of digital filters

Abstract

This paper presents a novel discrete particle swarm optimization (PSO) technique for the high-level synthesis of digital filter data-paths. In this technique, the cost associated with the final digital filter data-path is minimized for obtaining combined area-cum-time optimal digital filter data-paths subject to user-specified constraints on the number of the required arithmetic functional units. In the proposed technique, the digital filter data-path encoding is achieved by combining the information regarding the operation scheduling together with the information regarding the allocation and binding of operations to arithmetic functional units into a single particle. The scheduling, and allocation and binding information form the coordinate values of the particles in PSO. The salient feature of the resulting PSO technique is its fast convergence speed, achieved by ensuring that the (random) movement of the particles in the search space in the course of optimization are automatically guaranteed to preserve the data-dependency relationships in the original digital filter signal flow-graph without any recourse to backtracking. The usefulness of the proposed PSO technique is demonstrated through the application of it to the high-level synthesis of a benchmark elliptic wave digital filter. It is observed that the application of the PSO leads to substantially faster convergence speeds as compared to the corresponding genetic algorithms.