Structured stochastic optimization strategies for problems with ill-conditioned error surfaces

The paper compares the performance of several structured optimization strategies in adaptive signal processing problems that are characterized by ill-conditioned error surfaces. The genetic algorithm (GA), the particle swarm optimization (PSO) algorithm, and a new constrained random search (CRS) algorithm (Siddharth, P., 2004) are considered. When applied to adaptive filters, these structured stochastic search strategies are independent of the adaptive filter structure and are capable of converging to the global solution when applied in circumstances that create multi-modal mean square error surfaces.