Adaptive infinite impulse response (IIR) filters are more suitable in modelling real-world systems as they are inherently pole-zero structure and also require less parameters to achieve the same performance level of finite impulse response (FIR) filters. But they are not popular in practice due to lack of practical, efficient and robust global optimisation algorithms. This paper introduces a novel swarm intelligence (SI) based algorithm named as comprehensive learning particle swarm optimisation (CLPSO) to identify IIR systems. This variant of PSO enables the diversity of the swarm to be preserved to discourage premature convergence. Experiments were conducted to estimate the parameters of many benchmark IIR systems. The simulation results demonstrate improved estimation of these parameters by the CLPSO when compared with those obtained from standard PSO, the genetic algorithm (GA) and recursive least mean square(RLMS) algorithm. In addition, the new method offers significant improvement in convergence behaviour, lowest training time and least numbers of fitness evaluation during training of the model. The results of the new method also exhibit global search ability and absence of local minima phenomena in its reduced order models.
Keywords: infinite impulse response, adaptive IIR systems, IIR system identification, adaptive IIR filters, particle swarm optimisation, comprehensive learning PSO, applications, swarm intelligence, simulation, parameter estimation, convergence behaviour