Efficient Modeling of Linear Discrete Filters Using Ant Lion Optimizer

Abstract

In this work, a novel application of ant lion optimizer (ALO) has been presented for adaptive identification of infinite impulse response (IIR) filters. ALO is a nature-inspired, population-based, gradient-free meta-heuristic algorithm. It works based on the interaction between antlions and ants and uses Roulette wheel for selection of fitter antlions for catching ants. During the iterative process of optimization, performance of best antlion in each iteration is compared with elite antlion which ensured an optimum solution. To demonstrate the filter modeling efficacy of ALO, four IIR benchmark systems of different orders have been considered for equal and reduced ordered identifications. Modeling performance has been assessed using mean square error (MSE) between the actual and identified model performances, mean square deviation (MSD) between actual and identified IIR filter coefficients and rate of convergence. ALO outperformed the other two recent meta-heuristic algorithms, i.e., cuckoo search algorithm (CSA) and gravitational search algorithm (GSA), recently used for filter modeling. To further assess the robustness of obtained solutions, fifty independent identification trials were conducted and MSE and MSD values were analyzed statistically for standard deviations and means in addition to the statistical t test on MSE values. ALO offered the least standard deviations indicating the robust solutions. Further, in t test, higher positive t values again indicated the significant superiority of ALO over CSA and GSA for IIR filter modeling.