Evolving IIR approximants for FIR digital filters

This paper describes a new technique for approximating an FIR digital filter by a reduced order IIR filter. The technique yields the globally optimal IIR filter, in the sense of minimizing the output error between the given FIR filter and the IIR approximant to a random white noise input. The model order of the IIR filter can either be specified or it is taken as the one that minimizes the Akaike information criterion. Simulated evolutionary optimization, a multi-agent stochastic search technique is used to optimize the coefficients. The technique permits stipulation of additional constraints on the filter to make it stable, minimum-phase and any other designer specifications. Also being inherently parallel, the technique requires significantly less computation time, compared to algorithms that optimize coefficients serially. Simulation results indicate that the proposed method performs better than classical LMS and system identification based methods like the Stieglitz-McBride method.