An Evolutionary Computation Embedded IIR LMS Algorithm

An improved Infinite Impulse Response (IIR) Least Mean Squares (LMS) algorithm using parallel filters and evolutionary programming techniques is introduced. IIR filters have the attractive property that they require fewer computations than a corresponding FIR filter, but they are prone to instability and local minimum problems. Evolutionary algorithms are good in global optimization scenarios, but are computationally very expensive. Adaptive filter weights for a given step size and initial weight vectors may not lead to optimal solutions. In this paper we extend the IIR LMS algorithm by embedding an evolutionary computation, and also simultaneously implement multiple filters (different initial weight vectors) to achieve optimal solutions.