Design and Performance of Adaptive Systems Based on Bio-Inspired Optimization Strategies

While the theory and design of linear adaptive filters based on FIR filter structures is well developed and widely applied in practice, the same situation is not true for linear IIR or for nonlinear adaptive filters in general. The latter situation exists because both linear IIR structures and nonlinear structures sometimes produce multi-modal error surfaces on which stochastic gradient optimization strategies fail to reach the global minimum. This seminar begins with a concise review of state-of-the-art techniques in linear adaptive filtering, and then develops the need for nonlinear adaptive filters in applications such as nonlinear echo cancellation, nonlinear channel equalization, and acoustic channel identification. Several basic nonlinear adaptive structures will be discussed, including Volterra models, neural network models, and series cascade modular structures. Then three important evolutionary optimization algorithms will be introduced as potentially useful algorithms to deal with multimodal error surfaces. The three evolutionary algorithms to be considered are the simulated annealing, genetic, and particle swarm optimization (PSO) algorithms. Particular emphasis will be placed on the PSO techniques because they have not received much previous attention for adaptive filtering.