IIR All-Pass Filter Design Based on Neural Learning Algorithm

The least-squares design of IIR all-pass filter can be formulated as an eigenproblem by solving an appropriate positive-definite matrix. Based on Rayleigh's principle, the eigenfilter design can be achieved by solving a single eigenvector corresponding to the smallest eigenvalue of an associated matrix. In this paper, the minor component analysis-based neural learning approach is exploited to the design of eigenfilter. As the learning algorithm achieves convergence, the weight vector of the neuron would approach to the smallest eigenvector which results the optimal filter coefficients of eigenfilter design. Simulation results indicate that the proposed neural learning approach can achieve good performance.