RECURRENT RADIAL BASIS FUNCTION NETWORKS FOR ADAPTIVE NOISE CANCELLATION

Radical basis function neural network architectures are introduced for the nonlinear adaptive noise cancellation problem, Both FIR and IIR filter designs are considered, and it is shown that by exploiting the duality with system identification, the nonlinear IIR filter can be configured as a recurrent radial basis function network. Details of network training that is based on a combined k-means clustering and Givens routine, the inclusion of linear dynamic network links, and metrics for performance monitoring are also discussed. Examples are included to demonstrate the degree of noise suppression that can be achieved based on the new design.