Filtered-X CMAC: an efficient algorithm for active disturbance cancellation in nonlinear dynamical system

An algorithm for the convergent adaptation of a CMAC neural network in feedforward disturbance cancellation architectures is presented. This technique is a generalization of the Filtered-X LMS algorithm used in the case of linear adaptive filters. The fundamental advantage provided by the CMAC compensator is its effectiveness in systems with nonlinearities in the actuators, sensors, and signal transmission paths. The presented method also provides advantages over other nonlinear solutions due to its rapid convergence and minimal computational overhead. Two variants of the algorithm are considered. The full version uses a CMAC model of the secondary path while the reduced version approximates the secondary path with a linear FIR filler. Results are presented for an implementation of the algorithm on a laboratory acoustic duct model.