DSP based RBF neural modeling and control for active noise cancellation

This paper presents active control of acoustic noise using radial basis function (RBF) networks and a digital signal processor (DSP) real-time implementation. The neural control system consists of two stages: first, identification (modeling) of the secondary path of active noise control using RBF networks and its learning algorithm, and secondly neural control of the primary path based on the neural model obtained in the first stage. A tapped delay time is introduced in front of the neural controller, and another tapped delay line is inserted between controller neural networks and model neural networks. An algorithm referred to as FX-RBF is proposed to account for secondary path effects of the control system arising in active noise control. The resulting algorithm turns out to be the filtered-X version of the standard RBF learning algorithm. We address centralized and decentralized controller configurations and their DSP implementation is carried out. The effectiveness of the neural controller is demonstrated by applying the algorithm to active noise control within a 3-D enclosure to generate quiet zones around error microphones. Results of real-time experiments show 10-30 dB noise attenuation, better than those obtained by classical least mean-square techniques such as FX-LMS.