Noise Control of a Moving Evaluation Point Using Neural Networks

Abstract

This paper describes the noise control using neural networks during moving the evaluation point in a one-dimensional sound field. In the noise control, the Filtered-x LMS algorithm is widely used. However, this algorithm becomes uncontrollable when the secondary path characteristics from the control sound source to the evaluation point change by moving the evaluation point. To follow changes of the secondary path characteristics, we applied neural networks with the learning ability to the noise control system. So far, we have constructed the noise control system using neural networks and have shown the effectiveness of the control by numerical simulation. However, it has been reported that the control stability depends on the initial value of the internal parameter. In this paper, we consider that the cause is the neural network structure and propose a method of improving the control stability by changing the neural network structure related to its parameter. The effectiveness is investigated by comparing the control stability and performance of the proposed method with that of the conventional method in numerical simulation. The results show that the proposed method improves the control stability by 20% or more and has higher control performance than the conventional method.