Noise Control for a Moving Evaluation Point by Using Neural Networks Modifying Their Structures and Learning Rates

Abstract

In noise control, the Filtered-x LMS algorithm is generally used as an adaptive algorithm. This algorithm can achieve stable noise reduction with a simple control system. However, it has a disadvantage that the control cannot be performed when the secondary path characteristics change with the movement of an evaluation point. In this paper, we use a noise control system using neural networks with the learning ability to follow the change of the secondary path characteristics when an evaluation point moves. Conventional studies of noise control using neural networks have problems with their control performance and control success rate. To solve the problems, here we propose a novel structure of neural networks and a method to bias the learning rates according to the type of input signals for noise control system. The effectiveness of each method was investigated by numerical simulations. The results show that the proposed structure using the method to bias the learning rates has the control success rate of more than 90%, which is 30 points better than a conventional structure and that the noise reduction performance is approximately 16 dB, which is approximately 4 dB better than a conventional structure.

Keywords

• noise control
• adaptive control
• moving evaluation point
• neural network
• neural network structure
• learnig rate