Adaptive control of sound transmission with neural network algorithms

Abstract

The Artificial Neural Network is widely used for its self-learning, self-organizing adaptability as well as nonlinearity in nature. Active control of structural sound radiation and sound transmission through panels is widely researched recently for its effectiveness and efficiency in the low-frequency range. In this paper, a neural network-based feedforward adaptive controller using the modified Error Back Propagation Learning Algorithm is presented. The controller is realized with a TMS320C25 DSP Board monitored by an IBM PC compatible, and applied to control the sound transmission through a thin panel between two rooms. Experiments showed that the algorithm was superior in robustness and broadband performance. In the adaptive controller, the structure of a traditional multi-layered feedforward neural network (MFNN) is modified. Both filtered-X and filtered-U adaptive filters are discussed. Two auxiliary filters are designed to compensate the secondary signal feedback and the error delay. Random gradient estimation method is used to update the weights of MFNN. Several methods to speed the learning rate are also introduced. The causality of the feedforward controller is analyzed, by which the system behavior is greatly improved. Further strategy to improve the controller is also proposed. [Project supported by the Natural Science Fund of China.]