神经网络BP算法在有源消声中的应用

Neural Network BP Algorithm Applied to Adaptive Active Noise Control

为了克服传统自适应有源消声算法在应用中稳定性方面的不足，尝试将神经网络反向传播（ＢＰ）算法应用于有源消声技术。文中建立了基于ＢＰ算法的自适应有源消声（ＡＡＮＣ—ＢＰ）模型并给出该算法的递推公式。利用ＴＭＳ３２０Ｃ２５开发板实现了该算法功能。在半消声室中进行了单频和１００Ｈｚ带宽的消声实验，仅利用单个次级源结构便获得较好的消声效果。实验证明，基于ＢＰ算法的消声系统具有良好的稳定性。

The traditional adaptive active noise control (AANC) models that we know appear to be not quite satisfactory in robustness. We applied neural network BP (back-propagation) algorithm to improving its robustness. Fig.1 shows our AANC-BP sound attenuation model. We derived eqs.(1) through (8) for this model. In a semi-anechoic chamber, we obtained some sound reduction experimental data for a single frequency noise (Fig.3a) and for a 100 Hz bandwidth noise (Fig.3b), using our AANC-BP control system. Fig.3 was based on the use of only a single secondary source for suppressing noise. We observed an increase in robustness.