Abstract:
The modeling accuracy and control resulting in active noise control are affected by the nonlinear factors. The secondary path identification is optimized according to the active noise control (ANC) principle, to improve the accuracy and effect of noise control. The finite impulse response (FIR) model used to identify is replaced by the back propagation (BP) neural network, which performs better on the nonlinear factors. Based on least mean square (LMS) algorithm, the ANC algorithm under the secondary path model of neural network is deduced, and the iteration formula of coefficients is derived. The active noise control platform in a duct is built with the TMS320VC5509A as the core processor and the duct as the noise environment. The platform includes input, output and processing modules. The neural network model is trained with input and output signals as training samples. Signals of the secondary path are generated by the addictive white noise. The neural network improves the accuracy of the secondary path identification model as shown in the training results, indicating that the nonlinear factors of secondary path can be described by the neural network. The coefficients computed offline are loaded into the DSP and taken as the filtering parameters of the input signals. Under 500 Hz and 500+800 Hz noise source, the noise control experiment of FIR secondary path model and traditional ANC algorithm is compared with a neural network model and the optimized ANC algorithm. The results show that the algorithm is effective with good performance under the low-frequency noise of single and two-mixed frequency.