存在声反馈的前馈有源噪声控制系统性能分析

在一些应用场合,前馈有源噪声控制系统中次级源产生的声信号会反馈至参考传声器,影响参考信号质量和系统稳定,导致控制性能下降。引入了等效次级路径的概念,并通过等效次级路径与实际路径的相位偏差分析存在声反馈时的收敛性能。若某些频率的相位偏差大于90°,则这些频率附近将较难收敛,降噪性能下降,甚至导致系统不稳定。通过仿真和实验对单指向传声器声学方法、自适应滤波u型最小均方差(FuLMS)算法、反馈中和算法和在线建模算法共4种解决声反馈问题的方法的性能进行了比较。结果表明,4种方法都能提高存在声反馈时的前馈有源噪声控制系统的性能,有效解决声反馈引起的问题,但各有优缺点。单指向传声器方法最为方便,但低频指向性较差。FuLMS算法运算量较低,但不能保证收敛。反馈中和算法性能最好,但当系统时变时鲁棒性较差。在线建模算法不需要额外滤波器,但由于参数调节复杂,降噪性能稍差。

Performance analysis of feedforward active noise control systems with acoustic feedback

In some application scenarios of feedforward active noise control, the output of the secondary source is transmitted to the reference sensor. This contaminates the quality of reference signal and deteriorates the stability and performance of the active control system. Based on the concept of equivalent secondary path, this paper uses the phase deviation between the equivalent secondary path and the real secondary path to analyse the convergence behavior of the system with acoustic feedback. If the phase deviation exceeds 90 degrees at some frequencies, the control filter is difficult to converge around those frequencies. The noise reduction performance decreases, and the system may become unstable. Simulations and experiments are carried out to evaluate the performances of 4 approaches for solving the acoustic feedback induced problems, which includes the method of using a unidirectional microphone, the method of using the Filtered-u LMS (FuLMS) algorithm, the feedback neutralization method and the online modelling method. Both experiment and simulation results show that all 4 approaches can improve the noise reduction performance and are effective when there is acoustic feedback in the feedforward active control systems, but each approach has its own advantages and disadvantages. The unidirectional method is the most convenient method, but the directivity is poor in low frequency range. The FuLMS algorithm has low computational load; however the convergence is hard to be guaranteed. The feedback neutralization method has the best performance but has weak robustness when the system is time variant. The online modelling method doesn't require an extra filter, but its performance is slightly worse than other solutions due to complicated adjustment parameters.