基于自适应滤波算法的车载ANC仿真

Automotive ANC Simulation Based on Adaptive Filter Algorithm

本文利用自适应滤波器的原理对车载主动降噪系统ANC中的语音降噪功能进行一次模拟仿真。仿真的语音输入通过安卓车机系统和芯科科技的DSP芯片Si47925进行输入,以尽可能还原实车语音场景。本次设计的车载ANC系统以自适应滤波器为核心,以最小均方算法LMS为主进行自适应控制,对自适应滤波常用的四种算法:最小均方算法LMS、基于最小四阶矩阵的组合算法LMS/F、归一化最小均方算法NLMS和变步长最小均方算法VSS-LMS进行迭代推导,并基于Matlab软件,实现了这四种算法的代码实现,并对带噪语音进行数字滤波仿真。经过仿真发现,四种算法均能对带噪音频产生滤波降噪作用,其中LMS的权系数迭代式最简单和方便的,定步长LMS/F算法和LMS算法下的系统稳定性和降噪性相差不多,VSS-LMS算法和NLMS算法均可提高收敛速度,其中NLMS算法能实现较高的信噪比,并且稳态精度更加高。

In this paper, we use the principle of adaptive filter to perform a simulation of the voice noise re-duction function in the ANC of the automotive noise reduction system. The simulated speech input is performed through the Android car system and the DSP chip Si47925 from CoreTech to restore the real car speech scenario as much as possible. This designed automotive ANC system is based on the adaptive filter as the center and the least mean square algorithm LMS as the main adaptive control. Four algorithms commonly used for adaptive filtering: least mean square algorithm LMS, combined algorithm LMS/F based on the minimum fourth order matrix, normalized least mean square algorithm NLMS and variable step size least mean square algorithm VSS-LMS, will be derived, realize the code implementation of these four algorithms based on Matlab software, and digital fil-tering simulation is performed for noisy speech. According to the result of simulation, it is found that all four algorithms can realize noise reduction effects on noisy band frequencies, among which the LMS has the simplest and most convenient iterative weight coefficients, the stability and noise reduction of the system under the fixed-step LMS/F algorithm and the LMS algorithm are similar, the VSS-LMS algorithm and the NLMS algorithm can both improve the convergence speed, among which the NLMS algorithm can achieve a higher snr and the steady-state accuracy is much higher.