基于改进MVDR的麦克风阵列声源定位方法

Microphone Array Sound Source Localization Method Based on Improved MVDR

摘要: 针对传统最小方差无畸变响应(MVDR)算法在低信噪比环境下易受协方差矩阵估计误差影响，导致定位精度下降、主峰不稳定及旁瓣和伪峰增强的问题，提出一种基于对角加载与动态权值优化的麦克风阵列声源定位方法(DL-DWO-MVDR)。该方法通过对协方差矩阵进行对角加载，提高矩阵求逆稳定性；同时依据帧能量构建动态权值，使有效信息较强的帧在多帧空间谱融合中获得更高权重，降低低质量帧干扰。通过单声源和双声源仿真，对比DSB、传统MVDR与所提算法在不同信噪比和声源方位下的定位性能。结果表明，所提算法主峰更集中稳定，旁瓣和伪峰更低，且具有更好的双声源分辨能力。实物实验中平均定位误差小于2˚，单帧定位时间低于1 s，验证了方法的可行性。

Abstract: To address the degradation in localization accuracy, unstable main peaks, and increased sidelobes and spurious peaks of the conventional minimum variance distortionless response (MVDR) algorithm under low signal-to-noise ratio conditions, a microphone-array sound source localization method based on diagonal loading and dynamic weight optimization, termed DL-DWO-MVDR, is proposed. Diagonal loading is first applied to the covariance matrix to improve the numerical stability of matrix inversion. Dynamic weights are then constructed according to frame energy, allowing frames containing stronger effective information to contribute more to multi-frame spatial-spectrum fusion while reducing interference from low-quality frames. Single-source and dual-source simulations are conducted to compare the proposed method with delay-and-sum beamforming (DSB) and conventional MVDR under different signal-to-noise ratios and source directions. The results show that the proposed algorithm produces more concentrated and stable main peaks, lower sidelobes and spurious peaks, and better dual-source resolution. Physical experiments further demonstrate the feasibility of the method, with an average localization error of less than 2˚ and a single-frame localization time of less than 1 s.