基于稀疏贝叶斯学习的压缩球波束形成声源识别方法  

Compressive Spherical Beamforming Based on Sparse Bayesian Learning for Sound Source Identification

作　　者：褚志刚[1,2] 赵洋 杨亮[1,2] 张晋源[3] 杨洋 CHU Zhigang;ZHAO Yang;YANG Liang;ZHANG Jinyuan;YANG Yang(College of Mechanical and Vehicle Engineering,Chongqing University,Chongqing 400044;State Key Laboratory of Vehicle NVH and Safety Technology,Chongqing 401133;Faculty of Vehicle Engineering,Chongqing Industry Polytechnic College,Chongqing 401120)

机构地区：[1]重庆大学机械与运载工程学院,重庆400044 [2]汽车振动噪声和安全技术国家重点实验室,重庆401133 [3]重庆工业职业技术学院车辆工程学院,重庆401120

出　　处：《机械工程学报》2023年第2期1-13,共13页Journal of Mechanical Engineering

基　　金：国家自然科学基金(11774040);汽车噪声振动和安全技术国家重点实验室2022年度开放基金(NVHSKL-202202)资助项目。

摘　　要：基于实心球传声器阵列的压缩球波束形成具有声学成像全景、适宜中远距离测量而易于布置等优势,在汽车、飞机等噪声源识别领域具有广阔应用前景。新近提出的基于稀疏贝叶斯学习(Sparse Bayesian learning, SBL)的压缩球波束形成能够获得良好的低频声源识别性能,但由于其采用了第二类最大似然估计(Maximum type-Ⅱ likelihood estimation, MLE-Ⅱ)进而需要估计声源稀疏度,且抗噪声干扰能力和计算效率也有待提升,推广应用受限。为此,首先将压缩球波束形成数学模型求解问题转化为SBL框架下的源强分布最大后验(Maximum a posterior,MAP)估计问题,并采用期望最大化优化算法(Expectation maximization, EM)加以求解,提出无需稀疏度估计的MAP-EM压缩球波束形成方法;在此基础上,将多快拍复声压矩阵输入转换为多快拍平均的声压互谱矩阵输入,并基于互谱矩阵对角重构降噪建立了抗噪声干扰能力增强的EMAP-EM(Enhanced MAP-EM, EMAP-EM)压缩球波束形成方法。仿真和试验均表明,提出的MAP-EM和EMAP-EM压缩球波束形成均具有高的空间分辨率和计算效率,且EMAP-EM压缩球波束形成抗噪声干扰能力更强,尤其在低频、低信噪比环境中声源识别性能更佳。最后,分析了迭代次数和快拍数对MAP-EM和EMAP-EM压缩球波束形成性能的影响规律并给出推荐值。Compressive spherical beamforming(CSB) based on solid spherical microphone array enjoys the advantages of panoramic sound imaging, suitable for medium and long distance measurement and easy to arrange, so it has broad application prospects in the field of noise source identification such as automobiles and airplanes. The recently proposed CSB based on Maximum type-II likelihood estimation(MLE-II) under sparse Bayesian learning(SBL) framework can achieve good low-frequency sound source identification performance, but it needs to estimate the sound source sparsity, and the anti-noise interference ability and computational efficiency also need to be improved. To solve the above problems, MAP-EM-based CSB(MAP-EM-CSB) without sound source sparsity estimation is proposed, which transforms the mathematical model solving problem of CSB into the maximum a posteriori(MAP) estimation problem of the source intensity distribution under SBL framework, and then uses the expectation maximization(EM) algorithm to solve it. Further, the sound pressure complex matrix input of MAP-EM-CSB is converted into the average sound pressure cross-spectrum matrix input of multiple snapshots, and then the enhanced MAP-EM-CSB(EMAP-EM-CSB) is established based on the diagonal reconstruction and noise reduction of the cross-spectrum matrix. Both simulations and experiment show that the proposed MAP-EM-CSB and EMAP-EM-CSB have high spatial resolution and computational efficiency. Due to its strong anti-noise interference capability, EMAP-EM-CSB has better sound source identification performance, especially in low frequency and low signal-to-noise ratio environments. Finally, the effects of the number of iterations and snapshots on the performance of MAP-EM-CSB and EMAP-EM-CSB are analyzed, and the recommended values are obtained.

关 键 词：声源识别 球传声器阵列 压缩球波束形成 稀疏贝叶斯学习 

分 类 号：TB52[理学—物理]