Abstract: The acoustic source localization is a research subject with wide range of applications. To improve the localization accuracy, an acoustic source localization algorithm based on compressed sensing is proposed. By building redundant dictionary, the algorithm sets the location of unknown multiple-source nodes as a coefficient vector, and then uses sparse Bayesian compressed sensing algorithm to estimate acoustic source location. In order to accelerate the speed of algorithm, an efficient construction method for multi-resolution dictionary is presented. The localization accuracy is improved through the reduced localization space by iteration. Theoretical analysis and experimental results conclude that the acoustic source localization algorithm based on compressed sensing improves position ability of multi-source nodes and reduces the amount of sensor nodes needed. In addition, compared with the subspace algorithm, the proposed algorithm has superior performance.

Key words: acoustic source localization, compressed sensing, Bayes, sensor arrays

摘要： 声源定位是一个应用非常广泛的研究课题。针对阵列定位精度不高的问题，提出一种基于压缩感知的声源定位算法。通过构建冗余字典，该算法将网络中的多个未知源节点的位置作为一个系数向量，然后采用稀疏贝叶斯学习算法估计声源位置。为了增快算法的运行速度，提出一种有效的多分辨率字典构建方法，并迭代地减小定位空间，提高定位精度。实验结果显示，基于压缩感知的声源定位算法可以改善多源节点的定位能力，且有效地减少所需的传感器节点。此外，与基于子空间的算法比较显示，该算法的性能更优越。

关键词: 声源定位, 压缩感知, 贝叶斯, 传感器阵列