【摘要】 针对空间扩展声源的稀疏波达方向(DOA)估计问题，提出一种全变分规则化的二维DOA估计方法(简记为2DTV-CES)。首先利用扩展声源的空间分布成组性和声源相关性特征建立扩展声源模型；然后，通过构建二维广义阵列流形及其过完备表述，实现声源的稀疏表示；通过定义二维全变分，构建二维全变分正则项，实现对于声源结构特征的几何约束，促进解的分段常数轮廓的形成；最后结合全变分正则项与一般LASSO构建二维全变分稀疏DOA估计模型，由凸优化求解。理论分析表明，与传统二维DOA估计方法相比，所提方法避免了去相关处理、角度配对的步骤。仿真实验验证了方法的有效性，在涉及扩展源的DOA估计中，2DTV-CES方法性能明显优于一般LASSO方法和改进的ESPRIT方法，检测概率超过95%,实现了对扩展声源的二维波达方向的高精度快速估计。更多还原

【Abstract】 To solve the problem of sparse direction of arrival(DOA) estimation for spatially extended sound sources, a two-dimensional DOA estimation method based on total variational regularization was proposed, which was abbreviated as 2DTV-CES.Firstly, an extended sound source model was established by using the spatial distribution group of the extended sound source and the correlation characteristics of the sound source. The sparse representation of the sound source was realized by constructing the two-dimensional generalized array manifold and its over complete representation. Then, by defining the two-dimensional total variational and constructing the two-dimensional total variational regular term, the geometric constraints on the structural characteristics of the sound source were realized and the piecewise constant contour was formed.Finally, by combining total variational regularization and general least absolute shrinkage and selection operator(LASSO), a two-dimensional sparse DOA estimation model was established, which was solved by convex optimization.The theoretical analysis shows that compared with the traditional 2D DOA estimation method, the proposed method avoids the steps of de-correlation processing and angle matching. The simulation results verified the effectiveness of the proposed method. In the DOA estimation involving extended sources, the performance of the 2DTV-CES method is obviously better than that of the general LASSO method and the improved ESPRIT method. With the detection probability of over 95%, the proposed method can realize the fast and highly precise estimation of the 2D arrival direction of the extended sources.更多还原