基于互谱矩阵函数的多声源识别方法

Multi-source Identification Method Based on Cross-Spectral Matrix Function

为解决多声源识别中的混叠问题,提出一种基于互谱矩阵函数的多声源识别方法。首先,对与各峰值源相关的互谱矩阵进行特征值分解,并利用特征值和特征向量构造其函数;其次,在反卷积迭代过程中,将该函数用于计算移除各峰值源后的互谱矩阵函数和输出功率;最后,将与各峰值源相关的干净波束和剩余功率谱叠加,获得多声源的分布图像。仿真与实验结果表明,该算法的输出波束主瓣窄、旁瓣低,可有效提升传统反卷积算法的动态范围,实现多声源的高分辨率识别。

In order to solve the aliasing problem in multi-source identification,a high-resolution deconvolution algorithm based on cross-spectral matrix function is proposed.Firstly,the algorithm decomposes the cross-spectral matrix associated with each peak source into eigenvalues and constructs a matrix function.Secondly,the matrix function is used to update the cross-spectral matrix function and output power in the process of deconvolution iteration.Finally,the distribution images of multiple sound sources are obtained by superimposing each clean beam and residual power spectrum.Simulations and experiments show that the main lobe of the output beam of the algorithm is narrow and the side lobe is low.This algorithm can not only effectively improve the dynamic range of the traditional deconvolution algorithm,but also achieve high-resolution identification of multiple sound sources.