一种基于时频分析的欠定盲源分离算法

针对欠定盲源分离,提出了基于时频域信源数估计的盲源分离算法。通过短时傅里叶变换将信号变换至时频域进行分析。在混合矩阵估计环节,根据源信号稀疏性的假设,在时频域利用信号的时间频率比,确定与源信号对应的混合向量。在混合向量聚类过程中,利用同类向量间的距离以及向量不同聚类间的解析度,提出一种信源数估计方法,解决了在聚类算法中假设信源数已知的问题。在源信号分离环节,利用时频子空间中接收信号相关矩阵对角线元素与剩余元素的比值,提出一种子空间信源数估计算法,通过确定信源数,将子空间中欠定模型转换为超定或正定模型,以完成源信号分离。仿真证明算法对混合矩阵估计的正确性以及信号分离的可靠性。

An Underdetermined Blind Source Separation Algorithm Based on Time-Frequency Analysis

Aiming at underdetermined blind source separation (UBSS), an algorithm was proposed based on source estima.tion in time-frequency domain. First, the receipt signals in time-frequency domain was analyzed by using short time fouri.er transform (STFT), and according as the supposing of source sparseness, the corresponding vectors were ensured by usingthe time and frequency ratio of receipt signals. In the process of mixed vector clustering, a source estimation algorithm wasproposed based on the resolution between the different vector cluster and the distance between same vector cluster, thenthe problem that supposed the source number had been known before was resolved. In the process of source separation, asource estimation algorithm in time-frequency subspace was proposed based on the ratio between diagonal elements andothers of the receipt signal correlation matrix, in time-frequency subspace, the underdetermined model converted to over.determined or well-posed model, and complete the signal separation. At last, the simulation proved the reliability and cor.rectness of algorithm.