基于STFD和PARAFAC的欠定盲源分离信源个数估计算法

为更精确地进行欠定混合条件下的信源个数估计，提出了一种基于空间时频分布（STFP）与平行因子分析（PARAFAC）的欠定盲源分离信源个数估计算法.该算法从空间时频分布矩阵中选择符合单源时频支撑域的时频点构成1个3阶张量，然后由核连续诊断算法计算因子数量，估计出信源个数，并对3阶张量平行因子低秩分解的惟一性条件进行了分析.该算法不需要假设源信号必须满足稀疏性和独立性条件，也不要求其满足高斯分布条件.在计算机仿真实验中，当信噪比为-5 dB时，识别正确率提高18 dB，证明了改进算法的有效性. 

Source Number Estimation of Underdetermined Blind Mixing Signals Based on Spatial Time-Frequency Distribution and Parallel Factor Analysis

To estimate the number of sources of underdetermined blind mixing signals, a novel algorithm based on spatial time-frequency distribution (STFD) and parallel factor analysis (PARAFAC) was proposed. The time-frequency distribution matrices corresponding to the single auto-terms time-frequency (TF) points were stacked in a three-order tensor, and the core consistency diagnostic (CORCONDIA) was performed to estimate the number of sources. Then the uniqueness of the three order tensor low rank decomposition was analyzed. This algorithm does not need to assume that source signal must satisfy the sparse and independence conditions, and does not require that the signal meets Gaussian distribution. In simulation the recognition accuracy rate is increased by 18 dB when the signal-to-noise ratio is -5 dB, which demonstrates the proposed recognition algorithm is effective.