欠定条件下的盲分离算法

盲信号分离中当源信号个数大于观测信号个数，且源信号不是足够稀疏时，如果利用聚类算法进行分离，分离效果将会变差。为此提出一种在此欠定条件下新的盲信号分离算法。利用源信号的“稀疏性”估计混合矩阵，然后简化混合矩阵构造新的混合模型。由于源信号间具有的独立性，使得可以在新的混合模型中从观察信号的自相关函数中估计出源信号的频谱，从而达到分离出源信号的目的，且分离效果优于聚类算法。最后给出仿真试验实例，试验结果验证了算法的有效性。

Blind Source Separation in Underdetermined Mixtures

A new algorithm for the independent component analysis under the condition of more sources than sensors is presented. Under the condition,the clustering algorithm cannot obtain a better result if source signals are not sparse enough. The new algorithm can simplify the channel matrix and set up a new one by using the sparse characteristic of source signals. Based on the independence among source signals, the frequency spectrum of the source signals can be estimated from auto-correlation function of the observed signals in a new channel model,and then draw source signals. The algorithm can get better result than the clustering algorithm. Simulation experiments are available to support the proposed algorithm.