Underdetermined Blind Source Separation Source Signal Recovery

At present, most of the UBSS problems are based on compressed sensing(CS), and a premise of using CS theory is to ensure that the source signal is sparse enough. While the signal is not sufficiently sparse, the signal reconstruction accuracy will have a great impact. A sparse dictionary can be used for increase the sparsity of the signal. Compared with the traditional fixed dictionary, the learning dictionary is used to improve the reconstruction accuracy. The learning dictionary can obtain more structured features through training on different signals, and adaptively express different signal features. For the difference in the sparse of representation of each frame of signals, a variable step size adaptive threshold SAMP algorithm (VT-SAMP) based on inner product distribution is proposed, which can solve the accuracy and computational pressure with multiplied candidate atoms increased in the iterative process of the SAMP algorithm. Changing the threshold value can improve the accuracy of atoms selection. This article mainly uses the K-SVD dictionary learning algorithm and the VT-SAMP algorithm to separate the mixed signals to obtain the source signals. Compared with other traditional algorithms, this proposed algorithm has a higher reconstructed accuracy. Finally, it is verified through experiments that this method is superior to traditional signal reconstruction algorithms and has better recovery performance.