Research On UBSS Based On The Characteristics Criterion Of Signals

The technology of Blind Source Separation (BSS) has been widely used in signalprocessing, data mining, pattern recognition, channel estimation etc. In the past tenyears, the condition that the number of signal receiving antenna or the sensor is lessthan the number of sources of aliasing has been the hot spot in the field of BSSresearch.In the past, the researchers who studied in the field of BSS had mainly focused onthe well-determined and over-determined situation, using independent distributioncharacteristics of signals, a series of independent component analysis (ICA), principalcomponent analysis (PCA) as the theoretical foundation of the algorithms have beenstudied, and those algorithms have been effectively used to solve the two conditionsblind source separation problems. For the problem proposed for blind sourceseparation, the past algorithms would not be helpful; researchers propose a set of newtheories to solve such problems, including the analysis of high order statistics, sparsecomponent analysis (SCA) and non-negative matrix factorization (NMF).Based on the study of underdetermined blind source separation (UBSS) theory, Ihave studied various SCA and NMF algorithms. First, The signal’s characteristic ofsparseness is a necessary condition in BSS. This paper studied the sparsity of theimages in wavelet domain, and estimated the mix-matrix through the SCA clusteralgorithm. Second, based on the research of hierarchical NMF algorithm (HALS), thedeterminant criterion which can guarantee the uniqueness criterion matrixdecomposition is introduced into this algorithm. The improved HALS algorithm areapplied to the separation matrix sparsity constraint and determinant criterion constraint,and finally applied to UBSS. Through the simulation, in the underdetermined case, thealgorithm can be useful for both the sparse signals and normal sparse.