Researches And Realizations On Methods Of Reducing Blind Source Separation Based On The Temporal Structures

According to some statistical properties of source signal, the recovering process from observed signals to source signals is called blind signal separation under the unknown circumstances of source signal and transmission channel. It has been a hot research subject in the area of signal processing recently due to its high adaptability. Hence, it has a wide range of applications in medical signal processing, communication systems, array signal processing and speech signal separation.The signal separation with temporal structure is important content in blind signal separation, and it is also a problem in the field of signal processing. Aiming at temporal structure signal, this paper has researched the blind separation algorithm under the linear instantaneous mixing model. The main work stated as follows.1. It has introduced the hybrid model and separation principle of blind signal separation, the theory like separability, assumption and pretreatment of blind source separation, the analysis methods including principal component analysis and independent component analysis, and three kinds of algorithms based on temporal structure, namely, Algorithm for Multiple Unknown Signals Extraction(AMUSE), Second-Order Blind Identification(SOBI) and Jointly Approximate Diagonalisation of Eigenmarices (JADE). Finally, it has presented the separation evaluation criteria, including similarity coefficient and performance index.2. Blind source separation algorithm is realized by using inherent differences between source signals shown in some sides, in which the inherent differences are their time structural characteristics. The time structural characteristic of source signal is not changed in mixed signal after linear instantaneous mixtures. This paper has attempt to use the change degree of characteristic parameters of signal as a measurement of temporal structure, and discussed deeply the method which used the change degree of second-order characteristic parameters as measurement to constitute separation matrix, translate the blind signal separation into generalized eigenvalue to solve problems and achieve signal separation. At last, the effectiveness of the algorithm has tested by emulation.3. It has adopt the autoregressive (AR) to describe the source signal with temporal structure, and set the related sum of multistep delay time of source signal as discrimination and presented a new blind extraction algorithm based on linear forecast. Simulation results show that the new algorithm is valid, efficient, and excellent separation effect.