Improved Algorithm And Simulation Research For Independent Component Analysis

Independent component analysis (ICA) is a new method of data processing and analysis, which is regarded as a kind of expansion of a principal component analysis (PCA). ICA is a signal processing field research hotspot and its prior knowledge is rare, and non-Gaussian source signal as the research object. Only from the mix signal to recover or extract source signals which is statistically independent and non-Gaussian. That could be better separation out the independent of source signals which hidden in mixed signal. Because of its excellent separation performance, it has important application in the speech signal, data mining, financial data analysis, image processing, telecommunications and so on, and its theory and algorithms by the international community's attention. In recent years, independent component analysis has been considerable development, but there are still many problems needed to be further improved, including how to improve the performance of the algorithm, and the better application in the practical. This article emphatically introduces independent component analysis algorithm which is popular now days:FastICA and natural gradient method and puts forward the two new algorithms, which is verified through simulation experiment results. This main work is as follows:1. Introduces negative entropy-based fast independent component analysis algorithm (FastICA). Give a fifth-order convergence of Newton iteration, and on this basis is given an independent component analysis of the improved algorithm, the final image separation and source separation using two sets of simulation experiments to compare the improved algorithm with FastICA algorithm, verify the effectiveness of the algorithm.2. Expounds natural gradient algorithm of independent component analysis. Introduced a kind of symmetry and unsymmetrical generalized Gaussian model, and using the model to improve scoring function. The simulation using the signal separation experiments to verify the effectiveness of the improved algorithm by comparing the algorithm with the natural gradient method.