Blind Separation Based On Ica Theory Algorithms And Technology Research

Blind Source Separation (BSS) technology is a class method that it can separate signals only through the observations after the source signals mixed that theirs parameters are unknown. In reality, if we are not have any additional restrictive conditions, It is impossible to estimate the source signals. The reality is that we come into contact with the signal in many cases are in line with or similar to the kind of blind-separation model,so its potential applications are very popular,and the Independent Component Analysis (ICA) is the main method to the Blind Source Separation technology. For example, many technologies in the speech recognition system, communication signals and biomedical signal processing can using Blind Source Separation method. The methods of BSS and ICA are belong to a class of unsupervised learning algorithm, its algorithm theory and practical application are related the optimization of a number of areas of mathematics and neuroscience. The goal of ICA is to restore specified independent source signals (independent source) based on the sensor observations, and these observations are unable to directly access to the linear mixture signals. Unlike transform contrast-based interactions method, such as principal component analysis (PCA), ICA is not only second-order decorrelation), but also it can be extended to higher order statistics independent. And so that, it can obtain as much as possible the independence of the mixed signals. In other words, ICA is a methom that it can find a linear non-orthogonal coordinate system with multivariate datas, and the axis of the coordinate system depends on the original data in the second or higher order statistics, its purpose is to perform a linear transformation that makes the resulting variables as much as possible to statistically independent of each other.The first work of this paper is that it has done a research on on the adaptive variable Stepsize Gradient Algorithms for blind source separation.And then it has put forward a kind of Nonholonomic Variable step size natural gradient algorithm,and then applied to the mixed speech signals that compared with the traditional algorithm. on the other hand, Via the Empirical Mode Decomposition(EMD) method,we have analysised the problem of blind source separation based on Single Input Multiple Output (SIMO), A EMD-SIMO algorithm has proposed to the problem of single channel blind source separation.and it has simulated in the sinusoidal signal plus noise.