Research On Adaptive Blind Source Separation Algorithm Based On The Natural Gradient

Blind signal separation is a new technology for array processing and data analysis in the signal-processing field. Blind source separation aims to recover the source signal from observation signal according to the statistic characteristics of source signals without knowing prior information of the source signal and transmission channel. After twenty years of development, BBS has become an important issue in some areas and takes a key role in speech recognition, medical signal processing, remote sensing, radar and communication systems and seismic exploration.This paper firstly elaborates the basic theory of blind source separation in detail, including blind source separation model, the objective function for blind source separation, signal pretreatment, and thoroughly analyzed some common blind source separation algorithms, such as natural gradient method, fixed-point algorithm, EASI algorithm and negentropy maximum algorithm.Natural gradient algorithm in blind source separation is the focus of this paper. In the natural gradient algorithm, the convergence speed and stability error are two important indicators, but these two indicators are of inherent contradictions. In order to solve the shortcoming of natural gradient algorithm, this paper modified the nonlinear activation function in traditional algorithm and proposed an adaptive step factor based on exponential function and separation degree. The activate function and adaptive step were used respectively to natural gradient method and incomplete natural gradient method. Computer simulations show that the modified algorithm has faster convergence speed, smaller stability error and more desirable separation effect than traditional algorithm.