The Blind Source Separation Algorithms Based On Over-sampling And Time-frequency Analysis

With the rapid development of communication technology, blind source separation (Blind Signal Process, BSP) technology has become one of the hot topics by the researchers in the field of signal processing and neural network circles. This technology originated in the 1990s, whose process is that just use the information of observation signals to restore the source signals in the case of unknown source signal and transmission system. Blind source separation technology has been used widely in many fields.It is assumed that the signal is stationary in the problem of Classical blind source separation, but in practical many signals processed are non-stationary. Especially in communications, biomedical, radar field, most of the statistical properties of the signal is cyclostationary with a periodic. Therefore, the blind source separation is necessary and significant in the perspective of cyclostationarity.This paper studies blind source separation based on cyclostationary theory, which greatly includes the basic theoretical knowledge of blind source separation, its main algorithm, cyclostationarity theory, time-frequency analysis, etc.The major contribution of this paper includes:1. It summarizes the cyclo stationary theory and analyzes its application in the field of blind signal separation. Specifically it introduces the concept and principle of blind source separation, analyses the mathematical models of the instantaneous mixed blind source separation and convolutional mixed blind source separation, simply classifies blind source separation algorithm, and especially analyzes blind source separation algorithm based on the information theory.2. It analyzes the cyclostationary theory systematically, specifically introduces the concept of cyclostationarity, cyclic statistics, over-sampling model and the properties of signals.3. On the aspect of blind source separation, it focuses on blind source separation problems based on Cyclostationarity, and proposed a new iterative algorithm for blind source separation. The algorithm utilizes the cyclostationarity of source signals after oversampling, simplifies the mixed signals and introduces Cyclic Frequency into the separable matrix update equation. Through the combined treatment, it further improves the algorithm's convergence speed and convergence accuracy. Compared with traditional Infomax algorithm, the results show that the algorithm has rapid convergence speed, and better separation effect.4. It introduces the characteristics of cyclostationary signals to the time-frequency domain, better showing the characteristics that the signal frequency changes by time. Specifically it describes the theory of the second time frequency distribution and researches time-frequency analysis of cyclostationary signals in detail. It combines the joint matrix approximate diagonalization algorithm, and proposes a new blind source separation algorithm based on cyclostationary signal in Time-frequency distribution.