Blind Separation And Blind Multiuser Detection Technique Based On Independent Component Analysis

In recent years, blind source separation (BSS) has become a hot spot for researchers in many fields, such as artificial neural networks, statistical signal processing, information theory, and so on. As a kind of important method for BSS, independent component analysis (ICA) has been studied widely and deeply. Additionally, multi-user detection (MUD) technology, which is a key technology of the third generation mobile communication to combat multiple access interference (MAI), draws more and more attention from researchers. Blind adaptive MUD does not need the train sequence and spread sequence of other interfering users, and so becomes a research topic of MUD in the future.Based on the above background, this dissertation studies ICA and blind MUD, and presents a new algorithm for non-linear ICA and a new algorithm for MUD based on ICA and ICA with momentum term. In addition, the technology of blind MUD based on fraction lower order statistics (FLOS) in non-Gaussian channel noise is also studied. The whole dissertation consists of five chapters, and the second, third and fourth chapter is the main parts. The following is an abstract of every chapter. Chapter one is an introduction of the research background, present research situation, main contents and structure of the dissertation, the contribution and innovation of this dissertation is also pointed out.A new algorithm for non-linear ICA is presented in chapter two. Non-linear ICA is an extention of ICA and more fit for the practical environment, however, there is little mature algorithm for it because of the mathematical complex and nonseparability of many real systems just under the condition of statistical independence of the sources. In this chapter, a separation algorithm based on mutual information minimization for mono-nonlinear model is proposed, and its efficacy is proved by the contrast with the classical mutual information separation (MISEP) algorithm proposed by Almeida.In chapter three, a MUD algorithm based on ICA and ICA with momentum term is introduced. Multiple access interference is one of the important interferences in DS-CDMA system, which damages the performance of the system deeply. ICA , which is a method based on higher order statistics, can use the characteristics of mixed signals more efficiently than the traditional signal processing method based on second order statistics. ICA only needs the statistical independence of the source signals, and DS-CDMA system satisfies this point, so, combining ICA with the traditional signal detection technology can make the system more robust and efficient.Chapter four proposes a fractional lower order statistics based generalized constant modulus algorithm (FLOS-GCMA) to solve the problem of blind MUD in non-Gaussian channel noise. The comparison of FLOS-GCMA with the traditional constant modulus algorithm (CMA) and fractional lower order statistics based constant modulus algorithm (FLOS-CMA) in direct sequence code division multiple access (DS-CDMA) system shows that FLOS-GCMA has good performance both inα-stable distribution noise and Gaussian noise channel, which makes FLOS-GCMA has wide application.Chapter five is a conclusion of the whole dissertation and points out the future work to do.