Research On Blind Multiple User Detection Algorithms In DS-CDMA Communication Systems

Direct Sequence-Code Division Multiple Access (DS-CDMA) distributes mutually orthogonal (or as orthogonal as possible) code to different users for signal modulation in order to realize multiple users to access the system at the same time and frequency. But CDMA system is self-interfered system where Multi-Access Interference (MAI) and Near-far effect exist as main factors restricting system capacity. These years, considered as one of the key techniques for3G, Multiple User Detection (MUD) has made obvious progress. MUD reduces the MAI, solves the Near-far problem and improves system capacity. The blind MUD does not require training sequences and the interference user’s information. It only needs the prior knowledge of the signature waveform and the timing of the desired user. The blind MUD algorithms are the promising one with acceptable implementation complexity, while simultaneously eliminating the need of plenty of priori system information and can be implemented as a blind method.Firstly, DS-CDMA systems are introduced, theory of spread spectrum communication and characters of spread sequence are analyzed and essential reason of MAI is pointed out. Exiting MUD and their characters are analyzed,and meanwhile decorrelating and minimum mean square error (MMSE) algorithms are simulated separately.Secondly, this thesis has summarized the basis concept and theory on the independent component analysis (ICA), the kernel independent component analysis (KICA) and the blind MUD. Optimized algorithms and performance index of algorithms are summarized. On this basis, the simulations for the blind MUD algorithms based on FsatICA are studied in DS-CDMA.The applications for the blind MUD algorithms based on KICA are studied particularly in DS-CDMA. The earliest ICA algorithms are (in retrospect) based on contrast functions defined in terms of expectations of a single fixed nonlinear function. In this thesis, a new approach to the ICA problem is provided, which is based not on a single nonlinear function, but on an entire function space of candidate nonlinearities. The algorithms work with the functions in a reproducing kernel Hilbert space, and make use of the "kernel trick" to search over this space efficiently. KICA algorithms transform the singal into the high dimension space, initiating the KICA demix matrix with the desire user’s signature sequence, and iterating the demix matrix with adaptive methods. The simulation results show that, compared with the detector based on earliest ICA algorithms, the new detector provides better bit error rate performance in the same circumstances.Finally, the summary and the prospect on this research are given.