Study On Adaptive Multiple User Detection For CDMA Communication Systems

In the past two decades, be considered as one of the key techniques for the 3rd Generation (3G) mobile communication system, Multiple User Detection (MUD) for CDMA system made its obvious progress. Although a plenty of MUD algorithms have been proposed by many researchers, MUD for CDMA system can be essentially regarded as one kind of optimization problem in which the tradeoff between the computational complexity and detection performance should be elaborately balanced. Among all kinds of multiple user detectors for CDMA system, adaptive MUD is the promising one with acceptable implementation complexity, while simultaneously eliminating the need of a plenty of priori system information including the training sequence, thus can be implemented as a 'blind' MUD.In this thesis, the author demonstrates his close study on the topic of adaptive signal processing and MUD algorithms in detail. The main contribution of this thesis can be summarized as follows:1. Most MUD algorithms are developed with the assumption of perfect user signal synchronization and accurate knowledge on the user signal power. In fact, this assumption indicates that all wireless propagation channel parameters, such as multiple path time delay, amplitude and phase offset, should be estimated prior to the multiple user detecting. The author proposed a fast adaptive CDMA channel estimation algorithm both for the single channel and multiple channels cases. Performance analysis and simulation results show that our algorithm has faster convergence ability than the existing algorithms. The algorithm is then used in the novel MUD proposed in the thesis.Most existing adaptive filtering algorithms are only fit for real-valued processing. Complex-valued adaptive signal processing can provide more robustness and faster convergence ability than its real-valued counterpart. The author developed a Complex-valued Euclid Direction Set based2. adaptive learning algorithm useful for complex-valued signal processing as well as design of high performance MUD. The stability of the proposed algorithm is also studied.3. A novel two-stage complex-valued adaptive MUD algorithm featuring strong anti-MAI (Multiple Access Interference) capability is suggested. Comparative performance simulation with respect to other kinds of MUD shows that our algorithm outperforms existing adaptive MMSE MUDs.4. A Complex-valued Functional Link Neural Network (CFLNN) with its training algorithm is proposed. In the experiment related to a complex-valued nonlinear system identification task, it is shown that the CFLNN algorithm has equal or better performance as compared to the conventional MLP (Multiple Layer Perceptron). A new adaptive nonlinear MUD based on CFLNN was proposed, simulation results indicate that it has better detection performance than the complex-valued MLP based MUD.5. A new complex-valued adaptive blind MUD based on the well-known blind MOE criteria is developed, the algorithm has better performance than that of a blind MUD using conventional adaptive algorithm since a lower searching dimension is involved. Finally, a rigid body dynamics training algorithm is introduced to construct a blind MME adaptive MUD, which led to improved performance as compared to the conventional blind MMSE MUD...