ICA assisted blind multiuser detection in DS-CDMA systems.

摘要

Independent Component Analysis (ICA) is a statistical technique that involves the optimization of a contrast function to extract independent components from their observed mixtures. ICA techniques do not assume any prior knowledge of the underlying independent components or their mixing configuration. ICA algorithms exploit higher order statistics (HOS) of the observations, either by a direct computation or implicitly, by incorporating a suitable nonlinearity in the algorithm.; ICA algorithms incorporating a nonlinearity are attractive because they are robust to outliers in the data and are computationally simpler. However, the convergence of such algorithms can be proved only in the specific case of fourth-order nonlinearity. A part of this dissertation focuses on the issues related to the selection of this nonlinearity in negentropy maximization based ICA algorithms. We present a class of nonlinearities with which the unimodal character of the contrast function is preserved.; Next, the application of ICA to the problem of blind multiuser detection (MUD) in synchronous DS-CDMA systems is considered. Existing MUD approaches are based on second order statistics (SOS) of the data. ICA based MUD techniques exploit both SOS and HOS of the data and hence are expected to provide improved performance over the existing multiuser detectors. It is however, well known that ICA solutions are obtained with arbitrary scaling and permutation ambiguities. We propose to solve these ambiguities on the basis of a priori knowledge of the user signature codes. Various strategies to incorporate this information into the ICA framework are proposed. Performance comparisons with the existing techniques illustrate the usefulness of the proposed algorithms.; Finally, the application of ICA based MUD techniques is extended to asynchronous dispersive CDMA channels. A constrained ICA framework is utilized to incorporate prior information about the desired user's signature code and timing into the optimization of the contrast function under multiple constraints. A LMS based adaptive algorithm for the joint estimation of channel parameters and user data is proposed. The proposed algorithm is compared with a recent kurtosis based approach and improvements in convergence and achievable signal to interference plus noise ratio (SINR) are demonstrated via the computer simulations.