A general approach towards blind multiuser detection using higher order statistics
Malay Gupta, Balu Santhanam · 2006
Nongaussianity maximization based independent component analysis (ICA) algorithms are becoming increasing popular in signal processing and communications. These algorithms exploit higher order statistics (HOS) of the data either by an explicit computation or implicitly by nonlinearly transforming the observations. Learning algorithms based on explicit HOS are simple to analyze, but exhibit slow convergence. Introduction of suitable nonlinearities in the algorithm increases the convergence rate and robustness to outliers. In this paper, we propose a general nonlinearity based algorithm for blind adaptive multiuser detection in multipath channels. Implementation of the proposed HOS based detector is done via an extension of a recently proposed constrained less-complete ICA framework. Computer simulations illustrate the improved performance over the previously proposed kurtosis based detector