Online Gradient Ica With Individualized Learning Rates For Dynamic Environments With Communications Application

In time-varying environments, online learning algorithms are preferred over block adaptation methods. In this paper, a novel online gradient Independent Component Analysis (ICA) algorithm, named ICA with Individual Adaptation (IA-ICA), is proposed for dynamic situations. The algorithm is derived as a special case of the recently proposed Optimum Block Adaptive ICA (OBA/ICA) algorithm. Thanks to the individualization of the learning rates, IA-ICA outperforms the LMS-type online ICA algorithm in terms of both separation performance and convergence speed. The application of IA-ICA to interference suppression in wireless communications is also presented. Combined with diversity reception, the new technique represents an efficient digital interference rejection method. Simulations results confirm the effectiveness of the new technique. Also, it is shown experimentally that online ICA methods are applicable in the presence of thermal noise.