Author:

Keywords:

Science & Technology, Technology, Engineering, Electrical & Electronic, Engineering, Tensors, blind system identification, convolutive, cumulant, incomplete, coupling, CANONICAL POLYADIC DECOMPOSITION, JOINT DIAGONALIZATION, SEPARATION, UNIQUENESS, PARAFAC, ALGORITHMS, REDUCTION, STADIUS-17-128, C16/15/059#53326574, Networking & Telecommunications

Abstract:

© 1991-2012 IEEE. Blind system identification (BSI) is an important problem in signal processing, arising in applications such as wireless telecommunications, biomedical signal processing, and seismic signal processing. In the past decades, tensors have proven to be useful tools for these blind identification and separation problems. Most often, tensor-based methods based on fourth-order statistics are used, which have been studied extensively for independent component analysis and its convolutive extensions. However, these tensor-based methods have two main drawbacks: the accuracy is often limited by the estimation error of the statistics and the computation of these fourth-order statistics is time intensive. In this paper, we propose to counter these drawbacks for BSI by coupling the fourth-order statistics with second-order statistics and by using incomplete tensors. By doing so, we can obtain more accurate results or obtain results in a much faster way.