Spatially constrained parallel factor analysis for semi-blind beamforming

Parallel factor analysis (PARAFAC) has found numerous applications in blind signal processing, mainly due to its nice identifiability. However, the standard PARAFAC decomposition does not use prior information on the mixing procedure, which could actually be roughly estimated. As a result, the standard PARAFAC is ambiguious in permutation, and may converge slowly in the presence of collinearity. In this paper, we assume that the steering vector of the target source is coarsely known, and propose a spatially constrained PARAFAC algorithm by using this prior information. In addition, a semi-blind beamformer with multiple-invariance array is presented based on the above spatially constrained PARAFAC. Simulations are provided to verify the efficacy of the proposed method.