BLIND ESTIMATION OF A CLASS OF UNDER-DETERMINED CONVOLUTIVE MIMOSYSTEMS USING PARAFAC DECOMPOSITION OF OUTPUT TENSORS

摘要

We consider the problem of frequency domain identification ofrna convolutive multiple-input multiple-output (MIMO) system drivenrnby white, mutually independent unobservable inputs. In particular,rnwe improve upon a method recently proposed in [1] that usesrnPARAFAC decomposition of a tensor that is formed based on thirdorderrnstatistics of the system output. The approach of [1] utilizesrnonly one slice of the output tensor to recover one row of the systemrnresponse matrix. We here propose an approach that fully exploits therninformation in the output tensor. As a result, the proposed methodrnnot only achieves lower error values but also becomes applicable torna class of MIMO systems with more inputs than outputs. We alsornshow how by using higher order statistics, I.e., fourth- or sixth-orderrnstatistics, or pairs of third or Fourth-order tensors, one can furtherrnexpand the class of under-determined systems that are identifiable.