Blind MIMO system identification using PARAFAC decomposition of an output HOS-based tensor

We present a novel frequency domain approach for the identification of a multiple-input multiple-output (MIMO) system driven by white, mutually independent inputs. The system frequency response is obtained up to constant scaling and permutation ambiguities based on parallel factorization (PARAFAC) of a three-way tensor, which is constructed based on a higher-order statistics (HOS) of the system output. An important future of the proposed approach is that, unlike existing methods, no pre-whitening of the system outputs is needed, which translates to less estimation errors.