Reduced Complexity Blind Estimation of Under-Determined Convolutive Mimo Systems

We consider identification of an under-determined convolutive multiple-input multiple-output (MIMO) system driven by white, mutually independent unobservable inputs. In our recent work, we showed that an N _i-input and N_o-output system can be estimated within trivial ambiguities based on PARAFAC decomposition of a tensor containing K-th order statistics of the system output, where Kgesmax{(2N_i-1)/(N_o-1), 3}. In this paper we show that by using a tensor pair we can guarantee identifiability while using statistics of order smaller than in the single tensor case. We also provide an iterative identification scheme. The proposed tensor-pair approach results in complexity reduction as it involves lower dimensionality tensors and lower order statistics