Blind identification of MIMO-FIR systems: A generalized linear prediction approach

We address the blind identification of multiple input multiple output (MIMO) polynomial transfer functions with the number of inputs strictly less than the number of outputs. According to previous results, the knowledge of the second-order statistics of the output series allows to reduce the convolutive mixture to an instantaneous mixture of the inputs under mild assumptions. Our approach consists of second-order deconvolution via the so-called linear prediction approach with the possibility of further instantaneous mixture separation. Contrary to the standard linear prediction, we propose a statistically optimal least-squares identification procedure. The latter reveals a simple sub-optimal approach characterized by high estimation accuracy as well as robustness to the model order choice. The statistical inference about the asymptotic performance of these methods is validated by computer simulations. (C) 1999 Elsevier Science B.V. All rights reserved.