Convolutive Separation of I.I.D. Signals Based on Simultaneous Tensors Diagonalization

Abstract

This paper considers the problem of blind separation of a MIMO convolutive mixture of i.i.d. source signals. Separation criteria are considered for the overall extraction of source signals according to the use of so-called reference signals. We present a new MIMO contrast function using reference signals, which is moreover seen to have joint-diagonalization interpretation. A link with the PARAFAC decomposition is also emphasized. Finally the performance are investigated through computer simulations