Tensor-Based Blind Channel Identification

We propose a blind FIR channel identification method based on the parallel factor (Parafac) analysis of a 3rd-order tensor composed of the 4-th order output cumulants. Our algorithm is based on a single-step least squares (LS) minimization procedure instead of using classical three-step alternating least squares (ALS) methods. Using a Parafac-based decomposition, we avoid any kind of pre-processing such as the prewhitening operation, which is mandatory in most methods using higher-order statistics. Our method retrieves the channel vector without any permutation or scaling ambiguities. In addition, we establish a link between the cumulant tensor decomposition and the joint-diagonalization approach. Computer simulations illustrate the performance gains that our method provides with respect to other classical solutions. Initialization and convergence issues are also addressed.