Identificación ciega de sistemas SIMO con señal de entrada dispersa

Abstract

We consider the blind identification of FIR channels with a single input and multiple outputs when the input signal is sparse. The problem is equivalent to identifying the mixing matrix for underdetermined blind source separation, but with temporal correlation among the sources. The length of each channel is assumed known, or previously estimated. Exploiting the sparse character of the input signal, the algorithm solves sequentially the three identification problems: estimating the directions of each column of the channel matrix; estimating their L₂-norm; and finding the most likely order of the columns. The performance of the algorithm in additive noise and its computational cost are compared against subspace-based techniques.