An algorithm for dual-channel noiseless signal reconstruction and its performance analysis

We present an algorithm for estimating a two-input two-output system and recovering its inputs from noise-free observations of its outputs. The algorithm involves minimization of a set of cross-correlations between the reconstructed signals and a nonlinear function of these signals. A performance analysis of the algorithm is presented. It is shown that for the optimal choice of the nonlinear function, the performance of the algorithm is close to the Cramer-Rao lower bound (CRLB). Other choices of the function provide insights into the performance of moment/cumulant-based deconvolution techniques.