Robust blind channel equalization based on input decision information

This paper presents two new blind learning algorithms to achieve robust convergence for linear or nonlinear equalization. Rather than only using the output information contained in equalizer's output signals, the input decision information involved in the input signals is employed to assist the blind learning procedure. Based on this input information, two blind algorithms, Benveniste-Goursat input-output-decision (BG-IOD) and Stop-and-Go input-output-decision (SAG-IOD) are proposed. Extensive simulations show that the proposed algorithms are superior to existing algorithms such as stochastic quadratic distance (SQD) and dual mode constant modulus algorithm (DM-CMA) in terms of preventing local convergence for linear equalization with random initial conditions or nonlinear equalization using neural works.