A space-time semi-blind equalizer based on constant modulus and decision-direction

This paper considers the problem of directly computing inverse filters that can equalize and separate each user in a multi-user multipath environment. Some known but very small number of training symbols, while alone unable to yield a satisfactory filter, provide reasonable initializations that not only may speed up the convergence of an adaptive blind algorithm, but also guide the algorithm to an easier recovery of weak users. Training for each user does not have to be simultaneous. Decision-direction mode can be used to refine the filter coefficients after the phase ambiguity inherent to a blind approach is corrected for. Simulation results are provided to demonstrate the method's effectiveness.