We present a class of constraint LMS-like adaptive linear detection schemes that constitutes a generalization to the popular blind adaptive detector. We show that, contrary to the general belief, the conventional LMS and its constraint version, when in training mode, do not necessarily outperform the blind LMS of Honig et al. (1995). Trained algorithms uniformly outperform their blind counterparts only if they incorporate knowledge of the amplitude of the user of interest. Decision directed versions of such algorithms are shown to be equally efficient as their trained prototypes and significantly better than the blind versions.
展开▼