Techniques for performing decision feedback equalization are described. A feed-forward filter response and a feedback filter response are derived based on a channel estimate and a reliability parameter and further without constraint on the feedback filter response or with a constraint of no feedback for an on-time sample. The reliability parameter is indicative of the reliability of the feedback used for equalization and may be frequency dependent or frequency invariant. Different feed-forward and feedback filter responses may be derived with different constraints on the feedback filter and different assumptions for the reliability parameter. Equalization is performed with the feed-forward and feedback filter responses. If equalization is performed for multiple iterations then, for each iteration, the reliability parameter may be updated, the feed-forward and feedback filter responses may be derived based on the updated reliability parameter, and equalization may be performed with the filter responses for the iteration.