The massive use of transmit and receive antennas with perfect channel state information (CSI) in massive MIMO systems (Multiple Input Multiple Output) can lead array power gain increments proportional to the number of antennas. For CSI estimation pilots can be used, but in multi-user scenarios pilot contamination may occur which may compromise estimation, even when are used orthogonal sequences such as Chu sequences. In such conditions even a precoding does not cancel interference among users and the system performance can be compromised. It is shown that a decision directed scheme based on a iterative block frequency domain equalizer can be used to compensate the pilot contamination impact on channel estimation without sacrificing the complexity. Moreover, when the coherence time spans multiple data blocks, these most accurate channel estimates can be used to precod the subsequent blocks and thereby improve system performance. It is also presented a set of performance results that sustain our assumption.