Abstract: This paper proposed a both data- and pilot-aided channel estimation to deal with pilot contamination in time-division duplex massive multi-input multi-output (MIMO) systems. The paths of the desired user are distinguishable from those of interference users due to the approximate sparse nature in angle-domain of the spatial correlated massive MIMO channels, while the length of an uplink sub-frame is much larger than that of the pilot sequence. We proposed to first roughly estimate the channel by using uplink pilot sequence. Then, we regard the decoded uplink data symbols as the pilot sequence to find the positions of the significant coefficients in the angle-domain channel estimate. By extracting the corresponding channel components from the rough channel estimate, a refined channel estimate is obtained. Simulation results show that pilot contamination can be largely removed by the proposed channel estimate, no matter whether the same group of orthogonal pilot sequences are reused or different groups of pilot sequences with cross-correlation are applied among adjacent cells.