Performance of massive MIMO with zero-forcing beamforming and reduced downlink pilots

In massive MIMO systems, the number of terminals simultaneously served by the base station (BS) can be up to half of the coherence time in symbols when the cell sum-throughput is pursued. In this case, the conventional downlink training scheme for multiuser MIMO systems, in which the BS beamforms orthogonal pilot sequences of length no less than the number of terminals, is inefficient. We assume ZF beamforming is used and it is found that the desired signal can be much stronger than the interference at each terminal. Therefore, we propose to let the BS beamform non-orthogonal pilot sequences, which are of shorter length than that with the conventional scheme. We then derive a lower bound on the achievable rate and find that our proposed scheme is preferable from numerical results. Furthermore, we note that our proposed scheme performs similarly as the scheme with which each terminal uses the expected value of the effective channel gain to decode the received signal, but our proposed scheme is simpler when implementation is considered.