Transmitter channel state information (CSIT) is indispensable for the spectral efficiency gains offered by massive multiple-input multiple-output (MIMO) systems. In a frequency-division-duplexing (FDD) massive MIMO system, CSIT is typically acquired through downlink channel estimation and user feedback, but as the number of antennas increases, the overhead for CSI training and feedback per user grows, leading to a decrease in spectral efficiency. In this paper, we show that, using uplink pilots in FDD, the downlink sum spectral efficiency gain with perfect downlink CSIT is achievable when the number of antennas at a base station is infinite under some mild channel conditions. The key idea showing our result is the mean squared error-optimal downlink channel reconstruction method using uplink pilots, which exploits the geometry reciprocity of uplink and downlink channels. We also present a robust downlink precoding method harnessing the reconstructed channel with the error covariance matrix. Our system-level simulations show that our proposed precoding method can attain comparable sum spectral efficiency to zero-forcing precoding with perfect downlink CSIT, without CSI training and feedback.

翻译：发射端信道状态信息（CSIT）对于大规模多输入多输出（MIMO）系统实现频谱效率增益至关重要。在频分双工（FDD）大规模MIMO系统中，CSIT通常通过下行信道估计和用户反馈获取，但随着天线数量增加，每个用户的CSI训练与反馈开销随之增长，导致频谱效率下降。本文证明，在FDD系统中，利用上行导频，当基站天线数量在温和信道条件下趋于无穷时，可实现与完美下行CSIT相同的下行总频谱效率增益。该结论的核心思想在于：提出了一种利用上行导频实现均方误差最优的下行信道重构方法，该方法利用了上、下行信道的几何互易性。我们还提出了一种鲁棒的下行预编码方法，该方法利用重构信道及误差协方差矩阵。系统级仿真表明，所提预编码方法无需CSI训练与反馈即可达到与完美下行CSIT下的迫零预编码相当的总频谱效率。