Eigenvector decomposition (EVD) is an inevitable operation to obtain the precoders in practical massive multiple-input multiple-output (MIMO) systems. Due to the large antenna size and at finite computation resources at the base station (BS), the overwhelming computation complexity of EVD is one of the key limiting factors of the system performance. To address this problem, we propose an eigenvector prediction (EGVP) method by interpolating the precoding matrix with predicted eigenvectors. The basic idea is to exploit a few historical precoders to interpolate the rest of them without EVD of the channel state information (CSI). We transform the nonlinear EVD into a linear prediction problem and prove that the prediction of the eigenvectors can be achieved with a complex exponential model. Furthermore, a channel prediction method called fast matrix pencil prediction (FMPP) is proposed to cope with the CSI delay when applying the EGVP method in mobility environments. The asymptotic analysis demonstrates how many samples are needed to achieve asymptotically error-free eigenvector predictions and channel predictions. Finally, the simulation results demonstrate the spectral efficiency improvement of our scheme over the benchmarks and the robustness to different mobility scenarios.

翻译：特征向量分解（EVD）是实际大规模多输入多输出（MIMO）系统中获取预编码器的必要操作。由于基站（BS）天线规模庞大且计算资源有限，EVD庞大的计算复杂度成为制约系统性能的关键因素之一。针对此问题，本文提出一种通过预测特征向量插值预编码矩阵的特征向量预测（EGVP）方法。其核心思想是利用少量历史预编码器插值其余预编码器，从而避免对信道状态信息（CSI）进行EVD运算。我们将非线性EVD转化为线性预测问题，并证明特征向量预测可通过复指数模型实现。进一步，针对移动环境中应用EGVP方法面临的CSI时延问题，提出一种称为快速矩阵束预测（FMPP）的信道预测方法。渐近分析表明，实现渐近无误差特征向量预测和信道预测所需的最少样本数。最后，仿真结果验证了本方案相较于基准方案的频谱效率提升及其对不同移动场景的鲁棒性。