This paper proposes a joint iterative optimization based hybrid beamforming technique for massive MU-MIMO systems. The proposed technique jointly and iteratively optimizes the transmitter precoders and combiners, aiming to approach the global optimum solution for the system sum-rate maximization problem. The proposed technique develops an adaptive algorithm exploiting the stochastic gradients (SG) of the local beamformers and provides low-complexity closed-form solutions. Furthermore, an efficient adaptive scheme is developed based on the proposed adaptive algorithm and the closed-form solutions. The proposed algorithm requires the signal-to-interference-plus-noise ratio (SINR) feedback from each user and a limited size transition vector to be exchanged between the transmitter and receivers at each step to update beamformers locally. Analytic result shows that the proposed adaptive algorithm achieves low-complexity when the array size is large and is able to converge within a small number of iterations. Simulation result shows that the proposed technique is able to achieve superior performance comparing to the existing state-of-art techniques. In addition, the knowledge of instantaneous channel state information (CSI) is not required as the channels are also adaptively estimated with each coherence time which is a practical assumption since the CSI is usually unavailable or have time-varying nature in real-time applications.

中文翻译：

---

大规模MU-MIMO系统中基于联合迭代优化的低复杂度自适应混合波束成形

本文提出了一种针对大规模MU-MIMO系统的基于联合迭代优化的混合波束成形技术。所提出的技术联合和迭代地优化了发射机预编码器和组合器，旨在为系统求和率最大化问题寻求全局最优解。所提出的技术开发了一种利用局部波束形成器的随机梯度（SG）的自适应算法，并提供了低复杂度的封闭形式解决方案。此外，基于提出的自适应算法和封闭形式的解决方案，开发了一种有效的自适应方案。提出的算法要求在每个步骤中，在每个步骤中，发射机和接收机之间都要交换来自每个用户的信号干扰加噪声比（SINR）反馈，并在有限的大小转换向量之间进行交换，以本地更新波束成形器。分析结果表明，所提出的自适应算法在数组大小较大的情况下实现了低复杂度，并且能够在少量迭代中收敛。仿真结果表明，与现有技术相比，该技术能够获得更好的性能。另外，由于还使用每个相干时间来自适应地估计信道，因此不需要瞬时信道状态信息（CSI）的知识，这是一个实际的假设，因为CSI在实时应用中通常不可用或具有随时间变化的性质。仿真结果表明，与现有技术相比，该技术能够获得更好的性能。另外，由于还使用每个相干时间来自适应地估计信道，因此不需要瞬时信道状态信息（CSI）的知识，这是一个实际的假设，因为CSI在实时应用中通常不可用或具有随时间变化的性质。仿真结果表明，与现有技术相比，该技术能够获得更好的性能。另外，由于还使用每个相干时间来自适应地估计信道，因此不需要瞬时信道状态信息（CSI）的知识，这是一个实际的假设，因为CSI在实时应用中通常不可用或具有随时间变化的性质。