基于VSINR的分布式多小区MU-MIMO预编码矩阵设计

Distributed precoder design based on VSINR maximization for multi-cell MU-MIMO systems

本文提出一种基于虚拟信干噪比(VSINR,virtual signal-to-interference-plus-noise ratio)的多小区多用户分布式预编码设计方法.由于设计预编码矩阵来最大化系统吞吐量这一问题是非凸的,因此如何寻找一种高效的预编码矩阵设计方法,显得尤为重要.本文基于分布式的信漏噪比(SLNR,signal-to-leakage-plus-noise ratio)模型,通过设计合适的修正系数,寻求分布式VSINR最大化问题与原集中式吞吐量最大化问题之间的等效关系,建立Pareto最优态下的VSINR模型,使得分布式的VSINR模型与集中式的系统吞吐量模型在极值点处关于预编码矩阵的梯度值同时等于零,从而保证上述的两个最大化问题关于预编码矩阵同解;在此基础上,仅利用本地信道信息(CSI,channel stateinformation)即可分布式的求解相应的Pareto最优的预编码矩阵的闭式解.仿真结果表明,所提出的预编码矩阵求解算法,性能优于目前典型的几种方法,不仅能克服集中式全局CSI求解的额外开销,而且所涉及的迭代过程能够快速收敛到最优解,有效地提高小区平均吞吐量.

In this paper, we propose a distributed precoder design algorithm based on virtual signal-to-inter- ference-plus-noise ratio （VSINR） for multi-cell multi-user multiple-input multiple-output （MU-MIMO） system. As the problem of finding the optimal precoder to maximize the system throughput is non-convex and non-trivial, it is important to find low-complexity solutions. Motivated by the recent results in the distributed signal-to-leakage- plus-noise ratio （SLNR） model, we aim to establish equivalence between the distributed VSINR maximization problem and the original centralized throughput maximization problem by designing the proper correction factor, and the correction factor is constructed to satisfy the equivalence condition that the sub-gradients w.r.t the precoder of the VSINR model and the system throughput model become zero at the same time. Then we can build a VSINR model which is proved to be Pareto optimal, thus the entire system throughput can be optimized by using precoders based on an appropriate distributed VSINR model. Finally we can distributedly solve the corresponding Pareto optimal precoding matrix problem based on local CSI （CSI, channel state information） and a closed-form solution of optimal transmitter precoder can be derived. Simulation results show that the proposed precoder can enhance the system throughput with varying transmitting SNRs, active users and interference factors, and outperform other existing distributed precoder design algorithms such as ZF, MRT and SLNR algorithms. Besides, the iteration in our proposed algorithm converges fast within 3 iterations, the approximate optimum can be achieved, and within 20 iterations, and the global optimum can be obtained. Therefore, our proposed algorithm can bring a significant increase to the average cell throughput while effectively saving the system overhead.