Millimeter-wave (mmWave) massive multi-user multiple input multiple output (MU-MIMO) systems employ hybrid analog-digital precoding/combining design, to reduce the number of radio-frequency (RF) chains without large sum-rate performance loss. Committee machines combine several expert algorithms to yield better reliable estimates to those of its constituent experts. However, several committee machines approach mainly rely on the ‘quality’ of the joint support-set of expert algorithms combine by committee machine. In this paper, we present hybrid analog-digital precoder and combiner designs for mmWave massive MU-MIMO systems by exploiting the intersection of the estimated support-sets, namely, common support-set of ‘expert’ algorithms in a committee machine methodology. With $S$ denoting the sparsity level, the proposed algorithms perform two essential tasks to identify in which subspace, generated by not more than $S$ columns of the matrix of array response vectors, the optimal hybrid analog-digital precoders or combiners lie: Firstly, the selection of the committee machine common support set, which accepts the section where both the ‘experts’ agree. Secondly, if some column indices of the optimal hybrid analog-digital precoders or combiners do not lie in the current common support-set for the correct spanning space, $-$the algorithm searches over the matrix of array response vectors and selects column indices that exhibit the highest correlation with the signal residual, to acquire the remaining support. We derive the theoretical analysis for performance enhancement of the proposed algorithm. Simulation results demonstrate that the proposed design allows mmWave massive MU-MIMO systems to approach their unconstrained performance limits, and exhibits substantial sum-rate performance improvement over existing regularized channel diagonalization based schemes and the beam steering solution.

毫米波（mmWave）大规模多用户多输入多输出（MU-MIMO）系统采用混合模数预编码/组合设计，以减少射频（RF）链的数量，而不会造成较大的总和速率性能损失。委员会机器结合了几种专家算法，可以为其组成专家得出更好的可靠估计。但是，几种委员会机器方法主要依靠专家机器联合提供的专家算法联合支持集的“质量”。在本文中，我们通过利用估计的支持集的交集，即委员会机器方法中“专家”算法的通用支持集，介绍了毫米波大规模MU-MIMO系统的混合模数预编码器和组合器设计。和$\mathrm{小号}$ 表示稀疏性级别，所提出的算法执行两项基本任务，以识别不超过2个子空间的子空间。 $\mathrm{小号}$在阵列响应向量矩阵的两列中，最优的混合模数预编码器或组合器位于：首先，选择委员会机器通用支持集，该接受集接受了两个“专家”都同意的部分。其次，如果最佳混合模数预编码器或组合器的某些列索引不位于正确的扩展空间的当前公共支持集中，$-$该算法搜索阵列响应向量的矩阵，并选择与信号残差具有最高相关性的列索引，以获取剩余的支持。我们从理论上分析了所提出算法的性能。仿真结果表明，所提出的设计使mmWave大规模MU-MIMO系统能够接近其不受限制的性能极限，并且相对于现有的基于正则化信道对角化的方案和波束控制解决方案，其总和速率性能得到了显着改善。