基于字典学习的毫米波大规模MIMO系统混合预编码

针对毫米波大规模多输入多输出（MIMO）系统混合预编码方案设计的难题，采用字典学习的思想，提出了一种高精度混合预编码方案。该方案首先对全数字预编码矩阵的各列采用稀疏表示；进而按列将字典原子从稀疏表示中分离出来，通过对误差矩阵采用奇异值分解（SVD）来更新对应的字典原子，直到所有字典原子更新，以形成新的字典矩阵;最后，利用更新后的字典矩阵稀疏重构全数字预编码矩阵，以得到模拟预编码矩阵和数字预编码矩阵。仿真结果表明，相较于基于正交匹配追踪（OMP）的混合预编码方案，所提方案在提升系统频谱效率和降低误码率方面具有明显的优势。

Hybrid precoding based on dictionary learning in millimeter wave massive MIMO systems

For the difficult problem of designing the hybrid precoding for millimeter wave(MMW) massive multiple-input multiple-output(MIMO) systems,a high-precision hybrid precoding scheme based on dictionary learning is proposed. The scheme firstly sparsely represents each column of the full digital precoding matrix,then seperates each dictionary atom from the sparse representation by column and updates corresponding dictionary atom via performing singular value decomposition(SVD) on the error matrix. Until the entire dictionary atoms are completely updated,the new dictionary matrix is achieved. Finally,the analog precoding matrix and digital precoding matrix can be obtained by sparsely reconstructing the full digital precoding matrix. Simulation results show that the proposed scheme greatly outperforms the hybrid precoding method based on orthogonal matching pursuit(OMP) algorithm in terms of increasing spectrum efficiency and decreasing bit error rate(BER).