| 摘要: |
| 在C-RAN（Centralized,Cooperative,Cloud Radio Access Network）无线网络基于转移矩阵的负载预测方法中，虽然该预测矩阵具有稀疏特性，但是现有的技术缺乏对稀疏特性加以利用，从而造成计算复杂。针对此问题，提出了一种基于稀疏性的预测矩阵求解算法。该算法对网络状态转移矩阵进行分块迭代，每次等分4块，并分别定义4块矩阵的偏移量。当属于同一行的块矩阵的偏移量有一个是零矩阵时，直接得出所求矩阵对应块的元素全部为零，然后进行下一次迭代；当属于同一行的块矩阵偏移量都不为零矩阵时，通过对矩阵方程组变形处理，转换成迭代格式，然后分块处理。最后，结合仿真定量分析稀疏矩阵稀疏度的临界值问题，给出了稀疏度与计算量之间的关系，并证明了其合理性。仿真结果表明，所提算法能够在不影响预测准确度前提下，降低复杂度。 |
| 关键词: C-RAN网络 大规模稀疏矩阵 矩阵分块 稀疏度 |
| DOI： |
|
| 基金项目:国家高技术研究发展计划（863计划）项目(2014AA01A701)；重庆邮电大学博士基金启动项目 |
|
| A sparsity-based prediction matrix solving algorithm for C-RAN networks |
| KUANG Ling,LIU Zhanjun,TAN Xin,LIU Yang |
| (Chongqing Key Laboratory of Mobile Communications Technology，Chongqing University of Posts and Telecommunications,Chongqing 400065,China) |
| Abstract: |
| In load matrix based load prediction method in the C-RAN（Centralized,Cooperative,Cloud Radio Access Network）wireless network,although the prediction matrix has sparse characteristics,the existing technologies lack the use of sparse features,resulting in computational complexity.For this issue,an algorithm for predicting matrix based on sparsity is proposed.The algorithm performs block iteration on the network state transition matrix,and divides 4 blocks each time,and defines the offset of 4 matrices respectively.When one of block matrix belonging to the same column is tested as a zero matrix,it is directly derived that all the elements of the corresponding block of the requested matrix are zero,then the next iteration is proceeded;When none of the block matrices belonging to the same column has been checked out to be zero matrices,by transforming the matrix equations,it is converted into an iterative format and then processed in blocks.Finally,the critical value of sparse matrix sparsity is quantitatively analyzed by simulation,and the relationship between the sparsity and the computation is given and its rationality is proved.Simulation shows that the algorithm can reduce the computational complexity without influencing the calculation accuracy. |
| Key words: C-RAN large-scale sparse matrix matrix partition sparsity |
