基于ROAMP-Net的大规模MIMO系统智能信号检测方法

Intelligent signal detection method based on ROAMP-Net for massive MIMO systems

针对大规模多输入多输出(multiple-input multiple-output,MIMO)系统存在的信号检测计算复杂度高、检测精度不足等问题,参考OAMP-Net算法思想,引入残差结构,提出了一种新的智能信号检测网络模型ROAMP-Net。将正交近似消息传递(orthogonal approximate message passing,OAMP)估算信号的迭代过程展开为深度学习网络,同时引入残差结构,分别对各网络层的线性和非线性估计值进行逐层修正,有效防止估计误差的前向传播和过程积累,避免网络模型随着网络层数增加而发生性能退化,从而提高最终信号检测的准确度。针对不同调制方式和不同天线阵列的系列仿真实验结果表明,不同调制方式和天线阵列下ROAMP-Net在检测准确度上均有不错的性能表现。

The signal detection in massive multiple-input multiple-output(MIMO)systems usually confronts the challenges of high computation complexity and low detection accuracy.Artificial intelligence technologies have been widely applied to improve the performance of signal detection.OAMP-Net is a signal detection algorithm based on deep learning,and its comprehensive performance is relatively better than other typical signal detection algorithms.Inspired by the ideas of OAMP-Net,we propose a new intelligent signal detection model,i.e.ROAMP-Net,by introducing residual structure.In ROAMP-Net,the iteration of orthogonal approximate message passing(OAMP)is extended to a deep learning network.Meanwhile,to prevent the performance degradation of deep network with the increase of network layers,the model introduces residual structure to correct the linear and non-linear signal estimation layer by layer,so that the estimation errors would not be forwarded and accumulated.Consequently,high accuracy of signal detection can be expected.Simulation experimental tests suggest that ROAMP-Net outperforms many benchmarks on the accuracy of signal detection under different modulation methods and antenna arrays.