一种具有干扰识别功能的基于神经网络的自适应MIMO接收机
An Adaptive MIMO Receiver Based on Neural Network with Interference Recognition

对于多输入多输出(MIMO)系统，对数据量增长的需求促使科研者提出可以减少解码计算量并且保证接收准确性的新型技术。本文研究了基于机器学习的应用以提出一种自适应的干扰感知接收机。对于典型的接收机，干扰抑制合并(IRC)因其可以抑制干扰而能提供更好的性能，但其复杂度要高于最大比合并(MRC)。考虑到性能和计算复杂度的折中，本文提出了一种基于神经网络的自适应接收机，其可以根据信道状态在MRC和IRC间自适应切换。在本文提出的接收机中，从干扰相关矩阵中提取的特征及通过性能分析计算相应的标签用于训练神经网络。该接收机可以自动识别干扰等级并且选择合适的接收机。仿真表明本文提出的接收机可以实现更高的分类准确性，更低的计算复杂度及同IRC相近的性能。
For Multiple Input Multiple Output (MIMO) system, the increasing demand for large amount of data has motivated researchers to pursue novel techniques to reduce decoding computational complexity while ensuring accuracy. In this paper, we study the application of machine learning for proposing an adaptive interference aware receiver. For typical MIMO receivers, Interference Rejection Combining (IRC) provides better performance with regard to interference rejection, but its computational complexity is higher than Maximum Ratio Combining (MRC). Considering the tradeoff between performance and computational complexity, we propose a neural network based adaptive receiver, which can switch between MRC and IRC adaptively according to the channel condition. In our proposed receiver, the extracted features from interference covariance matrix and corresponding class label through performance analysis are used to train the neural network. Then the receiver can recognize interference level and select the appropriate receiver automatically. Simulation results demonstrate that our proposed receiver achieves higher classification accuracy, lower computational complexity and the similar performance as IRC.