This work proposes a transmitter antenna selection (TAS) method for multiple-input multiple-out (MIMO) nonorthogonal multiple access (NOMA) that is a promising multiple access technique for the fifth generation mobile communications systems. Specifically, we propose to activate the more suitable subset of base station antennas while allocating users into appropriate NOMA clusters such that the system operates in energy efficiency mode, selecting such appropriate antenna subset indexes that maximize the sum-rate (SR) of the NOMA-MIMO system. Once that the TAS based on exhaustive-search is very complex to be implemented in real communication systems, we propose an effective TAS method based on machine learning while keeping very promising performance. A convolutional neural network-based transmitter antenna selection (CNN-TAS) method is proposed for efficiently select antennas aiming at maximizing the system SR. Hence, extensive numerical results demonstrate that the CNN-TAS can suitably learn the problem, performing with very high accuracy choosing properly the antenna subset that maximizes the system spectral efficiency while reducing substantially the processing time of real-time operations.

中文翻译：

---

机器学习辅助的非正交多址多输入多输出系统中的天线选择

这项工作提出了一种用于多输入多输出 (MIMO) 非正交多址 (NOMA) 的发射机天线选择 (TAS) 方法，这是一种用于第五代移动通信系统的有前途的多址技术。具体来说，我们建议激活更合适的基站天线子集，同时将用户分配到合适的 NOMA 集群，以便系统在能效模式下运行，选择合适的天线子集索引来最大化 NOMA 的总速率（SR）-多输入多输出系统。一旦基于穷举搜索的 TAS 在实际通信系统中实施起来非常复杂，我们提出了一种基于机器学习的有效 TAS 方法，同时保持非常有希望的性能。提出了一种基于卷积神经网络的发射机天线选择 (CNN-TAS) 方法，用于有效选择天线，旨在最大化系统 SR。因此，大量的数值结果表明 CNN-TAS 可以适当地学习问题，以非常高的精度正确选择天线子集，最大限度地提高系统频谱效率，同时显着减少实时操作的处理时间。