A hybrid architecture for millimeter wave (mmW) massive MIMO systems is considered practically implementable due to low power consumption and high energy efficiency. However, due to the limited number of RF chains, user selection becomes necessary for such architecture. Traditional user selection algorithms suffer from high computational complexity and, therefore, may not be scalable in 5G and beyond wireless mobile communications. To address this issue, in this letter we propose a low complexity CNN framework for user selection. The proposed CNN accepts as input the channel matrix and gives as output the selected users. Simulation results show that the proposed CNN performs close to optimal exhaustive search in terms of achievable rate, with negligible computational complexity. In addition, CNN based user selection outperforms the evolutionary algorithm and the greedy algorithm in terms of both achievable rate and computational complexity. Finally, simulation results also show that the proposed CNN based user selection scheme is robust to channel imperfections.

中文翻译：

---

使用卷积神经网络的毫米波大规模 MIMO 系统中的用户选择

由于低功耗和高能效，毫米波 (mmW) 大规模 MIMO 系统的混合架构被认为具有实际可行性。然而，由于 RF 链的数量有限，因此此类架构需要用户选择。传统的用户选择算法计算复杂度高，因此可能无法在 5G 及无线移动通信之外进行扩展。为了解决这个问题，在这封信中，我们提出了一个用于用户选择的低复杂度 CNN 框架。所提出的 CNN 接受通道矩阵作为输入，并将选定的用户作为输出。仿真结果表明，所提出的 CNN 在可实现速率方面执行接近最优的穷举搜索，计算复杂度可以忽略不计。此外，基于 CNN 的用户选择在可实现速率和计算复杂度方面都优于进化算法和贪婪算法。最后，仿真结果还表明，所提出的基于 CNN 的用户选择方案对信道缺陷具有鲁棒性。