Multiple-Input Multiple-Output (MIMO) is one of the key technology components in the fifth generation communication (5G). Inevitably, the growing antenna number in a limited space makes mutual coupling (MC) effect severely influence the system performance of wireless communication. Therefore, accurate MC modeling is of great significance to eliminate MC effects. To this end, we propose a new MC modeling approach based on deep learning to model MC effect in massive MIMO system. For the sake of realizing the processing of complex data with deep neural network (DNN), we divide the complex data into real and imaginary parts, and deduce the expression of mean square error (MSE) of complex data. Then a baseband decoupling algorithm with hybrid beamforming structure has been proposed to eliminate the MC effect. Simulation results show that, in terms of MSE and bit error rate (BER), the proposed deep learning based MC modeling method has an impressive modeling accuracy and perfect performance of decoupling.

中文翻译：

---

基于深度学习的 MIMO 系统互耦建模和基带解耦算法

多输入多输出 (MIMO) 是第五代通信 (5G) 中的关键技术组件之一。不可避免地，有限空间内不断增长的天线数量使得互耦合（MC）效应严重影响无线通信的系统性能。因此，准确的 MC 建模对于消除 MC 效应具有重要意义。为此，我们提出了一种新的基于深度学习的 MC 建模方法来模拟大规模 MIMO 系统中的 MC 效应。为了实现深度神经网络（DNN）对复杂数据的处理，我们将复杂数据分为实部和虚部，推导了复杂数据的均方误差（MSE）表达式。然后提出了一种混合波束成形结构的基带解耦算法来消除MC效应。仿真结果表明，