基于深度学习的 MIMO 信号检测算法优化研究

RESEARCH ON OPTIMIZATION OF MIMO SIGNAL DETECTION ALGORITHM BASED ON DEEP LEARNING↑

郑沛聪

11749126

硕士

信息与通信工程

贡毅

2019-05-30

2019-07-15

哈尔滨工业大学

深圳

多输入多输出（Multi-input Multi-output，MIMO）技术被认为是在无线通信发展历史中最为关键的技术之一，并且也被广泛认为是推动第五代移动通信系统发展的关键技术之一。通过在信号发送端和接收端都配置多根天线，MIMO 通信系统可以在不需要增加额外信道带宽的情况下，提高频谱效率和信道容量。然而，随着发射天线和接收天线的数量的增加，信号检测算法的计算复杂度也会随之提高，这一问题限制了 MIMO 通信的发展。通常，一个高效的 MIMO 信号检测算法应在检测性能和计算复杂度之间取得平衡。因此，本论文在原有的 MIMO 信号检测算法的基础上，努力探索出一种复杂度低且检测准确度相对较高的检测算法。近年来，随着深度学习技术的发展，深度学习技术因其在信号调制识别，资源分配和物理层设计等通信领域问题的突出表现，引起学者们了极大的关注。然而，上述大部分研究都基于数据驱动的深度学习技术解决问题，这种方法将通信系统视为黑盒子并通过使用大量数据来训练网络。为了减少对数据和训练时间的需求，本文介绍了基于模型驱动的深度学习方法。基于模型驱动的深度学习方法基于通信领域知识构建网络。本论文首先研究了基于因子图的近似消息传递（Approximate Message Passing，AMP）MIMO 信号检测算法。然后，利用模型驱动的深度学习方法提 出了 AMP-Net，具体过程为将迭代 AMP 检测算法展开为逐层连接的类神经网络的网络结构，并在每层中添加可训练参数。结合训练数据，利用深度学习技术中的小批量随机梯度下降方法调整每一层中的参数。仿真结果表明，AMPNet 从检测性能、算法收敛速度都远远优于 AMP 检测算法。

Multiple-input Multiple-output (MIMO) systems are one of the most significant developments in wireless communication and have been greatly considered as one of the key technologies for the fifth generation (5G) mobile communication systems. MIMO systems by equipping multiple antennas at both the transmit side and the receive side can increase the spectral efficiency and the channel capacity without the need of extra channel bandwidth. However, one of the key challenges faced by MIMO systems is that the computational complexity of signal detection algorithm increases with the number of transmit antennas and receive antennas. In general, an efficient MIMO detection algorithm is supposed to strike a balance between the error rate performance and the computational complexity. The purpose of this thesis is to comprehensively investigate the detection algorithms of MIMO systems, and based on that, to develop new methods which can reduce the computational complexity while retain good system performance.In recent years, owing to strong learning ability from data, the combination of deep learning and communication has attracted great interests because of its outstanding performances in modulation recognition, resource allocation, and physical layer design. However, most of the existing works focus on data-driven deep learning approaches, which consider the communication system as a black box and train it by using a huge volume of data. In order to reduce the demand for labeled data and training time, the model-driven deep learning approaches are introduced in this paper. The model-driven deep learning approaches construct the network based on communication domain knowledge.In this thesis, we study the approximate message passing detection algorithm based on factor graphs in MIMO system. Then, we propose a novel detection network based on the model-driven deep learning method. The network, called Trainable-AMP-Net, is inspired by the approximate message passing (AMP) approach for MIMO detection. By unfolding the iterative AMP detection algorithm into a layer-wise structure and adding trainable parameters in each layer, we obtain neural-network-like architectures. Then, the trainable parameters are tuned by mini_x0002_batch stochastic gradient descent methods. Simulation results show that the convergence speed, the detection accuracy and the robustness of the detection algorithm are noticeably improved with the aid of fine-tuned parameters.

Massive MIMO 信号检测 近似消息传递算法 深度学习

Massive MIMO signal detection approximate message passing deep learning

中文

联合培养

南方科技大学

郑沛聪. 基于深度学习的 MIMO 信号检测算法优化研究[D]. 深圳. 哈尔滨工业大学,2019.