近年来，随着煤矿智能化建设的推进，矿井无线通信系统的标准也在不断提高。然而，矿井通信场景的复杂性导致信号衰落畸变严重，同时还存在信道估计困难、接收端信号解码复杂以及多个异构信息系统共存等问题。这些问题也造成了现有矿井无线通信系统信息传输质量低，接收设备复杂的情况，严重影响了矿井无线通信系统的智能化发展。利用深度学习的技术特点，可以为当前矿井无线通信系统面临的难题提供全新的解决方案。本文针对矿井环境下的多输入多输出正交频分复用(Multiple-Input Multiple-Output, Orthogonal Frequency Division Multiplexing, MIMO-OFDM)系统，在接收端引入深度学习技术来解决矿井无线通信系统信息恢复准确率低、接收端信号解码复杂等问题，有效提升了矿井无线通信系统的可靠性和智能性。主要工作有以下几个方面：

（1）针对复杂矿井环境下无线通信系统接收端信号恢复准确率低的问题，提出了一种基于卷积神经网络的矿井MIMO-OFDM智能接收算法。该接收算法采用改进的卷积神经网络(Convolutional Neural Networks, CNN)代替传统的接收端信号处理过程，实现接收端的整体优化并通过对IQ数据的自主学习实现原始信息的可靠恢复。在此算法思想上设计了一种智能接收算法模型的网络结构，在不同信道条件、不同调制方式、不同天线数目以及动态环境条件下进行了仿真实验，同时还对算法模型的应急能力进行了测试，仿真结果表明相较于传统的接收算法，该智能接收算法具有更佳的信息恢复性能。

（2）为了解决智能矿井通信系统融合发展所遇到的多模信号接收问题，提出了一种基于模型迁移的矿井多模信号智能接收算法。该算法采用两阶段训练的思想，首先使用初始数据集训练一个基准模型，然后对基准模型进行微调训练，从而实现对矿井多模信号的智能接收和恢复。在此算法思想上，针对不同输入位数信号、不同调制方式信号以及有无循环前缀信号进行实验测试，并且针对不同的信道多径场景对该算法进行测试，实验结果表明该算法提升了系统接收端对矿井多模信号的接收性能，并且极大的提升了接收端对矿井信息的恢复速度。

In recent years, with the advancement of intelligent construction in coal mines, the standard of wireless communication systems in mines has been improving. However, the complexity of mine communication scenarios leads to serious signal fading distortion, as well as problems such as difficult channel estimation, complex signal decoding at the receiver end and coexistence of multiple heterogeneous information systems. These problems also cause the existing mine wireless communication system to have low information transmission quality and complex receiving equipment, which seriously affects the intelligent development of mine wireless communication system. Using the technical features of deep learning, a new solution can be provided for the current problems faced by mine wireless communication systems. In this paper, for the Multiple-Input Multiple-Output, Orthogonal Frequency Division Multiplexing (MIMO-OFDM) system in the mine environment, deep learning technology is introduced at the receiver side to solve the problems of mine wireless communication system The problems of low information recovery accuracy and complex signal decoding at the receiver end have effectively improved the reliability and intelligence of the mine wireless communication system. The main work has the following aspects:

(1)For the problem of low accuracy of signal recovery at the receiver end of wireless communication system in complex mine environment, a convolutional neural network-based intelligent reception algorithm for MIMO-OFDM in mine is proposed. The receiver algorithm uses an improved Convolutional Neural Networks (CNN) instead of the traditional receiver signal processing to achieve the overall optimization of the receiver and the reliable recovery of the original information through the autonomous learning of IQ data. The network structure of an intelligent reception algorithm model is designed based on this algorithm idea, and simulation experiments are conducted under different channel conditions, different modulation methods, different number of antennas and dynamic environment conditions, and the emergency response capability of the algorithm model is also tested, and the simulation results show that the intelligent reception algorithm has better information recovery performance than the traditional reception algorithm.

(2)A model migration-based algorithm for intelligent reception of multimode signals in mines is proposed in order to solve the multimode signal reception problem encountered in the convergence development of intelligent mine communication systems. The algorithm adopts the idea of two-stage training, firstly training a benchmark model using the initial data set, and then fine-tuning the training of the benchmark model, so as to achieve intelligent reception and recovery of multimode signals in mines. Based on this algorithm idea, experimental tests are conducted for signals with different input bits, signals with different modulation methods and signals with or without cyclic prefixes, and the algorithm is tested for different channel multipath scenarios. The experimental results show that the algorithm improves the reception performance of mine multimode signals at the receiver side of the system and greatly improves the recovery speed of mine information at the receiver side.