5G海量机器类通信(massive Machine Type Communications,mMTC)场景，是针对未来大规模物联网应用而提出的，具有广泛的应用前景。不同业务类型的海量设备接入蜂窝物联网会导致导频碰撞加剧以及资源短缺等问题，进而影响接入性能。针对上述问题，本项目利用机器学习技术，设计一种基于机器学习的智能非正交非授权随机接入方案，以提升成功接入设备数及资源利用率，进而降低时延和减小功耗。首先，设计基于机器学习的信道信息估计方案以及智能数据检测算法，以缓解导频碰撞，实现信道信息的准确估计和数据信息的高效检测，进而提高成功接入设备数；然后，分析所设计的随机接入方案可支持的最大设备数，为后续资源分配方案的设计提供理论依据；最后，设计基于深度学习的活跃设备数预测模型并设计资源分配方案，以实现资源动态分配，提高资源利用率。本项目的研究，将为我国物联网的发展提供理论支撑和技术指导。

Massive Machine Type Communications (mMTC) in 5G is used to enable the future massive Internet of Things (IoT), which has broad application prospects. Massive MTC devices with different service classes accessing the cellular network will lead to the problems of pilot collision and resource shortages, which will further degrade the access performance. To resolve these problems, we propose a machine-learning-based intelligent non-orthogonal Grant-Free random access scheme by utilizing the machine learning technology. We first design a machine-learning-based channel information estimation scheme and a machine-learning-based intelligent data detection algorithm to alleviate pilot collision, estimate the channel information accurately and detect the data efficiently, and thus improve the number of successful devices. Then, we analyze the maximum number of supportable devices of the proposed random access scheme to provide theory support for the design of resource allocation schemes. Finally, we design a deep-learning-based the number of active devices prediction model and resource allocation schemes to achieve dynamic resource allocation and thus improve resource utilization. The research of this project will provide theory and technique supports for the development of the IoT.