In the second part of this paper, we develop a centralized packet
transmission scheduling scheme to pair with the protocol designed in Part I and
complete our medium access control (MAC) design for machine-type communications
in the industrial internet of things. For the networking scenario, fine-grained
scheduling that attends to each device becomes necessary, given stringent
quality of service (QoS) requirements and diversified service types, but
prohibitively complex for a large number of devices. To address this challenge,
we propose a scheduling solution in two steps. First, we develop algorithms for
device assignment based on the analytical results from Part I, when parameters
of the proposed protocol are given. Then, we train a deep neural network for
assisting in the determination of the protocol parameters. The two-step
approach ensures the accuracy and granularity necessary for satisfying the QoS
requirements and avoids excessive complexity from handling a large number of
devices. Integrating the distributed coordination in the protocol design from
Part I and the centralized scheduling from this part, the proposed MAC protocol
achieves high performance, demonstrated through extensive simulations. For
example, the results show that the proposed MAC can support 1000 devices under
an aggregated traffic load of 3000 packets per second with a single channel and
achieve <0.5ms average delay and <1% average collision probability among 50
high priority devices.

该论文提出了用于工业物联网中机器类型通信的媒体访问控制设计，其中包括集中式报文传输调度方案和基于深度神经网络的协议参数确定方法，以满足服务质量要求并避免处理大量设备引起的复杂性。