Multi-channel real-time access with starvation avoidance for heterogeneous data in smart factories

Abstract

In Industrial Wireless Control Networks (IWCNs), Industrial Devices (IDs) generate massive amounts of Data Packets (DPs) with different Quality of Service (QoS) requirements. However, most of the existing works set different priorities for differentiated transmission of heterogeneous data, and the high-priority DPs will access the channel immediately after they are generated. This may result in the access starvation of low-priority DPs in time-frequency resource-constrained IWCNs. In this paper, we study the collaborative transmission algorithm of heterogeneous data to avoid access starvation for lower-priority DPs while guaranteeing QoS for higher-priority DPs. Specifically, we first design an edge-assisted learning architecture with multi-access edge computing to assist the training of the algorithm. Then, to mitigate access conflicts among IDs, a gated recurrent unit enhanced Multi-Agent Deep Reinforcement Learning (MADRL) framework was adopted. Based on the framework, we propose a Multi-criteria Decision based dynamic Multi-channel Access (MDMA) algorithm, where high-priority DPs can consider waiting for access according to their own criteria to avoid preempting the channel access opportunity of low-priority DPs approaching the deadline. Extensive simulations show that the proposed MDMA algorithm outperforms the existing algorithms in terms of the average channel utilization rate and the average completion rate of heterogeneous data.