IEEE 802.15.4e 처리량 및 지연 최적화를 위한 TSCH 스케줄링 기법: 딥러닝 기반 접근법

Abstract

IEEE 802.15.4e time-slotted channel hopping (TSCH) sets a new standard for the industrial internet of things (IIoT) due to its simple architecture and productiveness for enhancing credibility in ultra-low-power absorption of industrial appliances. The performance of TSCH is also mainly dominated by the media access control (MAC) mechanism, which consists of refitment, enumeration, composition, and patronization of data transmission schedules that are not accurately prescribed. Most researchers are trying to establish many pragmatic scenarios. Their main approach is to schedule TSCH networks in a centralized way while framing scheduling problems as the nature of throughput and delay in the network. The main approach of this dissertation is to find a quicker and more exact solution for the scheduling of the TSCH network. We utilize the benefits of a deep learning scheme to reduce the execution time of IEEE 802.15.4e TSCH network scheduling. Firstly, we propose a Hungarian-based scheduling solution for TSCH networks by considering throughput and delay with fairness. The scheme proposed previously considered the only throughput for a TSCH network. We utilize maximum link weight alignment in a bipartite graph for TSCH networks to constitute the frames’ cell scheduling. In this dissertation, the weight of the bipartite graph is computed by considering both network throughput and delay. Here, we incorporate a window concept to determine moving average network throughput and delay. The throughput and delay parameters are multiplied by the corresponding moving average throughput and delay values to ensure fairness in the bipartite edge weight. Secondly, we propose a deep learning-based DNN scheme to reduce the execution time of scheduling. The proposed DNN scheme uses the Hungarian solution's training data. When the proposed DNN scheme accepts the weight of the bipartite edge as input, it will offer cell assignments. The proposed DNN scheme is remarkably accurate by learning the relationships between the Hungarian scheduling algorithm's input and output. As a result, it provides quick and precise rational results compared to the Hungarian-based scheduling algorithm. Thirdly, we design a scheduling method considering a TSCH network in coexistence method of interference network cluster (INC). The proposed dual-stage Hungarian-based scheduling method can do the transmission schedule of the TSCH network by avoiding collision from INC and made the throughput maximization of own network with minimizing the INC throughput. The learning-based DNN scheme is also utilized for reducing the execution time of scheduling.