Spatio-Temporal Traffic Prediction and Proactive Resource Management for Scalable LoRaWAN Networks

Abstract

Large scale Internet of Things (IoT) deployments such as Low Power Wide Area Networks (LPWANs) and the LoRaWAN have become central to the large-scale Internet of Things (IoT) deployment because of their long-range communications and low energy use. Nevertheless, the scalability and reliability of LoRaWAN networks is inherently limited by the unslotted ALOHA-based medium access scheme, fixed resource setup and severe regulatory constraints of duty-cycle. The difficulties are magnified in crowded and mobile setups like smart cities, where non-stationary and heavy traffic conditions result in severe packet collisions, high latency and poor quality of service (QoS). In this paper, a smart, predictive, and duty-cycle-aware resource management framework is proposed to be utilized by LoRaWAN networks that would reorganize network operation by transitioning it to reactive control to proactive decision-making. The proposed solution combines a hybrid Spatio-temporal traffic prediction model with Graph Convolutional Networks (GCN) and Gated Recurrent Units (GRU) as well as an adaptive resource management module that is placed at the edge of the network. Through learning of spatial contours between the entities in the network and time-based traffic dynamics, the framework effectively predicts short time congestion and pre-emptively modulates spreading factors, channels, and transmission scheduling and maintains the entire regulatory adherence. Extensive tests based on actual traffic traces and high-density simulation evidence that the suggested framework is far superior to the traditional LoRaWAN Adaptive Data Rate (ADR) schemes and the established machine learning model. GNCN-GRU model provides a 18% decrease in the error of traffic prediction relative to the conventional recurrent models with the resource adaptation being proactive which minimizes the packet collisions by up to 30 percent in the ultra-dense situation. Besides, the framework maintains up to 21% increase in the ratio of packet delivery at the 1000 nodes per gateway, and it ensures a sub-500ms latency of mission-critical traffic despite a rigid duty-cycle limit. The experiments of edge deployment prove the viability of the method, with a latency of inference of less than 42ms and a minimum of computational cost. Comprehensively, the findings indicate that edge Spatio-temporal intelligence is feasible and applicable to scalable, reliable, and regulation-friendly LoRaWAN operation, and the next-generation smart city and industrial IoT applications are feasible.