Online-Learning-Based Predictive Optimization of Uplink Scheduling for Industrial Internet-of-Things

Abstract

The industrial Internet of Things (IIoT) operates in dynamic environments where wireless channels are subject to rapid changes, posing significant challenges for reliable data transmission. This paper introduces a novel online learning approach to predictively optimize uplink scheduling for IIoT devices. In harsh industrial settings, the unpredictability of channel conditions and data arrivals necessitates immediate data transmission to ensure timeliness and representativeness. We propose a primal-dual online learning framework that integrates stochastic gradient descent (SGD) and online convex optimization (OCO) to generate predictive uplink schedules. By learning only from past channel changes and data arrivals, our predictive schedule can asymptotically minimize the amount of data dropped at the IIoT devices. We also accelerate the online learning by having the IIoT devices oversample their channels to reduce the penalty of the predictive schedule. The optimality loss is proved to asymptotically diminish with the decrease of SGD/OCO stepsizes and the increase of oversampling rate even in fast-changing IIoT environments. Simulation results validate the effectiveness of our approach, showing significant improvements in system throughput compared to state-of-the-art methods, especially in environments with rapidly changing wireless channels.