A Learning-Based Pre-Allocation Scheme for Low-Latency Access in Industrial Wireless Networks

Abstract

To promote the revolution of Industrial Internet of Things, the next generation communication system is expected to provide latency critical services in industry. However, for the traditional downlink-centric cellular systems, the timely delivery of packets cannot be guaranteed by the default dynamic access scheme due to complex signaling procedure. A promising solution to low-latency access is the resource pre-allocation scheme based on the semi-persistent scheduling (SPS) technique, however at the expense of low spectrum utilization. Aiming to make those pre-allocated resources more rewarding, a so-called DPre, a predictive pre-allocation scheme based on learning for low-latency uplink access in industrial wireless networks, is proposed in this paper. It intelligently explores the correlation of devices’ access behavior and device utility diversity through sequential learning. Thus, flexible and judicious per-allocation decisions in both time and frequency domains can be made in an on-demand manner. Moreover, with the proposed temporal-spatial utility metric, DPre is guaranteed to reserve for more informative devices. Both theoretical analysis and simulation validate its high spectrum utilization through accurate prediction and the potential to pre-allocate for valuable packets.