Resource Allocation and Coordination for Critical Messages Using Finite Memory Learning
Taehyeun Park, Walid Saad · 2016
In this paper, a novel framework for enabling machine-to-machine (M2M) communications between machine type devices (MTDs) having heterogeneous message types and under stringent resource constraints is proposed. In particular, an M2M communication system is considered in which two types of messages must seamlessly coexist: periodic, delay tolerant messages, such as meter readings, and critical, real-time messages that indicate the occurrence of critical events, such as a fire or a demand response. Due to the unpredictable occurrence of the critical messages, the MTDs must autonomously learn how to adjust their transmission parameters to ensure the timely delivery of the critical messages to the base station (BS). To address this problem, a novel approach based on the framework of sequential learning with finite memory is proposed. In this approach, the MTDs can learn how many critical messages exist, and collectively allocate the uplink transmission resources needed for the critical messages. Moreover, the proposed learning approach does not require the MTDs to be omniscient and only requires partial, finite information. The bounds on effectiveness of the proposed learning algorithm are derived. Simulation results show that, using the proposed scheme, the probability of successful transmission can quickly reach 99% with minimal memory requirement. The results also show that the proposed learning framework can quickly coordinate the MTDs with critical messages and prevent repeated transmission failures.