AoI-Aware Resource Allocation for Smart Multi-QoS Provisioning

Abstract

The Age of Information (AoI) has recently gained recognition as a critical quality-of-service (QoS) metric for quantifying the freshness of status updates, playing a crucial role in supporting massive ultra-reliable and low-latency communications (mURLLC) services. In mURLLC scenarios, due to the inherent system dynamics and varying environmental conditions, optimizing AoI under such multi-QoS constraints considering both delay and reliability often results in non-convex and computationally intractable problems. Motivated by the demonstrated efficacy of deep reinforcement learning (DRL) in addressing large-scale networking challenges, this work aims to apply DRL techniques to derive optimal resource allocation solutions in real time. Despite its potential, the effective integration of FBC in DRL-based AoI optimization remains underexplored, especially in addressing the challenge of simultaneously upper-bounding both delay and error-rate. To address these challenges, we propose a DRL-based framework for AoI-aware optimal resource allocation in mURLLC-driven multi-QoS schemes, leveraging AoI as a core metric within the finite blocklength regime. First, we design a wireless communication architecture and AoI-based modeling framework that incorporates FBC. Second, we proceed by deriving upper-bounded peak AoI and delay violation probabilities using stochastic network calculus (SNC). Subsequently, we formulate an optimization problem aimed at minimizing the peak AoI violation probability through FBC. Third, we develop DRL algorithms to determine optimal resource allocation policies that meet statistical delay and error-rate requirements for mURLLC. Finally, to validate the effectiveness of the developed schemes, we have executed a series of simulations.