Minimizing Task Age upon Decision for Low-Latency MEC Networks Task Offloading with Action-Masked Deep Reinforcement...

Area of Science:

• Wireless Communication Networks
• Edge Computing
• Information Theory

Background:

• Mobile Edge Computing (MEC) networks enable low-latency services by processing data closer to users.
• Finite Blocklength (FBL) codes are crucial for achieving low-latency transmissions in these networks.
• Timeliness of information for decision-making is critical in dynamic environments.

Purpose of the Study:

• To introduce and minimize the Age upon Decision (AuD) metric for task timeliness in MEC networks.
• To address dynamic task generation and random fading channels.
• To jointly optimize User Equipment (UE) selection and blocklength allocation for AuD minimization.

Main Methods:

• Formulating the AuD minimization problem.
• Transforming the optimization problem into a Markov Decision Process (MDP).
• Proposing the Error Probability-Controlled Action-Masked Proximal Policy Optimization (EMPPO) algorithm.

Main Results:

• The proposed EMPPO algorithm significantly reduces AuD compared to baseline methods.
• The design demonstrates robustness across various network conditions, including differing Signal-to-Noise Ratio (SNR) levels.
• Effectiveness is particularly noted in scenarios with low average SNR.

Conclusions:

• The EMPPO algorithm effectively minimizes task AuD in low-latency MEC networks.
• The approach enhances information timeliness for critical decision-making moments.
• The method shows strong potential for real-time applications requiring high data freshness.