JC5A: Service Delay Minimization for Aerial MEC-assisted Industrial Cyber-Physical Systems

Geng Sun, Jiaxu Wu, Zemin Sun, Long He, Jiacheng Wang, Dusit Niyato, Abbas Jamalipour, Shiwen Mao

Introduction

In the era of Industry 5.0, industrial environments are increasingly reliant on Industrial Sensor Devices (ISDs) for real-time monitoring and automatic control. These tasks are often computation-intensive and extremely delay-sensitive, such as high-speed robotics or AR/VR applications. However, ISDs typically possess limited processing power and battery life. While Mobile Edge Computing (MEC) aims to alleviate these constraints by placing servers closer to the network edge, traditional terrestrial MEC infrastructure is often hindered by high deployment costs and non-line-of-sight (NLoS) issues in complex industrial layouts.

System Architecture

To address these challenges, the authors propose a three-layer collaborative architecture consisting of ISDs at the bottom, a UAV cluster in the middle, and a Macro Base Station (MBS) at the top. UAVs serve as flexible, aerial edge servers that can move to optimize communication links.

The three-layer collaborative UAV-assisted MEC architecture for industrial environments.

This architecture enables a multi-tier offloading strategy: tasks can be processed locally on the ISD, offloaded to a "home" UAV, shared with neighboring UAVs in the cluster, or sent to the MBS. This collaborative approach effectively creates an "aerial cloud" to maximize resource utilization.

The JC5A Optimization Framework

The core contribution of this work is JC5A, a joint optimization framework designed to minimize service delays by simultaneously managing communication, computation, caching (the "3C" resources), and UAV trajectories. The resulting problem is a Mixed-Integer Non-Linear Programming (MINLP) problem, which is NP-hard.

To solve this efficiently, JC5A decomposes the original problem into three iterative subproblems:

The JC5A algorithm framework decomposing the original MINLP problem into three manageable subproblems.
  1. Computation Offloading and Service Caching (SP1): This subproblem determines where tasks are processed and which services are cached on UAVs. It is solved using the Block Successive Upper Bound Minimization Method of Multipliers (BSUMM), enabling parallelized and distributed decision-making.
  2. Resource Allocation (SP2): This manages bandwidth and CPU cycle distribution across the network. By treating it as a convex optimization problem, it can be solved using standard tools like CVX.
  3. UAV Trajectory Control (SP3): This optimizes flight paths to maintain high-quality communication links while respecting safety and speed constraints. The authors employ Successive Convex Approximation (SCA) to handle the non-convex geometry of the trajectory.

Key Insights: The 3C Synergy

A critical insight of this research is the integration of service caching as the "third pillar" alongside communication and computation. If a UAV lacks the specific software service required for a task, it cannot process it locally, leading to high-latency fetches from the MBS. By coupling caching decisions with trajectory and offloading, JC5A ensures that the right resources are in the right place at the right time.

Furthermore, the authors highlight that mathematical decomposition (JC5A) achieves better performance with lower overhead compared to Deep Reinforcement Learning (DRL) approaches, which often suffer from long training times and high sample complexity in industrial settings.

Results and Performance

The JC5A approach was evaluated in a simulated 1000m x 1000m industrial environment with 30 ISDs and 4 UAVs. Compared to benchmarks such as Local Computing (LC), Static UAVs (SU), and DRL methods like PPO and DDPG, JC5A demonstrated:

  • Significant Delay Reduction: JC5A consistently achieved the lowest Average Completion Delay (ACD).
  • Higher Efficiency: The Average Processing Rate (APR) was superior, indicating better hardware utilization.
  • Rapid Convergence: The algorithm demonstrated stable convergence within a few iterations, making it viable for real-time industrial adjustments.

Conclusion and Future Work

JC5A provides a scalable and mathematically grounded solution for aerial MEC in industrial settings. While current models assume fixed altitudes and quasi-static environments, future work aims to incorporate 3D trajectory optimization and energy harvesting (solar or RF) to further enhance the sustainability and flexibility of UAV-assisted ICPS.

Summary updated

BibTeX
@article{sun2025jc,
  author = {Geng Sun and Jiaxu Wu and Zemin Sun and Long He and Jiacheng Wang and Dusit Niyato and Abbas Jamalipour and Shiwen Mao},
  title = {{JC5A: Service Delay Minimization for Aerial MEC-assisted Industrial Cyber-Physical Systems}},
  journal = {IEEE Trans. Services Computing},
  year = {2025},
  url = {https://openreview.net/forum?id=3f56aqEoVR},
}
Figure