Online Optimization of Wireless Powered Mobile-Edge Computing for Heterogeneous Industrial Internet of Things

Hao Wu, Xinchen Lyu, Hui Tian

This paper, published in 2019, tackles the complex orchestration of Industrial Internet of Things (IIoT) networks where devices are both wirelessly powered and reliant on edge computing. It introduces an online scheduling framework that balances the need for high data throughput with the reality of hardware limitations and delayed network information.

TL;DR

The authors propose Energy-aware Resource Scheduling (ERS), an online algorithm that optimizes data collection and offloading in IIoT networks powered by Wireless Power Transfer (WPT). The framework uniquely handles a mix of device types (those with and without energy storage) and remains asymptotically optimal even when the network state information (NSI) is outdated or delayed.

What Problem Does This Paper Address?

In a modern "smart factory," thousands of heterogeneous devices (sensors, actuators, controllers) collect data that must be processed at a Mobile-Edge Computing (MEC) node. This environment presents three major hurdles:

  1. Sustainability: Replacing batteries for thousands of devices is impractical; they must harvest energy from wireless signals (WPT).
  2. Heterogeneity: Some devices have high-quality rechargeable batteries (Type-II), while others have low-capacity storage that must be used immediately (Type-I).
  3. Information Delay: In a massive network, the central Access Point (AP) cannot know the exact battery or queue status of every device in real-time due to signaling overhead.

The authors seek to maximize "system utility"—a combination of total throughput and fairness—without requiring "non-causal" (future) knowledge of channel conditions.

Methods

The researchers leverage Lyapunov Optimization to transform a long-term stochastic problem into a series of simpler, per-time-slot sub-problems.

  • ERS-RN (Real-time NSI): The baseline algorithm assumes the AP knows everything instantly. It decouples the problem into two parts:
  • Distributed Data Collection: Each device decides how much data to sense based on its current local queue.
  • Joint Energy and Time Allocation: The AP determines how much time to spend on WPT (downlink) versus data offloading (uplink) for each device.
  • ERS-ON (Outdated NSI): This is the more practical extension. It accounts for a feedback interval ($m$) and signaling overhead ($\epsilon$). The AP makes decisions based on the last known state of the devices, and the authors mathematically prove that the resulting "optimality loss" is bounded.
  • Convex Optimization: The sub-problems are formulated as convex issues, allowing them to be solved efficiently using standard tools like CVX.

Datasets

The paper does not use a specific machine learning dataset. Instead, it evaluates the algorithm via MATLAB simulations using parameters derived from:

  • 3GPP Specifications (TR 45.820 and TR 38.825): Used to model realistic path loss and spectrum bandwidth (0.2 MHz).
  • Synthetic Channel Models: Rayleigh fading models were used to simulate the short-term fluctuations of wireless signals in an industrial warehouse setting.

Key Insights and Ideas

  • The $[O(1/V), O(V)]$ Tradeoff: The algorithm uses a control parameter $V$. Increasing $V$ pushes the system closer to the theoretical maximum throughput but causes data queues (and thus latency) to grow linearly. This allows operators to tune the system based on whether they prioritize speed or low latency.
  • Solving the "Doubly Near-Far" Problem: In WPT-MEC systems, a device far from the AP suffers twice: it receives less power and has a weaker signal to send data back. By incorporating Proportional Fairness into the utility function, the ERS algorithm ensures distant devices aren't "starved" of resources.
  • The Value of Heterogeneity: The authors show that treating all devices as if they have no storage (Type-I) leads to massive throughput loss. Type-II devices can "save" energy during poor channel conditions and "spend" it when the channel is clear, a strategy the ERS algorithm exploits.

Key Results

  • Asymptotic Optimality: The ERS algorithm achieves a utility within $B_1/V$ of the theoretical offline optimum (which would require knowing the future).
  • Throughput Gains: ERS-ON significantly outperformed benchmark schemes like Greedy algorithms (which ignore fairness) and Homogeneous-only optimization.
  • Robustness to Stale Info: Even with outdated NSI (where $m=4$ slots), the algorithm maintained high fairness (Jain’s Index near 1.0) and only suffered a linear decrease in throughput as signaling overhead ($\epsilon$) increased.
  • Stability: The algorithm successfully kept all data queues bounded, preventing the "buffer bloat" often seen in simpler greedy scheduling.

Limitations and Future Work

  • Single AP Focus: The current model assumes a single Access Point. In a massive factory, inter-cell interference between multiple APs would be a significant factor.
  • I.I.D. Assumption: The proof relies on independent and identically distributed (i.i.d.) block fading. Real industrial environments might have more complex, correlated shadowing (e.g., moving heavy machinery).
  • Future Directions: The authors suggest integrating Machine-to-Machine (M2M) communication, where devices can relay data for each other to improve scalability and reduce the burden on the AP.

Summary updated

Abstract

A spurt of progress in wireless power transfer (WPT) and mobile edge computing (MEC) provides a promising approach for Industrial Internet of Things (IIoT) to enhance the quality and productivity of manufacturing. Scheduling in such a scenario is challenging due to congested wireless channels, time-dependent energy constraints, complicated device heterogeneity, and prohibitive signaling overheads. In this article, we first propose an online algorithm, called energy-aware resource scheduling (ERS), to maximize the system utility comprising throughput and fairness, with consideration on both system sustainability and stability. Based on Lyapunov optimization and convex optimization techniques, the proposed algorithm achieves asymptotic optimality for heterogeneous IIoT systems without prior knowledge of network state information (NSI). Subsequently, we extend the ERS algorithm to a more realistic scenario where the overhead and delay of NSI feedbacks are non-negligible. The optimal scheduling decisions of the scenario are provided, and the optimality loss on system utility under outdated NSI is analyzed. The simulations verify our theoretical claims and demonstrate the gains of our proposed ERS algorithm over alternative benchmark schemes.
BibTeX
@article{wu2019online,
  author = {Hao Wu and Xinchen Lyu and Hui Tian},
  title = {{Online Optimization of Wireless Powered Mobile-Edge Computing for Heterogeneous Industrial Internet of Things}},
  journal = {IEEE Internet of Things Journal},
  year = {2019},
  url = {https://openreview.net/forum?id=BJQwGVHskf},
}

Loading related content...

Figure