Optimal Energy Allocation and Task Offloading Policy for Wireless Powered Mobile Edge Computing Systems

Feng Wang, Jie Xu, Shuguang Cui

TL;DR

This paper optimizes the interplay between wireless power transfer (WPT) and mobile edge computing (MEC) in a dynamic environment. By jointly scheduling how an energy transmitter beams power to a user and how that user allocates tasks between local processing and edge offloading, the authors minimize total energy consumption while respecting the "causality" of both energy and task arrivals.

What Problem Does This Paper Address?

In the world of the Internet of Things (IoT), many devices are low-power and "battery-free," relying on wireless power harvested from the environment or dedicated transmitters. At the same time, these devices often need to perform complex computations that exceed their local capabilities, leading them to offload tasks to the "edge"—servers located at nearby access points.

Most existing research treats this as a static problem: the channel is constant, and the tasks are already there. In reality, wireless channels fluctuate (fading), and tasks arrive in unpredictable bursts. This paper tackles the joint optimization of energy and task allocation over a finite time horizon. The core challenge is "causality": a device cannot spend energy it hasn't harvested yet, nor can it execute a task that hasn't arrived yet. The authors seek to minimize the energy spent by the transmitter while ensuring every bit of data is processed before a strict deadline.

Methods

The authors model a single-user system with a multi-antenna energy transmitter (ET), a user, and an MEC-integrated access point. They use convex optimization to solve the problem in two main stages:

  1. Offline Optimization: Assuming perfect "look-ahead" knowledge of future channel states (CSI) and task arrivals (TSI). They derive well-structured optimal policies using Karush-Kuhn-Tucker (KKT) conditions.
  2. For static channels, they prove the ET should transmit energy uniformly, while the user follows a "staircase" task allocation (processing more bits as the deadline approaches).
  3. For time-varying channels, they introduce the concept of Causality Dominating Slots (CDS)—specific moments when the channel is exceptionally good compared to the past—and show that energy should only be transmitted during these "peaks."
  4. Online Optimization: Since real-world devices don't have a crystal ball, the authors developed heuristic online designs. These algorithms use the "staircase" and "water-filling" structures discovered in the offline analysis but adapt them based on past observations and statistical averages of future arrivals.

Datasets

As a theoretical study in communications engineering, this paper does not use a standard "training" dataset. Instead, it evaluates its algorithms using Monte Carlo simulations based on:

  • Rician Fading Models: To simulate realistic distance-dependent wireless channels.
  • Randomized Task Arrivals: Tasks are modeled as uniformly distributed bit-arrivals ($A_i \sim \mathcal{U}[0, A_{max}]$).
  • Numerical Parameters: Standardized values for CPU cycles per bit, noise power, and RF-to-DC conversion efficiency ($\eta = 0.3$).

Key Insights and Ideas

  • The Staircase Property: The optimal strategy for a user is to increase the computation intensity over time. Because energy and tasks accumulate, the "buffer" is most efficiently cleared by ramping up towards the deadline rather than working at a constant, lower rate that might leave the device idle early on.
  • Energy Sporadicity: In fading channels, the transmitter shouldn't be "always on." It is much more energy-efficient to wait for a "causality dominating slot"—a peak in channel gain—and blast energy then, storing it in the user's battery for later use.
  • Staircase Water-Filling: For offloading, the optimal policy resembles a "water-filling" approach where the "water level" (computation level) increases in steps. If the cost of offloading (due to a poor channel) is higher than the current computation level, the device should wait or compute locally.

Key Results

  • Significant Energy Savings: The proposed joint allocation scheme significantly outperforms benchmark strategies like "local computing only" or "full offloading."
  • Proximity to Optimal: The heuristic online designs, which only know the past and the average of the future, perform remarkably close to the "perfect knowledge" offline solutions.
  • Distance Sensitivity: The paper quantifies the trade-off between WPT and offloading. As the user moves further from the energy transmitter (and closer to the AP), the system naturally shifts from local computing to more aggressive offloading to save the now-scarce harvested energy.

Limitations and Future Work

  • Linear Energy Harvesting: The authors assume a linear model for RF-to-DC conversion. In practice, rectifiers are non-linear and saturate at high power levels. Extending this to non-linear models would add significant complexity.
  • Single-User Focus: The current model handles one user. In a real MEC cell, multiple users would compete for the ET’s beamforming and the AP’s computation resources, requiring more complex interference management and multi-user scheduling.
  • Task Partitioning: The paper assumes tasks are "infinitely partitionable" (bit-by-bit). Future work could look at "binary offloading," where a task must either be done entirely locally or entirely at the edge, which turns the problem into a much harder combinatorial (NP-hard) challenge.

Summary updated

BibTeX
@article{wang2020optimal,
  author = {Feng Wang and Jie Xu and Shuguang Cui},
  title = {{Optimal Energy Allocation and Task Offloading Policy for Wireless Powered Mobile Edge Computing Systems}},
  journal = {IEEE Trans. Wireless Communications},
  year = {2020},
  url = {https://openreview.net/forum?id=QJwR9FLtBQ},
}

Loading related content...

Figure