Introduction
In the evolving landscape of the Internet of Things (IoT), many devices are designed to be low-power or even "battery-free," harvesting energy from wireless signals to perform their functions. However, these devices often face a computational bottleneck: the energy required for complex processing exceeds their local harvesting capabilities. This necessitates a synergy between Wireless Power Transfer (WPT) and Mobile Edge Computing (MEC), where tasks can be offloaded to nearby servers.
This paper addresses the joint optimization of energy beamforming and task offloading in a dynamic environment. Unlike static models, this research accounts for fluctuating wireless channels and unpredictable task arrivals, focusing on the critical constraints of energy and task "causality"—the principle that a device cannot consume energy or process data that has not yet arrived.
System Architecture
The researchers model a single-user system comprising three primary entities: a multi-antenna Energy Transmitter (ET), a user device equipped with a CPU and energy harvester, and an Access Point (AP) integrated with an MEC server.
As shown in Figure 1, the ET uses Maximum Ratio Transmission (MRT) beamforming to send power to the user. The user then manages two buffers: one for harvested energy and one for incoming computation tasks ($A_i$). The user must decide in each time slot how much data to process locally ($\ell_i$) and how much to offload ($d_i$) to the MEC server, all while staying within the limits of its harvested energy.
Methodology and Parameters
The optimization problem is formulated to minimize the total energy consumption of the ET over a finite time horizon. The study employs two distinct approaches:
- Offline Optimization: Assuming perfect "look-ahead" knowledge of future channel states and task arrivals, the authors use Karush-Kuhn-Tucker (KKT) conditions to derive optimal policies.
- Online Optimization: For practical implementation where future states are unknown, heuristic algorithms are developed based on the structural properties discovered in the offline analysis.
Figure 2 details the parameters used to define the system's constraints, including the RF-to-DC conversion efficiency ($\eta$), the number of CPU cycles required per bit ($C$), and the power of additive white Gaussian noise ($\sigma^2$).
Key Insights: The Staircase and Energy Peaks
The research reveals several fundamental properties of optimal energy and task management:
- The Staircase Property: In static channels, the optimal task processing rate follows a "staircase" pattern. The user should gradually increase computation intensity as the deadline approaches. This prevents the device from being idle early on and ensures the most efficient use of accumulated energy and data buffers.
- Energy Sporadicity in Fading Channels: When wireless channels fluctuate, the ET should not transmit power continuously. Instead, it should wait for Causality Dominating Slots (CDS)—moments when the channel gain is at a peak—to transmit energy at high power. This energy is then stored by the user for later use during periods of poorer channel quality.
- Staircase Water-Filling: For offloading, the optimal policy combines the staircase intensity increase with a "water-filling" approach. If the energy cost of offloading is too high due to a weak communication channel, the device prioritizes local computing or waits for a better channel state.
Results and Performance
Monte Carlo simulations using Rician fading models demonstrate that this joint allocation scheme significantly outperforms traditional benchmarks, such as "local computing only" or "full offloading." Notably, the heuristic online designs perform remarkably close to the theoretical offline optimum, even without perfect knowledge of future arrivals.
The study also highlights distance sensitivity: as a user moves further from the ET and closer to the AP, the system intelligently shifts from local computing to aggressive offloading to compensate for the reduction in harvested energy.
Limitations and Future Outlook
While providing a robust framework, the current model assumes a linear energy harvesting model and focuses on a single-user scenario. Future research may explore non-linear rectifier circuits and multi-user interference management. Additionally, moving from "infinitely partitionable" tasks to "binary offloading" (where tasks cannot be split) presents a significant combinatorial challenge for future optimization efforts.
Summary updated