Online Scheduling for Energy Minimization in Wireless-Powered Mobile Edge Computing

This repository contains the Python research implementation associated with the IEEE WCNC 2022 paper "Online Scheduling for Energy Minimization in Wireless Powered Mobile Edge Computing." It simulates a multi-access-point, multi-device wireless-powered mobile edge computing (WP-MEC) system and evaluates online scheduling policies for wireless power transfer, local computation, and computation offloading.

Research snapshot. This is the original simulation-oriented codebase, preserved and documented for research reference. It is not currently a bit-exact reproduction package for every number or figure in the paper.

Related paper

Xingqiu He, Yuhang Shen, Xiong Wang, Sheng Wang, Shizhong Xu, and Jing Ren, "Online Scheduling for Energy Minimization in Wireless Powered Mobile Edge Computing," 2022 IEEE Wireless Communications and Networking Conference (WCNC), pp. 1146-1151, 2022.

• IEEE DOI: 10.1109/WCNC51071.2022.9771592
• Open-access preprint: arXiv:2104.10893
• DBLP record

@inproceedings{he2022online,
  author    = {Xingqiu He and Yuhang Shen and Xiong Wang and Sheng Wang and
               Shizhong Xu and Jing Ren},
  title     = {Online Scheduling for Energy Minimization in Wireless Powered
               Mobile Edge Computing},
  booktitle = {2022 IEEE Wireless Communications and Networking Conference
               (WCNC)},
  pages     = {1146--1151},
  year      = {2022},
  doi       = {10.1109/WCNC51071.2022.9771592}
}

Implemented model

The simulator represents wireless devices (WDs) served by multiple access points (APs). In each time slot, a scheduling policy coordinates:

1. wireless power transfer from an AP;
2. local computation at each WD;
3. computation offloading and WD-to-AP association;
4. transmit power, CPU frequency, and time allocation;
5. queue and battery-state updates.

The code includes three policy paths:

ID	Policy	Implementation
0	Proposed WP-MEC online policy	Algorithm.executeWPMEC()
1	Local-computation-only baseline (LCO)	Algorithm.executeLCO()
2	Full-offloading baseline (FO)	Algorithm.executeFO()

The proposed path uses Lyapunov-style queue and battery terms, closed-form updates where available, numerical constrained solves, and Hungarian assignment for WD-to-AP association.

Repository map

Path	Purpose
Scheduler.py	Main experiment driver and parameter sweeps.
Algorithm.py	Proposed policy and baseline implementations.
Environment.py	Queue, battery, channel, and arrival-state transitions.
Parameter.py	System parameters and optional topology generation.
GenerateMap.py	Generates reusable AP/WD topology files.
Auxiliary.py	Numerical helper functions.

Environment

The documented compatibility environment uses Python 3.11 with NumPy 1.24 and SciPy 1.10:

conda env create -f environment.yml
conda activate wpmec-scheduling

This environment has been created successfully on Windows. Module imports and a reduced one-slot LCO check pass. The proposed WP-MEC path, FO path, and full default sweep have not been validated as regression tests in this environment.

Running the simulation

The default script runs the M sweep for the three policies and writes change_m.mat in the repository root:

python Scheduler.py

Scheduler.py has no if __name__ == "__main__" guard, so importing it also starts this sweep. The default workload evaluates five values of M, three policies, and 500 time slots per policy/value pair; expect a long-running numerical experiment rather than a quick-start command.

The experiment selector near the bottom of Scheduler.py uses the following IDs:

• 1: sweep the Lyapunov trade-off parameter V;
• 2: sweep the number of wireless devices N;
• 3: sweep the number of access points M.

For example, type = {1, 2, 3} runs all three sweeps. Each output MATLAB file contains energy and queue-length series for the proposed policy, LCO, and FO.

Topology files are not consumed by the checked-in default experiment. To generate artifacts for inspection or manual integration:

mkdir data
python GenerateMap.py

Reproducibility notes

• Parameter.py initializes NumPy with seed 47, but the three policies run sequentially and do not currently replay an identical pre-generated random trace.
• Environment.step() regenerates the complete uplink/downlink channel matrices inside the per-device loop. A single time slot therefore resamples them N times, and different device updates can observe different channel snapshots. This should be corrected before interpreting policy comparisons.
• Several code constants use numerical scaling for queues, batteries, and bit units. Treat the checked-in configuration as an implementation snapshot, not as a complete paper-parameter manifest.
• The repository does not include the original plotting script or a golden set of paper-figure outputs.
• The current implementation retains historical NumPy array-to-scalar operations. They are accepted by the documented NumPy version, but should be made explicit before adopting newer numerical-library releases.

Scope and limitations

This code is intended for simulation and research inspection. It is not a production MEC controller, does not communicate with physical APs or devices, and has not been safety-qualified for deployment. Before using numerical results in a publication, record the commit, environment, random trace, configuration, and generated output files.

License status

This historical repository does not yet include a software license. The provenance and reuse terms of inherited code must be confirmed before a new license can be applied. Until then, public availability should not be interpreted as permission to redistribute or relicense the source.