PaperlayerView exact revision on arXiv

Source-linked AI summary

Mobile Edge Computing in Unmanned Aerial Vehicle Networks

Fuhui Zhou, Rose Qingyang Hu, Zan Li, Yuhao Wang

arXiv:1910.10523v1eess.SPcs.GT

TL;DR

Mobile devices face computation-intensive workloads despite limited computational capability and battery capacity. The paper surveys three UAV-enabled MEC architectures, clarifies implementation issues, and discusses challenges and open issues, concluding that research remains in its early stage.

  • Problem

    Computation-intensive applications strain mobile devices with limited computational capability and finite battery capacity, motivating MEC-based task offloading.

  • Method

    The paper presents a comprehensive survey of state-of-the-art UAV-enabled MEC research, including three architectures, implementation issues, challenges, and open research issues.

  • Results

    The survey clarifies important implementation issues and discusses key challenges and open issues in UAV-enabled MEC networks.

  • Takeaways & Limitations

    The paper provides guidance for future research directions while characterizing UAV-enabled MEC research as being in its early stage.

Abstract

from arXiv · show

Unmanned aerial vehicle (UAV)-enabled communication networks are promising in the fifth and beyond wireless communication systems. In this paper, we shed light on three UAV-enabled mobile edge computing (MEC) architectures. Those architectures have been receiving ever increasing research attention for improving computation performance and decreasing execution latency by integrating UAV into MEC networks. We present a comprehensive survey for the state-of-the-art research in this domain. Important implementation issues are clarified. Moreover, in order to provide an enlightening guidance for future research directions, key challenges and open issues are discussed.

I. INTRODUCTION

Computation-intensive mobile applications strain devices with limited computational capability and battery capacity, while MEC enables task offloading to nearby servers. UAV-enabled MEC extends these benefits through flexible deployment, short-distance LoS links, and optimized trajectories, but the research area remains early-stage and requires systematic synthesis.

  • Computation-intensive applications challenge mobile devices because they typically have limited computational capability and finite battery capacity.
  • MEC addresses this challenge by allowing users to offload partial or complete computation-intensive tasks to servers deployed near end users.This deployment improves computation performance in a cost-effective and energy-saving manner.
  • UAVs can function as task-executing users, relays assisting computation offloading, or MEC servers within UAV-enabled MEC architectures.
  • UAV-enabled MEC networks can be flexibly deployed in wilderness, deserts, and complex terrains where terrestrial MEC may be inconvenient or unreliable to establish.They are also useful when conventional terrestrial MEC systems are destroyed by natural disasters.
  • Short-distance LoS links and UAV trajectory optimization can improve computation offloading, result downloading, and user computation performance.
  • Research and development in UAV-enabled MEC networks remain at an early stage, with limited investigations and unresolved technical challenges.The article responds with a comprehensive survey of research efforts, implementation issues, challenges, and open research questions.

II. STATE OF THE ART

The state-of-the-art discussion covers potential application scenarios and three UAV-enabled MEC architectures, including support for hotspots and environments where terrestrial MEC is impractical. UAV assistance can reduce outage probability and improve user QoE in computation-intensive settings.

  • Overview: The state-of-the-art review introduces potential application scenarios and three UAV-enabled MEC architectures before surveying ongoing research efforts.
  • Hotspots: In hotspots with massive demand for computation-intensive services, UAV-enabled MEC can assist terrestrial MEC systems with high-volume computation tasks.The cited example describes more than one million people simultaneously using navigation applications in Beijing during rush hour.
  • Hotspots: UAV assistance in hotspots can greatly decrease outage probability and improve user QoE.
  • Remote and complex terrains: In wildernesses, deserts, and complex terrains, terrestrial MEC may be difficult or unaffordable to establish, making UAV-enabled MEC a desirable alternative.

3) Battle fields:

UAV-enabled MEC research addresses computation needs where terrestrial MEC is unreliable and organizes deployments into three architectures based on the UAV’s role. Studies examine resource allocation, trajectories, access schemes, and offloading decisions, but the research remains early-stage.

  • 3) Battle fields:: Battlefield and disaster scenarios create computational demands where reliable terrestrial MEC systems may be unavailable or destroyed.UAV-enabled MEC networks are described as useful for real-time battlefield estimation and rescue or reconstruction tasks.
  • Architectures: The three architectures treat the UAV as a computation-task user, an MEC server, or a relay for users’ offloading.Their selection depends on the UAV’s computation and communication capabilities and on the scale of computation bits.
  • Recent advances: Sequential-game strategies for the first architecture demonstrated Nash equilibrium and a tradeoff between energy consumption and delay.Other work jointly optimized UAV trajectory and offloading time to minimize mission completion time.
  • Challenges and deployment: Existing studies remain constrained by assumptions such as hovering UAVs, unoptimized trajectories, omitted hover energy, and limited computation capability.The three architectures can also coexist and cooperate, with selection depending on the UAV’s function and computation scale.
  • Recent advances: In the UAV-as-MEC-server architecture, studies optimized offloading, downloading, local computing, energy, trajectories, and multiple-access schemes.Reported results favored NOMA over OFDMA and partial over binary computation offloading, while trajectory optimization improved computation performance.
  • Recent advances: For relay-based offloading, optimizing the UAV trajectory and using cyclical multiple access significantly improved computation performance.The relay architecture supports users when the direct offloading link to the MEC server is poor or terrestrial MEC is unavailable.

III. TECHNICAL DETAILS FOR IMPLEMENTATION

Implementation issues in UAV-enabled MEC networks are highlighted as a distinct part of the paper. The overview is presented in Fig. 4.

  • III. TECHNICAL DETAILS FOR IMPLEMENTATION: Key implementation elements in UAV-enabled MEC networks are highlighted in Fig. 4.The passage identifies Fig. 4 as the overview location for these elements.
  • III. TECHNICAL DETAILS FOR IMPLEMENTATION: Fig. 4 provides an overview of the implementation elements discussed for UAV-enabled MEC networks.The supplied passage does not specify the individual elements.
  • III. TECHNICAL DETAILS FOR IMPLEMENTATION: Implementation considerations are treated as part of the technical detail for UAV-enabled MEC networks.The passage introduces this treatment without enumerating specific mechanisms or constraints.

A. Operation Modes

UAV-enabled MEC networks support partial and binary computation offloading, with distinct flexibility and complexity trade-offs. Partial offloading can improve computation performance but requires more complex circuitry and protocols.

  • A. Operation Modes: Partial offloading partitions a task between MEC execution and local computation.Examples include locally processing low-computation face-recognition tasks while offloading higher-computation recognition tasks.
  • A. Operation Modes: Binary computation executes each task wholly either locally or through complete offloading.For channel-state-information estimation, correlated raw data samples must be computed together.
  • A. Operation Modes: Under partial offloading, UAVs can allocate communication and computation resources dynamically based on channel-state information.The UAV can simultaneously offload tasks and perform local computation.
  • A. Operation Modes: Binary computation limits resource-allocation flexibility because offloading and local computing cannot occur simultaneously.This constraint follows from executing each task as a whole under the binary mode.
  • A. Operation Modes: Partial offloading can achieve better computation performance than binary computation, but requires more complex circuitry and protocols.The preferred mode depends on UAV structure and computation-task characteristics.

B. UAV Computing Techniques

UAV-enabled MEC computing techniques combine local-computing choices with communication strategies for task offloading. Dynamic frequency scaling and advanced multiple-access schemes can improve performance while introducing complexity or design considerations.

  • B. UAV Computing Techniques: Dynamic voltage and frequency scaling can outperform constant-rate local computing in energy consumption, throughput, and latency.It adjusts CPU frequency according to computation-task scale, whereas fixed-frequency circuits compute at a constant rate.
  • B. UAV Computing Techniques: Terrestrial MEC communication techniques can also be applied to UAV-enabled MEC, with UAV-specific functions available to improve conventional offloading.The survey distinguishes OMA techniques such as OFDMA and TDMA from NOMA approaches.
  • B. UAV Computing Techniques: OAM-based cyclical multiple access schedules users according to UAV proximity and channel conditions to improve offloading efficiency.The reported offloading throughput with CMA is larger than with TDMA.
  • B. UAV Computing Techniques: NOMA can provide higher computation performance gain than OMA when serving multiple users for uplink offloading and downlink result transmission.NOMA allows multiple UAVs or users to share the same physical resource and can improve connectivity and spectral efficiency.

2) Multiple antennas techniques:

UAV-enabled MEC design must account for antenna, duplexing, security, and resource-allocation choices. These choices balance offloading efficiency, secure communication, computation performance, energy use, and operational cost.

  • 2) Multiple antennas techniques:: Multiple antennas improve offloading efficiency through spatial-diversity gain, while UAV flying direction affects beamforming design.Beamforming must account for trajectory effects unlike conventional terrestrial MEC designs.
  • 2) Multiple antennas techniques:: Full duplex enables simultaneous task offloading and result downloading, improving offloading efficiency over half duplex.The benefit comes with more complex protocol and circuit design, and requires effective interference control.
  • 2) Multiple antennas techniques:: UAV-enabled MEC remains vulnerable to eavesdropping because line-of-sight links can expose confidential tasks and results.Encryption can add computation overhead that conflicts with latency and power concerns, motivating physical-layer security techniques.
  • 2) Multiple antennas techniques:: Physical-layer security leverages wireless-channel characteristics to realize secure communications in UAV-enabled MEC networks.Trajectory optimization can keep the UAV closer to legitimate users and farther from eavesdroppers.
  • 2) Multiple antennas techniques:: Resource allocation jointly considers offloading, local computing, and UAV flight resources to optimize performance and economical operation.Optimizable variables include bandwidth, power, time, CPU frequency, trajectory, speed, and acceleration.
  • 2) Multiple antennas techniques:: Resource-allocation objectives include computation-bit maximization, energy minimization, efficiency maximization, cost minimization, completion-time minimization, and fairness.Trajectory design can trade computation performance against UAV operating cost.

1) Computation bits maximization:

Computation-bit maximization and energy minimization emphasize different objectives, while computation-efficiency maximization targets their trade-off. UAV-enabled MEC optimization must account for energy consumed by computing, offloading, and flight.

  • 1) Computation bits maximization:: Computation-bit maximization seeks to maximize total computation bits produced through offloading and local computing.When the UAV is a user or MEC server, offloading, local computing, and flight should be jointly optimized; as a relay, offloading and flight are emphasized.
  • 1) Computation bits maximization:: Energy consumption spans local computing, task offloading, and UAV flight.It depends on CPU frequency and computation time, transmit power and offloading time, and flight weight, speed, acceleration, and duration.
  • 1) Computation bits maximization:: Resource-allocation schemes have been designed to minimize total energy consumed across these three processes.Two different flight-energy models have been proposed in prior work.
  • 1) Computation bits maximization:: Optimizing only computation bits or energy consumption cannot achieve a good trade-off between the two metrics.Computation-efficiency maximization instead maximizes computation bits per Joule.
  • 1) Computation bits maximization:: Computation efficiency concerns the computation process, whereas energy efficiency concerns information transmission.The paper explicitly treats them as different definitions.

4) Cost minimization:

The section identifies optimization and deployment challenges in UAV-enabled MEC networks, spanning cost, completion time, resource allocation, beamforming, security, computation efficiency, cooperation, and machine learning.

  • Cost minimization: Cost minimization balances energy overhead and execution latency in UAV-enabled MEC networks.
  • Completion time minimization: Completion-time objectives differ between partial and binary offloading while satisfying minimum computation-bit requirements.
  • Resource allocation: Multiple-user multiple-UAV resource allocation is challenging because prior schemes mainly considered one UAV or one user, while UAV operation time and batteries are limited.
  • Operation selection: Binary local-computation or offloading decisions create mixed integer non-convex operation-selection problems, especially with multiple users.
  • Beamforming and security: Dynamic UAV directions complicate beamforming design, while security remains insufficiently studied and physical-layer security is identified as a promising direction.
  • Computation efficiency: Limited UAV battery energy makes computation efficiency vital, but binary computation introduces especially difficult intractable fractional forms.
  • UAV–terrestrial cooperation: Future work includes cooperation with terrestrial MEC networks, including joint resource allocation and interference coordination on shared frequency bands.
  • Machine learning: Machine learning may support intelligent UAV control and computation performance, but multi-agent reinforcement learning for noncooperating UAVs remains challenging.

V. CONCLUSIONS

UAV-enabled MEC networks can improve computation performance and reduce execution latency. The article surveys recent advances, clarifies implementation issues, and discusses challenges and open issues while noting that the field remains at an early stage.

  • UAV-enabled MEC networks are promising for improving computation performance and reducing execution latency.
  • The article presents a comprehensive survey of recent advances in UAV-enabled MEC networks.
  • Implementation issues are highlighted to facilitate applications of UAV-enabled MEC networks.
  • Key challenges and open issues are discussed to guide future research directions.
  • The research area is still at an early stage and requires extensive efforts to reach maturity.

VI. BIOGRAPHIES

The biographies identify the paper’s authors and describe their academic affiliations, positions, and research interests in wireless communications, networking, signal processing, and related areas.

  • Fuhui Zhou received his Ph.D. from Xidian University in 2016.
  • Yuhao Wang is a professor at Nanchang University and leads academic and research organizations there.
  • Rose Qingyang Hu is a professor and associate dean for research at Utah State University and directs its Communications Network Innovation Lab.
  • The authors’ listed research interests span mobile edge computing, machine learning, physical-layer security, resource allocation, sensing, and signal processing.
  • Zan Li is a professor at Xidian University whose research interests include wireless communications and signal processing.
Loading 1910.10523v1…