Advancements in Mobile Edge Computing and Open RAN: Leveraging Artificial Intelligence and Machine Learning for Wireless Systems

Fatemeh Afghah; Ryan Barker; Tolunay Seyfi

REVIEW 4 major objections 5 minor 11 references

Advancements in Mobile Edge Computing and Open RAN: Leveraging Artificial Intelligence and Machine Learning for Wireless Systems

T0 review · 4 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This review paper argues that deep reinforcement learning and related AI/ML methods deliver concrete, quantifiable gains in mobile edge computing and Open RAN, reporting improvements such as 23% lower energy use and 21% fewer task failures.

desk verdict A transparent, clearly-labeled reading report with no new results and several mismatched citations that undermine its only quantitative evidence. read the letter →

arxiv 2502.02886 v3 pith:6ZTLGPVM submitted 2025-02-05 cs.NI

Ryan Barker , Tolunay Seyfi , Fatemeh Afghah This is my paper · ORCID

classification cs.NI

keywords mobileedgecomputingOpenRANdeepreinforcementlearningcomputationoffloadingxAppsnetworkslicingself-healingnetworksnext-generationwireless

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This is a review paper that argues AI and machine learning, especially deep reinforcement learning (DRL), are the key enablers for mobile edge computing (MEC) and Open RAN (ORAN) in next-generation wireless networks. It surveys the literature and highlights concrete performance numbers: DROO, a DRL-based offloading algorithm, cuts energy consumption by 23% versus heuristics, and DRL-based xApps in the ColO-RAN testbed reduce task failures by 21%. The paper frames these gains as evidence that closed-loop, AI-driven control can make networks more adaptive, energy-efficient, and self-healing. A sympathetic reader would take away that DRL has moved from theory to validated testbed results in both domains.

What carries the argument

The mechanism carrying the argument is the closed-loop control loop enabled by the near-real-time RAN Intelligent Controller (RIC) and its xApps in ORAN, paired with the DRL-based offloading policy in MEC. The DROO algorithm uses deep reinforcement learning to map observed network states to offloading decisions without an explicit model of the system, letting it adapt to changing user demand and channel conditions. In ORAN, xApps hosting DDQN, SVM, k-NN, and decision-tree models run inside the near-real-time RIC to steer traffic, allocate resources, predict interference, and preempt congestion. These mechanisms are what turn network optimization into a learnable, adaptive decision problem, and they are what the paper's reported performance gains are attributed to.

What would settle it

Run DROO in a live, wireless-powered MEC testbed under the same conditions as the original study, measuring energy consumption and task failure rate against the same heuristic and MDP baselines; if the 23% energy reduction and sub-5% failure rate do not reproduce, the paper's central quantitative claim fails. Similarly, deploy the DDQN traffic-steering xApp on an ORAN testbed and compare load-balancing metrics against a static baseline to test the 17% figure.

Watch

Extended reading notes

Core claim

The central claim is that AI/ML, and deep reinforcement learning in particular, transforms MEC and ORAN from static, model-dependent systems into adaptive, learning-driven ones. In MEC, the DROO algorithm learns optimal computation-offloading decisions for wireless-powered networks, reporting a 23% reduction in energy consumption relative to heuristic approaches, a 17% reduction in average task completion time relative to an MDP-based energy-harvesting method, and a task failure rate below 5% versus 12% for heuristics. In ORAN, the ColO-RAN testbed integrates DRL-based xApps on the near-real-time RIC and achieves a 21% reduction in task failures; other ML-based xApps are reported to cut interference by 25%, improve load balancing by 17%, improve handover success by 18%, and reduce congestion events by 22%. The paper presents these numbers as evidence that closed-loop, AI-driven control is the path to low-latency, reliable, self-healing next-generation radio networks.

Load-bearing premise

The load-bearing premise is that the performance percentages quoted from the cited papers are accurate, correctly attributed, and will transfer from the specific testbeds and simulations to production networks.

Editorial extensions

If this is right

DRL-based offloading could let mobile devices in wireless-powered MEC networks make near-optimal offloading decisions without knowing the system dynamics, saving energy and reducing task failures in practice.
ORAN operators could deploy self-healing xApps that detect and mitigate congestion, interference, and handover failures in real time, improving QoS for video streaming, gaming, and VR.
The near-real-time RIC becomes the central decision-making hub, enabling closed-loop control that adapts to changing traffic and radio conditions within milliseconds.
The cited numbers provide concrete, quantifiable targets for validating AI-driven RAN management in experimental testbeds and future field trials.
Combining DRL in MEC with DRL in ORAN could lead to end-to-end optimization across the edge and radio access network, though the paper does not demonstrate this integration.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The reported gains come from specific testbeds and simulations; the paper does not establish that they survive in commercial deployments with non-stationary traffic, hardware heterogeneity, and imperfect telemetry.
An implicit implication is that MEC and ORAN are complementary halves of the same AI control problem: offloading decisions at the edge and radio resource decisions at the RAN both depend on latency, load, and channel state, so joint optimization could compound the reported gains.
A testable extension would be to benchmark DROO not just against the older heuristic and MDP baselines in the paper, but against modern offline-optimal and model-predictive offloading schemes in a live edge testbed, to see whether the 23% energy gain persists.
The paper's suggestion to add delay-tolerant and federated learning mechanisms to DROO implies that its current performance depends on continuous, low-latency feedback; how much of the 23% gain would degrade under intermittent connectivity is left unquantified.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Colleague,

This is a reading report, not a research paper. The authors say so themselves, both in the abstract and in Section II. So the right question is not whether the results are new, but whether the summary is accurate and useful. On that, I have mixed feelings.

What the paper does well: it gives a clean tour of two important areas—DROO for MEC offloading and DRL-based xApps in ORAN—and it spells out real limitations (delayed feedback, exploration-exploitation, fault tolerance) that a newcomer would want to see. The structure is sensible, and the choice of key papers (Huang et al., ColO-RAN, the traffic-steering work) is defensible.

The soft spot is real and load-bearing: the quantified performance claims are attached to the wrong references in several places. In Section III.B, the 17% task-completion-time improvement attributes the MDP energy-harvesting baseline to [1][3], but [3] is the fog/IoT paper; the actual MDP paper is [4]. The <5% vs 12% task-failure comparison cites [1][2], and [2] is not a heuristic offloading baseline. In Section IV.B, the traffic-steering load-balancing figure is cited to [7] instead of [10], the ColO-RAN task-failure number to [9] instead of [8], and the congestion figure to [8] instead of [9]. Because these percentages are the only quantitative evidence for the paper's central "AI/ML helps" claim, the mismatch means a reader cannot verify the claim from the cited sources without going to the original papers and checking. That is more than a typo; it is a reproducibility problem in a review.

I do not think this is dishonest. The authors seem to have summarized from memory and made attribution errors. But it does mean the report's evidence is hearsay until fixed.

Who is this for? A newcomer to MEC/ORAN who wants a quick map of the literature might get some orientation. A researcher would not cite it for any number. It does not deserve peer review as a research contribution. If it is meant as a survey for a venue like IEEE Communications Surveys & Tutorials, it would need a full rewrite with accurate citations and more systematic coverage.

Recommendation: desk reject for a research track; if it comes to you as a survey, send it back for major revision on citations before any reviewer sees it.

Referee Report

4 major / 5 minor

Summary. This paper is a literature survey (self-described as a 'reading report') reviewing recent work on AI/ML, particularly deep reinforcement learning (DRL), for computation offloading in mobile edge computing (MEC) and for closed-loop control via xApps in Open RAN (ORAN). It describes the DROO algorithm, reports quantitative gains over heuristic and MDP-based baselines, surveys ORAN architecture, and summarizes performance improvements from several xApps for traffic steering, network slicing, interference management, handover, and congestion avoidance. The paper contains no new experiments, derivations, or original data; its contribution is a synthesis of selected prior work and a discussion of limitations and future directions.

Significance. If accurate, this survey would serve as a compact entry point to DRL for MEC and ORAN, particularly for its explicit enumeration of DROO's limitations (feedback dependence, exploration-exploitation trade-offs, fault tolerance) and its organization around self-healing ORAN use cases. The paper's evidentiary value rests entirely on faithfully reporting cited results, and the citation mismatches detailed below currently prevent verification of the central quantitative claims. The paper makes specific, falsifiable numeric claims (e.g., 23% energy reduction, 21% task-failure reduction) that can be checked against the primary literature, but those checks fail without correct references. The paper provides no reproducible code or datasets; its strength is the breadth of topics covered rather than technical depth.

major comments (4)

[III.A and III.B] References [3] and [4] are swapped. The MDP-based energy-harvesting dynamic computation offloading work is [4] (Mao, Zhang, and Letaief), but the text attributes it to [3] (Chiang and Zhang, 'Fog and IoT'). Conversely, the Fog and IoT overview is [3], but it appears as [4] in the paragraph on fog computing. The 17% task-completion-time improvement claim in Section III.B cites [1][3]; it should cite [1][4] (or [1] alone if the comparison is internal to the DROO paper). As written, a reader cannot verify the comparison.
[III.B] The claim that 'DROO exhibited a lower task failure rate, averaging less than 5% compared to the 12% failure rate seen with heuristic-based methods [1][2]' cites [2], which is 'Wireless Powered Communication: Opportunities and Challenges,' a paper that does not present a heuristic offloading baseline. The baseline is presumably reported in [1] (the DROO paper) or needs a different supporting reference. This citation error directly affects the reliability of the stated task-failure comparison.
[IV.B] The 17% load-balancing improvement from the DDQN-based traffic-steering xApp is attributed to reference [7] (the X5g testbed paper), but the actual traffic-steering work is reference [10] (Lacava et al., 'Programmable and Customized Intelligence for Traffic Steering in 5G Networks Using Open RAN Architectures'). The entire traffic-steering paragraph is meant to summarize [10], so this mismatch is load-bearing for the self-healing section's quantitative evidence.
[IV.B] The 21% task-failure reduction from DRL-based xApps in ColO-RAN is attributed to reference [9] (O-RAN Performance Analyzer), but the described ColO-RAN experimentation is reference [8] (Bonati et al.). Conversely, the 22% congestion-event reduction is attributed to [8], but the O-RAN Performance Analyzer is [9]. These two citations are cross-swapped, undermining the verifiability of two separate quantitative claims in the same paragraph.

minor comments (5)

[Abstract and I] The manuscript self-identifies as a 'reading report,' which is informal for a journal. Please reframe the paper as a survey/tutorial and define the scope and selection criteria for included works.
[References] The reference list format is inconsistent: some entries include URLs and arXiv identifiers ([7], [9], [10]) while others provide only bibliographic data. Standardize all entries to a single citation style.
[IV.B] The text states that NVIDIA ARC has validated AI-based mechanisms but provides no citation for this claim; either add a supporting reference or remove the specific mention.
[Throughout] The naming of ColO-RAN is inconsistent (ColO-RAN, Colo-RAN, COLO-RAN) and 'xApp' is not defined at first use. Please standardize terminology and define acronyms.
[III.B] The phrase 'the heuristic approaches presented in the paper "Wireless Powered Communication: Opportunities and Challenges"' conflates the heuristic baselines of [1] with a paper that does not discuss heuristic offloading; rephrase to distinguish the cited references from the baseline methods of [1].

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the paper is a literature survey whose quantitative claims are attributed to external sources, not derived from its own fitted inputs or self-citations.

full rationale

This manuscript is a reading report that summarizes prior published work on MEC and ORAN. It does not introduce new equations, fit parameters, or claim a first-principles derivation. The energy, latency, and failure-rate figures for DROO are presented as results reported in the cited literature (e.g., Huang, Bi, and Zhang), and the ORAN xApp figures are attributed to external testbed and algorithm papers. The analysis identifies limitations and suggests future directions without deriving any output from the paper's own assumptions. No step in the paper defines a quantity in terms of the very quantity it claims to predict, and no load-bearing conclusion is justified solely by a self-citation. The citation mismatches noted elsewhere (e.g., a 17% load-balancing figure attached to reference [7] rather than [10]) are factual accuracy concerns about source attribution, not evidence of circular reasoning. Accordingly, the appropriate circularity score is 0.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The report makes no original derivations, so the ledger mainly records the secondary-source assumptions it inherits. There are no fitted parameters or invented entities. The key unstated premise is that all quoted performance numbers from cited papers are accurate and transferable.

assumptions (3)

domain assumption The performance improvements reported in the cited papers are accurately measured and reproducible in the contexts described.
The whole report's conclusions depend on accepting secondary-source numbers, e.g., Section III.B 23% energy reduction and Section IV.B 17% load balancing, without independent verification.
domain assumption ORAN architecture components (Near-RT RIC, xApps, O-CU, O-DU) function as described in the cited O-RAN standards literature.
Section IV.A assumes the standard architecture and interfaces, citing references 5 and 6, as background.
domain assumption The comparison methods (heuristic, MDP, fog) used in the underlying studies are representative baselines.
Section III.A compares DROO to heuristic and MDP-based approaches without defining the baseline algorithms or conditions.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Advancements in Mobile Edge Computing and Open RAN: Leveraging Artificial Intelligence and Machine Learning for Wireless Systems." pith.science (2026). https://pith.science/paper/6ZTLGPVM

@misc{pith2026250202886,
  author       = {Pith},
  title        = {Pith review of: Advancements in Mobile Edge Computing and Open RAN: Leveraging Artificial Intelligence and Machine Learning for Wireless Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6ZTLGPVM}},
  note         = {Machine review of arXiv:2502.02886}
}

read the original abstract

Mobile Edge Computing (MEC) and Open Radio Access Networks (ORAN) are transformative technologies in the development of next-generation wireless communication systems. MEC pushes computational resources closer to end-users, enabling low latency and efficient processing, while ORAN promotes interoperability and openness in radio networks, thereby fostering innovation. This paper explores recent advancements in these two domains, with a particular focus on how Artificial Intelligence (AI) and Machine Learning (ML) techniques are being utilized to solve complex wireless challenges. In MEC, Deep Reinforcement Learning (DRL) is leveraged for optimizing computation offloading, ensuring energy-efficient solutions, and meeting Quality of Service (QoS) requirements. In ORAN, AI/ML is used to develop intelligent xApps for network slicing, scheduling, and online training to enhance network adaptability. This reading report provides an in-depth analysis of multiple key papers, discusses the methodologies employed, and highlights the impact of these technologies in improving network efficiency and scalability.

Figures

Figures reproduced from arXiv: 2502.02886 by the authors.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

11 extracted references · 9 canonical work pages

[1]

Huang, S

L. Huang, S. Bi, and Y.-J. A. Zhang, ``Deep reinforcement learning for online computation offloading in wireless powered mobile-edge computing networks,'' IEEE Transactions on Mobile Computing, vol. 19, no. 11, pp. 2581--2593, 2020

work page 2020
[3]

Chiang and T

M. Chiang and T. Zhang, ``Fog and iot: An overview of research opportunities,'' IEEE Internet of Things Journal, vol. 3, no. 6, pp. 854--864, 2016

work page 2016
[4]

Y. Mao, J. Zhang, and K. B. Letaief, ``Dynamic computation offloading for mobile-edge computing with energy harvesting devices,'' IEEE Journal on Selected Areas in Communications, vol. 34, no. 12, pp. 3590--3605, 2016

work page 2016
[2]

S. Bi, C. K. Ho, and R. Zhang, ``Wireless powered communication: opportunities and challenges,'' IEEE Communications Magazine, vol. 53, no. 4, pp. 117--125, 2015

work page 2015
[7]

Villa, I

D. Villa, I. Khan, F. Kaltenberger, N. Hedberg, R. S. da Silva, S. Maxenti, L. Bonati, A. Kelkar, C. Dick, E. Baena, J. M. Jornet, T. Melodia, M. Polese, and D. Koutsonikolas, ``X5g: An open, programmable, multi-vendor, end-to-end, private 5g o-ran testbed with nvidia arc and openairinterface,'' 2024. [Online]. Available: https://arxiv.org/abs/2406.15935

arXiv 2024
[10]

Lacava, M

A. Lacava, M. Polese, R. Sivaraj, R. Soundrarajan, B. S. Bhati, T. Singh, T. Zugno, F. Cuomo, and T. Melodia, ``Programmable and customized intelligence for traffic steering in 5g networks using open ran architectures,'' 2022. [Online]. Available: https://arxiv.org/abs/2209.14171

arXiv 2022
[9]

Kouchaki, S

M. Kouchaki, S. H. Natanzi, M. Zhang, B. Tang, and V. Marojevic, ``O-ran performance analyzer: Platform design, development, and deployment,'' IEEE Communications Magazine, 2024. [Online]. Available: https://par.nsf.gov/biblio/10544903

work page arXiv 2024
[8]

Bonati, S

L. Bonati, S. D’Oro, S. Basagni, and T. Melodia, ``Coloran: Developing machine learning-based xapps for open ran closed-loop control on programmable experimental platforms,'' IEEE Transactions on Network and Service Management, vol. 18, no. 4, pp. 4625--4639, Dec 2021

work page 2021

Show all 11 references

[5]

Polese, L

M. Polese, L. Bonati, S. D’Oro, S. Basagni, and T. Melodia, ``Understanding o-ran: Architecture, interfaces, algorithms, security, and research challenges,'' IEEE Communications Surveys & Tutorials, vol. 25, no. 2, pp. 1376--1411, 2023

2023
[6]

Lacava, M

A. Lacava, M. Bordin, M. Polese, R. Sivaraj, T. Zugno, F. Cuomo, and T. Melodia, ``ns-o-ran: Simulating o-ran 5g systems in ns-3,'' in Proceedings of the 2023 Workshop on ns-3, ser. WNS3 2023. 1em plus 0.5em minus 0.4em ACM, Jun. 2023, p. 35–44. [Online]. Available: http://dx....

2023
[11]

adobe:ns:meta/

11em plus .33em minus .07em 4000 4000 100 4000 4000 500 `\.=1000 = #1 \@IEEEnotcompsoconly \@IEEEcompsoconly #1 * [1] 0pt [0pt][0pt] #1 * [1] 0pt [0pt][0pt] #1 * \| ** #1 \@IEEEauthorblockNstyle \@IEEEcompsocnotconfonly \@IEEEauthorblockAstyle \@IEEEcompsocnotconfonly \@IEEEco...

1999

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.

[1] [1]

Huang, S

L. Huang, S. Bi, and Y.-J. A. Zhang, ``Deep reinforcement learning for online computation offloading in wireless powered mobile-edge computing networks,'' IEEE Transactions on Mobile Computing, vol. 19, no. 11, pp. 2581--2593, 2020

work page 2020

[2] [3]

Chiang and T

M. Chiang and T. Zhang, ``Fog and iot: An overview of research opportunities,'' IEEE Internet of Things Journal, vol. 3, no. 6, pp. 854--864, 2016

work page 2016

[3] [4]

Y. Mao, J. Zhang, and K. B. Letaief, ``Dynamic computation offloading for mobile-edge computing with energy harvesting devices,'' IEEE Journal on Selected Areas in Communications, vol. 34, no. 12, pp. 3590--3605, 2016

work page 2016

[4] [2]

S. Bi, C. K. Ho, and R. Zhang, ``Wireless powered communication: opportunities and challenges,'' IEEE Communications Magazine, vol. 53, no. 4, pp. 117--125, 2015

work page 2015

[5] [7]

Villa, I

D. Villa, I. Khan, F. Kaltenberger, N. Hedberg, R. S. da Silva, S. Maxenti, L. Bonati, A. Kelkar, C. Dick, E. Baena, J. M. Jornet, T. Melodia, M. Polese, and D. Koutsonikolas, ``X5g: An open, programmable, multi-vendor, end-to-end, private 5g o-ran testbed with nvidia arc and openairinterface,'' 2024. [Online]. Available: https://arxiv.org/abs/2406.15935

arXiv 2024

[6] [10]

Lacava, M

A. Lacava, M. Polese, R. Sivaraj, R. Soundrarajan, B. S. Bhati, T. Singh, T. Zugno, F. Cuomo, and T. Melodia, ``Programmable and customized intelligence for traffic steering in 5g networks using open ran architectures,'' 2022. [Online]. Available: https://arxiv.org/abs/2209.14171

arXiv 2022

[7] [9]

Kouchaki, S

M. Kouchaki, S. H. Natanzi, M. Zhang, B. Tang, and V. Marojevic, ``O-ran performance analyzer: Platform design, development, and deployment,'' IEEE Communications Magazine, 2024. [Online]. Available: https://par.nsf.gov/biblio/10544903

work page arXiv 2024

[8] [8]

Bonati, S

L. Bonati, S. D’Oro, S. Basagni, and T. Melodia, ``Coloran: Developing machine learning-based xapps for open ran closed-loop control on programmable experimental platforms,'' IEEE Transactions on Network and Service Management, vol. 18, no. 4, pp. 4625--4639, Dec 2021

work page 2021