Pith. sign in

REVIEW 4 major objections 4 minor 19 references

QoS-Aware Proportional Fairness Scheduling for Multi-Flow 5G UEs: A Smart Factory Perspective

T0 review · 4 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read A per-flow 5G scheduler keeps control-loop deadline misses under 2 percent while staying fair to video and telemetry traffic.

desk verdict A useful open-source Simu5G extension for multi-QFI scheduling, but the evaluation lacks the per-flow PF baseline needed to attribute gains to QoS-awareness. read the letter →

arxiv 2508.21783 v1 pith:QEMJ2I4Q submitted 2025-08-29 cs.NI

Mohamed Seliem , Utz Roedig , Cormac Sreenan , Dirk Pesch This is my paper · ORCID
classification cs.NI
keywords 5GQoSFlowIdentifierproportionalfairnesssmartfactoryURLLCdeadlinecomplianceSimu5Gresourceallocation
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

The paper tries to establish that one private 5G base station can serve the mixed traffic of a smart factory—tight-deadline robot control, live video, and background telemetry—without starving any flow. To test this, the authors extend an existing 5G network simulator so that each user device can carry several QoS flows at once, each tagged with its QoS Flow Identifier, and they add a MAC-layer scheduler that scores every flow by a weighted mix of delay urgency, guaranteed-bit-rate shortfall, and priority, divided by that flow's recent throughput. In a simulated six-device factory cell, the scheduler keeps deadline violations for the most time-critical class below 2 percent, maintains flow-level fairness above 0.9, and satisfies guaranteed bit rates above 95 percent, while comparison schedulers either miss deadlines or starve low-priority flows. If these numbers hold, the contribution is a practical recipe for testing and provisioning private 5G for industrial workloads, plus a scheduler design that could generalize beyond factories.

What carries the argument

The central object is the QoS-flow-aware proportional-fairness score: M_i(t) = U_i(t)/Rbar_i(t), a utility-per-recent-throughput ratio computed for every active QFI. The numerator U_i(t) is a tunable weighted sum of delay urgency, guaranteed-bit-rate shortfall, and priority, while the denominator is the flow's averaged throughput; this combination gives time-critical flows a boost when they fall behind while still letting well-served flows yield resources. The paper implements this score inside the simulator's MAC scheduler, supported by newly added SDAP-layer tagging, per-QFI PDCP/RLC instances, and a centralized QFI context manager that feeds flow state into each scheduling decision.

What would settle it

Run the identical traffic mix with Max C/I and Static Priority modified to operate per QFI with the same QFI tagging and context; if their QFI-1 violation rates fall to the same roughly 2 percent level, the utility-weighting mechanism is not what produces the headline result.

Watch

Extended reading notes

Core claim

The central claim is that per-flow QFI awareness can be combined with proportional fairness to satisfy heterogeneous industrial QoS without sacrificing throughput or fairness. The scheduler's metric is M_i(t) = U_i(t)/Rbar_i(t), where U_i(t) = alpha_i*D_i(t) + beta_i*G_i(t) + gamma_i*P_i(t); D_i is delay urgency, G_i is GBR fulfillment, P_i is the flow's priority scalar, and Rbar_i is the flow's exponential moving average throughput. At each transmission time interval the scheduler ranks active QFIs by this score and allocates resource blocks in that order. The paper reports that under high contention this keeps deadline violations for the strictest URLLC flow below 2 percent, compared with

Load-bearing premise

The comparison assumes the baseline schedulers are applied at the same per-flow granularity as QoS-PF; if they schedule per device instead, the reported deadline gains could come from finer resource granularity rather than from QoS awareness.

Editorial extensions

If this is right

  • A single private 5G cell can carry URLLC control loops, eMBB video, and best-effort telemetry without hard slicing, as long as flows carry QFI tags and are scheduled by their utility scores.
  • Factory designers can use the open simulator extensions to check whether a proposed traffic mix meets its delay budgets before deployment, instead of relying on costly live trials.
  • The reported near-linear runtime growth—under 2 ms per TTI at 40 UEs with three flows each—suggests per-flow QoS scoring is cheap enough for realistic cell sizes.
  • Operators can shift the scheduler from delay-first to fairness-first behavior by retuning three weights, without changing the underlying allocation mechanism.
  • The same scheduler logic is claimed to carry over to other multi-flow verticals such as tele-surgery, connected vehicles, and AR/VR, where URLLC and eMBB traffic share one device.

Reading between the lines

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

  • A controlled ablation that runs the comparison schedulers at the same per-QFI granularity as QoS-PF would separate the contribution of the utility weighting from the contribution of finer-grained scheduling itself; the paper does not report that comparison.
  • The three utility weights are static in this paper; an obvious extension is to adapt alpha, beta, and gamma online from measured violations and fairness, turning the sensitivity analysis into a closed-loop controller.
  • Because the QFI context lives at the MAC layer, the same score could be lifted to uplink scheduling, where each device must decide how to split its transmission grants among its own flows.
  • The modular utility function can host non-linear urgency models, such as learned deadline predictors, without touching the proportional-fairness denominator that enforces long-term sharing.
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.

Referee Report

4 major / 4 minor

Summary. The paper extends the Simu5G simulator to model multiple QoS flows per UE with QFI/5QI tagging and introduces a QoS-aware proportional fairness (QoS-PF) scheduler. The scheduling metric (Eq. 1) is M_i(t) = U_i(t) / Rbar_i(t), where U_i(t) (Eq. 2) combines delay urgency, GBR deficit, and a priority scalar with tunable weights. The evaluation uses a 6-UE, three-flow smart-factory scenario (control, sensor, video) and compares QoS-PF against Max C/I and Static Priority. The authors report lower delays and deadline violations, higher GBR satisfaction, and improved Jain's fairness for QoS-PF, plus a sensitivity analysis and a scalability study. The central claim is that QoS-PF improves deadline adherence and fairness without compromising throughput.

Significance. If validated, the contribution would be useful to the industrial-5G simulation community: the open-source Simu5G extensions for per-QFI modeling and SDAP-like tagging address a real gap, and the scheduler design is plausible and modular. The paper explicitly promises reproducibility through public code and configuration files. However, the current evaluation does not isolate the QoS-aware component of the scheduler. Because Eq. (2) reduces to standard proportional fairness when U_i is constant, the absence of a per-QFI non-QoS PF baseline means the reported gains could come from the PF mechanism rather than from the delay/GBR/priority utility. In addition, several internal inconsistencies in QFI labels, baseline granularity, and the channel model prevent the reported numbers from supporting the abstract's claim as written. The core ideas are worth pursuing, but the evidence needs substantial strengthening.

major comments (4)
  1. [§III.C and §V] The central attribution of the gains to QoS-awareness is not identified. Eq. (2) defines U_i(t) = α_i D_i(t) + β_i G_i(t) + γ_i P_i(t), and Eq. (1) is M_i = U_i / Rbar_i. When U_i is constant, this is exactly standard proportional fairness. The evaluation in §IV.B compares QoS-PF only against Max C/I and Static Priority; no per-QFI non-QoS PF baseline is included. Since PF itself favors low-throughput flows, the reported differences (e.g., <2% vs >20% violation in Fig. 5, fairness >0.9 in Fig. 6) may be due to the PF denominator rather than to the delay/GBR/priority utility. Add a standard per-QFI PF baseline (U_i = 1) at the same scheduling granularity, and ideally also a per-UE PF baseline, to separate the effects of granularity and QoS-awareness. This affects all performance evidence in Figs. 4–6 and Table III.
  2. [§IV.B and §V.A] The baseline configuration is under-specified, and one assumption contradicts the results. Section IV.B says Max C/I assigns resources to 'the flow or UE' without stating which. If the stock Simu5G baselines operate per-UE while QoS-PF operates per-QFI, the comparison conflates scheduling granularity with QoS-awareness. Furthermore, §IV.A states 'All UEs operate under the same channel conditions with fixed LoS propagation', yet §V.A says 'Max C/I heavily favors high-SINR users' and Fig. 6 reports fairness below 0.6 for Max C/I. With identical channel conditions, Max C/I has no user-level channel differentiation and would not produce the stated fairness gap. Specify the exact objects (QFI or UE) that each baseline schedules, and reconcile the channel model with Figure 6.
  3. [§V.A and Table III] The results are not tied to the simulated traffic classes. Table II defines the scenario as QFI 1/2/3 with 5QI 85/6/9, but §V.A begins 'Figure 4 shows ... QFIs 5, 7, and 8' and later refers to 'QFIs 7 and 8'. Table III lists only QFI 5 and QFI 8, and reports GBR satisfaction only for QFI 5—not for QFI 1, which is the control flow at the center of the deadline claims. The reader cannot map the reported delays, violations, and throughputs to the described applications. Renumber all tables and figure captions to match the scenario QFIs, and report per-QFI metrics for all three flows.
  4. [§V.B] The sensitivity analysis in Fig. 7 reports percentage changes (18% average-latency reduction for the Delay-Tuned configuration, 9% fairness improvement for Fairness-Tuned) without confidence intervals or a statistical test. Since §IV.A claims 20 Monte Carlo runs, error bars should be feasible. This issue is secondary to the main comparison, but as presented the trade-off between Delay-Tuned and Fairness-Tuned is not statistically supported.
minor comments (4)
  1. [§II] The text says a robotic arm 'may transmit sub-millisecond motion control commands (QFI 1) alongside periodic telemetry updates (QFI 1)'. Both flows should not share the same QFI if they have different QoS profiles; the telemetry flow should use a distinct QFI (e.g., QFI 5).
  2. [§V.A and Fig. 5] The text reports 'over 20%' deadline violation for Max C/I, but the y-axis of Fig. 5 ends at 14%. The figure and text need to be reconciled, or the bar is truncated.
  3. [§IV.C, Eq. (3)] The Jain's fairness index formula has a formatting problem: J = (∑x_i)^2 / (n · ∑x_i^2) is the intended expression; the current rendering is ambiguous.
  4. [§V.C and Fig. 8] The scalability claim says per-TTI runtime 'remained under 2 ms' for 40 UEs and is 'well within real-time scheduling budgets'. A 5G TTI is typically 1 ms, so a 2 ms scheduling time would not fit in a real-time budget. Clarify whether the reported time is wall-clock simulation time or a normalized scheduling time, and compare against the relevant TTI.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the QoS-PF scheduler is an evaluated control law; its utility function and the reported KPIs are causally related but not definitionally identical.

full rationale

The paper's central claim is that a MAC scheduler with metric M_i(t)=U_i(t)/Rbar_i(t), U_i=alpha_i D_i + beta_i G_i + gamma_i P_i, improves deadline adherence, GBR satisfaction, and fairness. This is an empirical comparison against Max C/I and Static Priority, not a derivation of one equation from another. D_i (deadline proximity) is a scheduling input; the deadline violation ratio is a measured outcome. A flow near its deadline gets scheduled, so lower violations are the intended mechanism, not a fitted coincidence. The same holds for GBR fulfillment and Jain's fairness: the scheduler is designed to favor under-served flows via the PF denominator and QoS terms, and the paper measures whether this happens without throughput loss. Circularity would require, e.g., fitting alpha,beta,gamma to the reported delays and then 'predicting' those delays, which is not done; the weights are fixed and varied only in sensitivity analysis. Self-citations [8], [14], [17] are background or limitation statements; [17] notes Simu5G's missing SDAP, and the paper's contribution is the implementation itself, so no load-bearing argument reduces to a self-citation. Methodological weaknesses exist but are not circular: no per-flow PF baseline is included (Section IV.B), so the specific contribution of QoS-awareness vs. PF granularity is not isolated; the Max C/I description is ambiguous regarding flow vs. UE granularity; and Figure 4/Table III QFI labels (5,7,8) conflict with the scenario QFIs (1,2,3). These affect result attribution and reproducibility, but they do not make any stated result equivalent to its input by construction.

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

The central claim rests on unstated hyperparameters (utility weights, smoothing constant, rate caps) and on standard but unverified domain assumptions about 3GPP QoS semantics and the fidelity of Simu5G's PHY/MAC model. No new physical or system entities are introduced; the 'centralized QFI context manager' is internal software glue, not a new entity with independent evidence requirements.

free parameters (3)
  • alpha_i, beta_i, gamma_i utility weights = Balanced default: alpha=0.4, beta=0.3, gamma=0.3; Delay-Tuned: 0.7/0.2/0.1; Fairness-Tuned: 0.2/0.2/0.6
    The scheduler's utility function U_i(t)=alpha_i*D_i(t)+beta_i*G_i(t)+gamma_i*P_i(t) uses per-flow tunable coefficients. The 'Balanced' setting is selected as default without a principled optimization; the sensitivity analysis (Section V.B) shows performance depends strongly on these values, making them free parameters affecting all central results.
  • Exponential moving average smoothing constant for Rbar_i(t) = unspecified
    The average throughput Rbar_i(t) in Eq. (1) is an exponential moving average, but the paper never states the smoothing factor or averaging window. This parameter controls the fairness-responsiveness trade-off and is a free choice of the authors.
  • PRB allocation rate caps and buffer thresholds = unspecified
    Step 4 of the scheduler algorithm (Section III.C) allocates PRBs 'subject to flow-specific rate caps and buffer constraints'. These caps are not quantified, so the resource allocation behavior cannot be reproduced independently.
assumptions (3)
  • domain assumption 3GPP QoS semantics (5QI tables, QFI mapping) as defined in TS 23.501 are applied correctly in the simulator.
    The paper assumes the 3GPP QoS model is faithfully implemented in the SDAP/QFI extension (Section III.A). If the mapping or scheduling semantics deviate from 3GPP, the validity of the QoS labels and the scheduler's behavior would be affected.
  • domain assumption The Simu5G PHY/MAC model accurately represents NR radio resource allocation behavior.
    All results are simulation outputs. The paper relies on Simu5G's channel model, PRB allocation, and protocol stack as an adequate representation of real 5G NR, a premise that is not independently validated in this work (Section IV.A).
  • domain assumption The modeled traffic patterns (Table II) are representative of smart factory applications.
    The scenario uses 6 UEs with control, sensor, and video flows. The paper asserts these reflect industrial use cases, but does not validate the traffic models against measurements or standards-derived traces (Section IV.A).

how reviews work

0 comments
Cite this review

Pith. "Pith review of QoS-Aware Proportional Fairness Scheduling for Multi-Flow 5G UEs: A Smart Factory Perspective." pith.science (2026). https://pith.science/paper/QEMJ2I4Q

@misc{pith2026250821783,
  author       = {Pith},
  title        = {Pith review of: QoS-Aware Proportional Fairness Scheduling for Multi-Flow 5G UEs: A Smart Factory Perspective},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QEMJ2I4Q}},
  note         = {Machine review of arXiv:2508.21783}
}
read the original abstract

Private 5G networks are emerging as key enablers for smart factories, where a single device often handles multiple concurrent traffic flows with distinct Quality of Service (QoS) requirements. Existing simulation frameworks, however, lack the fidelity to model such multi-flow behavior at the QoS Flow Identifier (QFI) level. This paper addresses this gap by extending Simu5G to support per-QFI modeling and by introducing a novel QoS-aware Proportional Fairness (QoS-PF) scheduler. The scheduler dynamically balances delay, Guaranteed Bit Rate (GBR), and priority metrics to optimize resource allocation across heterogeneous flows. We evaluate the proposed approach in a realistic smart factory scenario featuring edge-hosted machine vision, real-time control loops, and bulk data transfer. Results show that QoS-PF improves deadline adherence and fairness without compromising throughput. All extensions are implemented in a modular and open-source manner to support future research. Our work provides both a methodological and architectural foundation for simulating and analyzing advanced QoS policies in industrial 5G deployments.

Figures

Figures reproduced from arXiv: 2508.21783 by the authors.

Figure 1
Figure 1. Smart factory architecture with heterogeneous devices and multi-QFI [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 3
Figure 3. gNB-side architecture showing the extended Simu5G NR NIC stack. [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 6
Figure 6. Jain’s fairness index across scheduling strategies. [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗
Figures from the paper (3 more)
Figure 5
Figure 5. Figure 5: Deadline violation ratio for QFIs 1 and 2 under high load. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 8
Figure 8. Figure 8: Per-TTI runtime overhead under increasing UE count. [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 7
Figure 7. Figure 7: Sensitivity of performance metrics to scheduling weight configura [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

19 extracted references · 13 canonical work pages

  1. [1]

    Lo Bello and W

    L. Lo Bello and W. Steiner, ”A Perspective on IEEE Time-Sensitive Networking for Industrial Communication and Automation Systems,” in Proceedings of the IEEE, vol. 107, no. 6, pp. 1094-1120, June 2019, doi: 10.1109/JPROC.2019.2905334

    work page arXiv 2019

  2. [2]

    R. Ali, Y . B. Zikria, A. K. Bashir, S. Garg and H. S. Kim, ”URLLC for 5G and Beyond: Requirements, Enabling Incumbent Technologies and Network Intelligence,” in IEEE Access, vol. 9, pp. 67064-67095, 2021, doi: 10.1109/ACCESS.2021.3073806

    work page arXiv 2021

  3. [3]

    Wen et al., ”Private 5G Networks: Concepts, Architectures, and Research Landscape,” in IEEE Journal of Selected Topics in Signal Processing, vol

    M. Wen et al., ”Private 5G Networks: Concepts, Architectures, and Research Landscape,” in IEEE Journal of Selected Topics in Signal Processing, vol. 16, no. 1, pp. 7-25, Jan. 2022, doi: 10.1109/JSTSP.2021.3137669

    arXiv 2022

  4. [4]

    Mahmood et al., ”Industrial IoT in 5G-and-Beyond Networks: Vision, Architecture, and Design Trends,” in IEEE Transactions on Industrial Informatics, vol

    A. Mahmood et al., ”Industrial IoT in 5G-and-Beyond Networks: Vision, Architecture, and Design Trends,” in IEEE Transactions on Industrial Informatics, vol. 18, no. 6, pp. 4122-4137, June 2022, doi: 10.1109/TII.2021.3115697

    work page arXiv 2022

  5. [5]

    3GPP TS 23.501 19.3.0, ”System architecture for the 5G System (5GS),” 3rd Generation Partnership Project (3GPP), 2025

    work page 2025

  6. [6]

    Nardini, G

    G. Nardini, G. Stea and A. Virdis, ”Scalable Real-Time Emulation of 5G Networks With Simu5G,” in IEEE Access, vol. 9, pp. 148504-148520, 2021, doi: 10.1109/ACCESS.2021.3123873

    work page arXiv 2021

  7. [7]

    Y . Xu, G. Gui, H. Gacanin and F. Adachi, ”A Survey on Re- source Allocation for 5G Heterogeneous Networks: Current Research, Future Trends, and Challenges,” in IEEE Communications Surveys & Tutorials, vol. 23, no. 2, pp. 668-695, Secondquarter 2021, doi: 10.1109/COMST.2021.3059896

    work page arXiv 2021

  8. [8]

    Seliem, D

    M. Seliem, D. Pesch, U. Roedig and C. J. Sreenan, ”Compara- tive Analysis of 5G and Wi-Fi Integration with TSN for Indus- trial Applications,” 2025 IEEE Wireless Communications and Net- working Conference (WCNC), Milan, Italy, 2025, pp. 1-6, doi: 10.1109/WCNC61545.2025.10978747

    arXiv 2025

Show all 19 references
  1. [9]

    State of Industry 5.0—Analysis and Identifica- tion of Current Research Trends

    Akundi, A.; Euresti, D.; Luna, S.; Ankobiah, W.; Lopes, A.; Edinbarough, I. State of Industry 5.0—Analysis and Identifica- tion of Current Research Trends. Appl. Syst. Innov. 2022, 5, 27. https://doi.org/10.3390/asi5010027

    2022 doi

  2. [10]

    Pedro Coelho, Catarina Bessa, Jorge Landeck, Cristov ˜ao Silva, In- dustry 5.0: The Arising of a Concept, Procedia Computer Sci- ence, V olume 217, 2023, Pages 1137-1144, ISSN 1877-0509, https://doi.org/10.1016/j.procs.2022.12.312

    2023 doi

  3. [11]

    S. R. Pokhrel, J. Ding, J. Park, O. -S. Park and J. Choi, ”Towards Enabling Critical mMTC: A Review of URLLC Within mMTC,” in IEEE Access, vol. 8, pp. 131796-131813, 2020, doi: 10.1109/AC- CESS.2020.3010271

    2020

  4. [12]

    B. S. Khan, S. Jangsher, A. Ahmed and A. Al-Dweik, ”URLLC and eMBB in 5G Industrial IoT: A Survey,” in IEEE Open Journal of the Communications Society, vol. 3, pp. 1134-1163, 2022, doi: 10.1109/OJCOMS.2022.3189013

    2022

  5. [13]

    Natale Patriciello, Sandra Lagen, Biljana Bojovic, Lorenza Giup- poni, An E2E simulator for 5G NR networks, Simulation Modelling Practice and Theory, V olume 96, 2019, 101933, ISSN 1569-190X, https://doi.org/10.1016/j.simpat.2019.101933

    2019

  6. [14]

    Seliem and D

    M. Seliem and D. Pesch, ”Software-Defined Time Sensitive Net- works (SD-TSN) for Industrial Automation,” 2022 14th Interna- tional Conference on Computational Intelligence and Communica- tion Networks (CICN), Al-Khobar, Saudi Arabia, 2022, pp. 1-7, doi: 10.1109/CICN56167.2022.10008262

    2022

  7. [15]

    Varga, A. (2010). OMNeT++. In: Wehrle, K., G ¨unes ¸, M., Gross, J. (eds) Modeling and Tools for Network Simulation. Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-642-12331-3 3

    2010 doi

  8. [16]

    Nardini, G., Sabella, D., Stea, G., Thakkar, P., & Virdis, A. (2020). Simu5G – An OMNeT++ library for end-to-end perfor- mance evaluation of 5G networks. IEEE Access, 8, 181176–181191. https://doi.org/10.1109/ACCESS.2020.3028550

    2020

  9. [17]

    ”SDAP-based QoS Flow Multiplexing Support in Simu5G for 5G NR Simulation.” arXiv preprint arXiv:2508.12785 (2025)

    Seliem, Mohamed, Utz Roedig, Cormac Sreenan, and Dirk Pesch. ”SDAP-based QoS Flow Multiplexing Support in Simu5G for 5G NR Simulation.” arXiv preprint arXiv:2508.12785 (2025)

    2025 arXiv

  10. [18]

    Debnath, M

    R. Debnath, M. S. Akinci, D. Ajith and S. Steinhorst, ”5GTQ: QoS- Aware 5G-TSN Simulation Framework,” 2023 IEEE 98th Vehicular Technology Conference (VTC2023-Fall), Hong Kong, Hong Kong, 2023, pp. 1-7, doi: 10.1109/VTC2023-Fall60731.2023.10333533

    2023

  11. [19]

    Nardini, G

    G. Nardini, G. Stea, A. Virdis, D. Sabella and P. Thakkar, ”Using Simu5G as a Realtime Network Emulator to Test MEC Apps in an End- To-End 5G Testbed,” 2020 IEEE 31st Annual International Symposium on Personal, Indoor and Mobile Radio Communications, London, UK, 2020, pp. 1-7,...

    2020

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.