pith. sign in

arxiv: 2508.03287 · v1 · submitted 2025-08-05 · 💻 cs.NI · cs.DC

Directives for Function Offloading in 5G Networks Based on a Performance Characteristics Analysis

Falk Dettinger , Matthias Wei{\ss} , Daniel Baumann , Martin Sommer , Michael Weyrich This is my paper

Pith reviewed 2026-05-19 00:57 UTC · model grok-4.3

classification 💻 cs.NI cs.DC
keywords 5G networkscloud offloadingvehicle AIlatencyround trip timepacket error ratenon-standalone 5Gfunction offloading
0
0 comments X

The pith

Real-world 5G tests indicate that cloud offloading for vehicle AI functions is suitable only when round-trip time exceeds 150 milliseconds.

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

This paper tests how well current 5G networks can handle sending vehicle AI tasks like object and emotion recognition to the cloud. The researchers collected data on latency, signal quality, and errors while driving an 8.8 km route covering cities, countryside, and woods. They compared different cloud locations and ways to run the software. The results show stable connections with very few packet losses, but they conclude that offloading only makes sense if the round trip takes more than 150 ms. This provides practical rules for when cars should process data locally versus sending it away to save energy.

Core claim

Measurements in non-standalone 5G networks along an 8.8 km route in Baden-Württemberg show an average signal quality of 84% with no connectivity interruptions and packet error rates below 0.1%. Transfer times vary by location and server connections while processing times depend on hardware. Cloud offloading appears suitable only when round trip time exceeds 150 ms.

What carries the argument

Performance characteristics analysis of latency, round trip time, packet delivery ratio, and signal quality for AI algorithms executed on cloudlet and cloud platforms using conventional, containerized, and orchestrated deployments.

If this is right

  • Offloading decisions in vehicles should incorporate real-time round trip time checks to decide between local and cloud execution.
  • 5G non-standalone networks provide reliable packet delivery for vehicle functions despite varying terrain.
  • Processing hardware in the cloud has a larger impact on overall time than network transfer in some cases.
  • Deployment strategies like container orchestration affect transfer times depending on backend network connections.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The 150 ms threshold may shift with the rollout of standalone 5G networks that promise lower latency.
  • Directives could extend to other vehicle functions such as path planning or sensor fusion if similar performance holds.
  • Repeating the tests in different geographic regions would test how universal the location-dependent transfer times are.

Load-bearing premise

The performance data gathered on one 8.8 km route in Germany using two particular AI tasks can guide offloading decisions for vehicle functions in general 5G setups.

What would settle it

Demonstrating reliable and beneficial cloud offloading for similar AI vehicle functions at round trip times below 150 ms on a 5G network would disprove the main directive.

Figures

Figures reproduced from arXiv: 2508.03287 by Daniel Baumann, Falk Dettinger, Martin Sommer, Matthias Wei{\ss}, Michael Weyrich.

Figure 1
Figure 1. Figure 1: Overview of the experimental setup. The vehicle is connected to the 5G [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Map outlining the route in southern Germany including the 5G Non [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Box plots for executing the object recognition algorithm in the BWCloud in Mannheim and the Telekom Cloudlet in Frankfurt. The functions are [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Box plots for executing the emotion recognition algorithm in the BWCloud in Mannheim and the Telekom Cloudlet in Frankfurt. The functions [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
read the original abstract

Cloud-based offloading helps address energy consumption and performance challenges in executing resource-intensive vehicle algorithms. Utilizing 5G, with its low latency and high bandwidth, enables seamless vehicle-to-cloud integration. Currently, only non-standalone 5G is publicly available, and real-world applications remain underexplored compared to theoretical studies. This paper evaluates 5G non-standalone networks for cloud execution of vehicle functions, focusing on latency, Round Trip Time, and packet delivery. Tests used two AI-based algorithms -- emotion recognition and object recognition -- along an 8.8 km route in Baden-W\"urttemberg, Germany, encompassing urban, rural, and forested areas. Two platforms were analyzed: a cloudlet in Frankfurt and a cloud in Mannheim, employing various deployment strategies like conventional applications and containerized and container-orchestrated setups. Key findings highlight an average signal quality of 84 %, with no connectivity interruptions despite minor drops in built-up areas. Packet analysis revealed a Packet Error Rate below 0.1 % for both algorithms. Transfer times varied significantly depending on the geographical location and the backend servers' network connections, while processing times were mainly influenced by the computation hardware in use. Additionally, cloud offloading seems only be a suitable option, when a round trip time of more than 150 ms is possible.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 1 minor

Summary. The manuscript reports real-world 5G NSA measurements for offloading two AI vehicle functions (emotion recognition and object recognition) to a cloudlet in Frankfurt and a cloud in Mannheim. Experiments were conducted along an 8.8 km route in Baden-Württemberg covering urban, rural, and forested terrain using conventional, containerized, and orchestrated deployments. Key results include 84 % average signal quality, packet error rate below 0.1 %, location- and backend-dependent transfer times, and hardware-dependent processing times. From these data the authors derive directives for function offloading, notably that cloud offloading is suitable only when RTT exceeds 150 ms.

Significance. The direct field measurements of signal quality, low packet error rates, and observed timing variations supply concrete empirical data on currently available 5G non-standalone performance for vehicular offloading—an area the paper correctly notes is underexplored relative to theoretical work. If the 150 ms threshold and route representativeness can be substantiated, the results would offer practical guidance for deployment decisions.

major comments (1)
  1. [Abstract] Abstract: the claim that 'cloud offloading seems only be a suitable option, when a round trip time of more than 150 ms is possible' is presented without any referenced table, figure, plot, or statistical test that identifies 150 ms as an inflection point or suitability threshold. The text states that transfer times vary with geography and backend connections, yet supplies no derivation linking the measured times to the specific 150 ms cutoff.
minor comments (1)
  1. [Abstract] Abstract: grammatical correction needed—'seems only be a suitable option' should read 'seems to be a suitable option only'.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address the major comment point by point below and have made revisions to improve clarity and substantiation where the feedback indicates a need.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that 'cloud offloading seems only be a suitable option, when a round trip time of more than 150 ms is possible' is presented without any referenced table, figure, plot, or statistical test that identifies 150 ms as an inflection point or suitability threshold. The text states that transfer times vary with geography and backend connections, yet supplies no derivation linking the measured times to the specific 150 ms cutoff.

    Authors: We appreciate the referee's observation that the abstract presents the 150 ms threshold without an explicit link to supporting data or derivation. This threshold was identified from the experimental results on transfer times, which vary by location along the route and by backend (cloudlet vs. cloud), combined with the hardware-dependent processing times for the two AI functions. Specifically, our measurements indicated that total offloading latency (transfer plus remote processing) only yields a practical advantage over local vehicle execution when RTT exceeds this value. We acknowledge, however, that the abstract does not reference the relevant results or provide a clear derivation. In the revised manuscript we will update the abstract to include a direct reference to the results section and add a new figure that plots total offloading time against measured RTT values, with the 150 ms point marked as the crossover where cloud offloading becomes suitable. This addition will supply the requested substantiation while preserving the original empirical observations. revision: yes

Circularity Check

0 steps flagged

No circularity: directives derived from direct empirical measurements

full rationale

The paper reports results from real-world 5G network tests along an 8.8 km route using two AI algorithms on cloudlet and cloud platforms, measuring signal quality, packet error rates, transfer times, and processing times. The 150 ms RTT suitability statement is an interpretive summary of these observed performance characteristics rather than any derivation, equation, fitted parameter, or self-citation that reduces to the input data by construction. No load-bearing steps invoke uniqueness theorems, ansatzes, or renamings; the work remains self-contained as experimental data collection and post-hoc interpretation without circular reduction.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on empirical observations from field tests rather than mathematical derivations or new postulates.

free parameters (1)
  • RTT suitability threshold = 150 ms
    The 150 ms value is presented as the point where offloading becomes suitable based on observed transfer and processing time variations in the experiments.
axioms (1)
  • domain assumption The tested route encompassing urban, rural, and forested areas and the two AI algorithms are representative of typical resource-intensive vehicle functions and operating conditions.
    Invoked implicitly to generalize the performance findings and directives beyond the specific 8.8 km test in Baden-Württemberg.

pith-pipeline@v0.9.0 · 5780 in / 1530 out tokens · 76413 ms · 2026-05-19T00:57:23.188330+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

22 extracted references · 22 canonical work pages

  1. [1]

    Revealing the complexity of automotive software,

    V . Antinyan, “Revealing the complexity of automotive software,” in Proceedings of the 28th ACM Joint Meeting on European Software Engineering Conference and Symposium on the F oundations of Software Engineering, P. Devanbu, M. Cohen, and T. Zimmermann, Eds. New York, NY , USA: ACM, 2020, pp. 1525–1528

    work page 2020

  2. [2]

    Artificial intelligence and sensor technology in the automotive industry: An overview,

    S. M. Ammal, M. Kathiresh, and R. Neelaveni, “Artificial intelligence and sensor technology in the automotive industry: An overview,” in Automotive Embedded Systems , ser. EAI/Springer Innovations in Com- munication and Computing, M. Kathiresh and R. Neelaveni, Eds. Cham: Springer International Publishing, 2021, pp. 145–164

    work page 2021

  3. [3]

    Connected vehicle: Ontology, taxonomy and use cases,

    D. Baumann, M. Sommer, F. Dettinger, T. R ¨osch, M. Weyrich, and E. Sax, “Connected vehicle: Ontology, taxonomy and use cases,” in 2024 IEEE International Systems Conference (SysCon), Montreal, 15th - 18th April 2024 . Institute of Electrical and Electronics Engineers (IEEE), 2024, p. 1–6

    work page 2024

  4. [4]

    Total cost of ownership: Cloud-based vs. onboard vehicle software components,

    D. Baumann, M. Sommer, E. Sax, F. Dettinger, and M. Weyrich, “Total cost of ownership: Cloud-based vs. onboard vehicle software components,” in The 1st International Conference on Systems Scalability and Expandability . International Academy, Research, and Industry Association (IARIA), 2024, p. 6 S

    work page 2024

  5. [5]

    Vehicular cloud com- puting in the dawn of 5g,

    A. Falchetti, C. Azurdia-Meza, and S. Cespedes, “Vehicular cloud com- puting in the dawn of 5g,” in 2015 CHILEAN conference on electrical, electronics engineering, information and communication technologies (CHILECON). IEEE, 2015, pp. 301–305

    work page 2015

  6. [6]

    A taxonomy and survey of edge cloud computing for intelligent transportation systems and connected vehicles,

    P. Arthurs, L. Gillam, P. Krause, N. Wang, K. Halder, and A. Mouzakitis, “A taxonomy and survey of edge cloud computing for intelligent transportation systems and connected vehicles,” IEEE Transactions on Intelligent Transportation Systems, vol. 23, no. 7, pp. 6206–6221, 2021

    work page 2021

  7. [7]

    Autonomous vehicles in 5g and beyond: A survey,

    S. Hakak, T. R. Gadekallu, P. K. R. Maddikunta, S. P. Ramu, C. De Al- wis, M. Liyanage et al. , “Autonomous vehicles in 5g and beyond: A survey,” V ehicular Communications, vol. 39, p. 100551, 2023

    work page 2023

  8. [8]

    5g nsa performance: A mea- surement study,

    J. Moreno, M. Contini, and A. Aguiar, “5g nsa performance: A mea- surement study,” in 2024 19th Wireless On-Demand Network Systems and Services Conference (WONS) . IEEE, 2024, pp. 1–8

    work page 2024

  9. [9]

    Comparison of 5g networks non- standalone architecture (nsa) and standalone architecture (sa),

    M. Alnaas and O. Alhodairy, “Comparison of 5g networks non- standalone architecture (nsa) and standalone architecture (sa),” network, vol. 11, p. 12, 2024

    work page 2024

  10. [10]

    Chapter 9 - localization and tracking,

    K. Witrisal, C. Anton-Haro, S. Grebien, W. Joseph, E. Leitinger, X. Li, J. A. Del Peral-Rosado, D. Plets, J. Vil `a-Valls, and T. Wilding, “Chapter 9 - localization and tracking,” in Inclusive Radio Communications for 5G and Beyond , C. Oestges and F. Quitin, Eds. Academic Press, 2021, pp. 253–293. [Online]. Available: https: //www.sciencedirect.com/scien...

    work page 2021

  11. [11]

    A comparison of the v2x communication systems: Its-g5 and c-v2x,

    V . Mannoni, V . Berg, S. Sesia, and E. Perraud, “A comparison of the v2x communication systems: Its-g5 and c-v2x,” in 2019 IEEE 89th V ehicular Technology Conference (VTC2019-Spring) , 2019, pp. 1–5

    work page 2019

  12. [12]

    Understanding computation time: a critical discussion of time as a computational performance metric,

    D. Harris-Birtill and R. Harris-Birtill, “Understanding computation time: a critical discussion of time as a computational performance metric,” in Time in V ariance. Brill, 2021, pp. 220–248

    work page 2021

  13. [13]

    A holistic survey of multipath wireless video streaming,

    S. Afzal, V . Testoni, C. E. Rothenberg, P. Kolan, and I. Bouazizi, “A holistic survey of multipath wireless video streaming,” Journal of Network and Computer Applications , vol. 212, p. 103581, 2023. [Online]. Available: https://www.sciencedirect.com/science/article/pii/ S1084804522002223

    work page 2023

  14. [14]

    A systematic review of quality of service in wireless sensor networks using machine learning: Recent trend and future vision,

    M. Pundir and J. K. Sandhu, “A systematic review of quality of service in wireless sensor networks using machine learning: Recent trend and future vision,” Journal of Network and Computer Applications, vol. 188, p. 103084, 2021. [Online]. Available: https: //www.sciencedirect.com/science/article/pii/S1084804521001065

    work page 2021

  15. [15]

    A survey on performance, current and future usage of vehicle-to- everything communication standards,

    F. Dettinger, M. Weiß, D. Dittler, J. St¨umpfle, M. Artelt, and M. Weyrich, “A survey on performance, current and future usage of vehicle-to- everything communication standards,” arXiv preprint arXiv:2410.10264, 2024

    work page arXiv 2024

  16. [16]

    Vehicular cloud computing through dynamic computation offloading,

    A. Ashok, P. Steenkiste, and F. Bai, “Vehicular cloud computing through dynamic computation offloading,” Computer Communications, vol. 120, pp. 125–137, 2018

    work page 2018

  17. [17]

    Process for the identification of vehicle functions for cloud offloading,

    M. Sommer, D. Baumann, T. R ¨osch, F. Dettinger, E. Sax, and M. Weyrich, “Process for the identification of vehicle functions for cloud offloading,” in Intelligent Computing , K. Arai, Ed. Cham: Springer Nature Switzerland, 2024, pp. 596–608

    work page 2024

  18. [18]

    Offloading au- tonomous vehicle machine learning algorithms to the 5g edge: A proof of concept implementation,

    V . Boutin, A. Hannart, A. Essaidi, and B. Sans `o, “Offloading au- tonomous vehicle machine learning algorithms to the 5g edge: A proof of concept implementation,” in 2021 IEEE 4th 5G World F orum (5GWF) , 2021, pp. 269–274

    work page 2021

  19. [19]

    Intelligent task offloading for heterogeneous v2x communications,

    K. Xiong, S. Leng, C. Huang, C. Yuen, and Y . L. Guan, “Intelligent task offloading for heterogeneous v2x communications,” IEEE Transactions on Intelligent Transportation Systems , vol. 22, no. 4, pp. 2226–2238, 2021

    work page 2021

  20. [20]

    5G-M2 EVB kit,

    Quectel, “5G-M2 EVB kit,” https://www.quectel.com/product/ 5g-m2-evb-kit/#specifications, online; accessed on: 12. Febraura 2025

    work page 2025

  21. [21]

    5G RM520N series,

    ——, “5G RM520N series,” https://www.quectel.com/product/ 5g-rm520n-series/#summary, online; accessed on: 12. Febraura 2025

    work page 2025

  22. [22]

    Large-scale modeling of future automotive data traffic towards the edge cloud,

    R. Protzmann, K. K. Huebner, N. Ascheuer, U. Bauknecht, T. Enderle, U. Gebhard, C. Raack, and A. Witt, “Large-scale modeling of future automotive data traffic towards the edge cloud,” in Photonic Networks; 20th ITG-Symposium . VDE, 2019, pp. 1–3

    work page 2019