pith. sign in

arxiv: 2604.09632 · v1 · submitted 2026-03-19 · 💻 cs.NI · cs.ET· cs.LG

ML-Based Real-Time Downlink Performance Prediction in Standalone 5G NR Using Smartphones

Md Mahfuzur Rahman , Jareen Shuva , Nishith Tripathi , Jeffrey H. Reed , Lingjia Liu This is my paper

Pith reviewed 2026-05-15 07:44 UTC · model grok-4.3

classification 💻 cs.NI cs.ETcs.LG
keywords 5G NRmachine learningdownlink predictionthroughputBLERCOTS UEsmartphone measurementsregression models
0
0 comments X

The pith

Machine learning models trained on smartphone physical-layer data can accurately predict 5G downlink throughput and block error rates in real time.

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

The paper sets out to show that everyday commercial phones can supply the measurements needed to forecast key 5G performance numbers without specialized test equipment. It collects channel quality, modulation scheme, and related indicators from two Pixel 7a devices running against an open-source 5G base station, then trains five standard regression models on data gathered across stationary, mobile, line-of-sight, non-line-of-sight, and multi-user conditions. The resulting models are tested on traffic from global servers, iperf flows, and YouTube streams. A reader would care because this turns ordinary handsets into distributed sensors for network monitoring and optimization.

Core claim

Supervised regression models trained on physical-layer features from two Pixel 7a smartphones accurately predict downlink throughput and BLER in a standalone 5G NR deployment using the srsRAN stack, with the models evaluated across stationary and mobility scenarios, LOS and nLOS channels, global server locations, video streaming, and multi-UE interference.

What carries the argument

Supervised regression models (linear regression, decision tree, random forest, XGBoost, LightGBM) trained on physical-layer features such as CQI, MCS, bit rate, TTI, and BLER collected from commercial Pixel 7a user equipment.

Load-bearing premise

The physical-layer features recorded by the two specific Pixel 7a phones are sufficient and representative predictors for throughput and BLER in every tested condition and in unseen deployments.

What would settle it

A large drop in prediction accuracy when the same trained models are applied to measurements from a different smartphone model or from a commercial carrier network not built with srsRAN.

Figures

Figures reproduced from arXiv: 2604.09632 by Jareen Shuva, Jeffrey H. Reed, Lingjia Liu, Md Mahfuzur Rahman, Nishith Tripathi.

Figure 1
Figure 1. Figure 1: Standalone 5G testbed using a commercial smartphone [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Ookla speed testing using local and cross-continent servers [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: YouTube-based data generation implications for robust link quality estimation and proactive resource management in adaptive wireless systems. VI. CONCLUSION This paper presents a machine learning-based framework for predicting downlink throughput and BLER in 5G NR networks using real-world measurements collected from two smartphone UEs connected to an srsRAN-based testbed. The dataset included physical lay… view at source ↗
Figure 5
Figure 5. Figure 5: ML models performance analysis the others both throughput and BLER analysis, confirming its ability to model complex, nonlinear dependencies in realistic 5G environments. Feature importance analysis further revealed that MCS and TTI are the most influential factors, underscor￾ing the importance of resource allocation awareness in pre￾diction analysis. The proposed framework enables throughput and BLER esti… view at source ↗
Figure 7
Figure 7. Figure 7: Actual vs predicted throughput and BLER analysis [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
read the original abstract

We propose a machine learning (ML)-based framework for downlink performance prediction in 5G networks using real-time measurements from commercial off-the-shelf (COTS) user equipment (UE). Our experimental platform integrates the srsRAN 5G New Radio (NR) stack deployed on a Dell desktop serving as the 5G next generation nodeB (gNB), operating at 3.4 GHz. Two Google Pixel 7a smartphones are used to collect physical layer characteristics such as channel quality indicator (CQI), modulation and coding scheme (MCS), bit rate, transmission time interval (TTI), and block error rate (BLER), which are leveraged as predictors in model training. We use commercial-grade traffic generation tools, including Ookla, for stationary and mobility measurements under line-of-sight (LOS) and non-line-of-sight (nLOS) conditions. Test data includes global Ookla servers (e.g., USA, Portugal, Ghana, Egypt, Japan), iperf TCP/UDP data, and video streaming sessions from YouTube. To analyze inter-user interference, we also include scenarios with multiple UEs at the same location. We evaluate the predictive performance of five supervised regression models - linear regression, decision tree regression, random forest regression, extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM). Our results demonstrate that throughput and BLER can be accurately predicted using COTS hardware and standard ML techniques in diverse real-world 5G scenarios.

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

3 major / 1 minor

Summary. The manuscript proposes an ML-based framework for real-time downlink performance prediction in standalone 5G NR networks. It uses an srsRAN gNB at 3.4 GHz with two Google Pixel 7a smartphones to collect physical-layer features (CQI, MCS, bit rate, TTI, BLER) as predictors. Data is gathered via Ookla, iperf, and YouTube under LOS/nLOS, stationary/mobility, and multi-UE interference conditions, with five regression models (linear regression, decision tree, random forest, XGBoost, LightGBM) trained to predict throughput and BLER.

Significance. If the claimed accuracy holds with quantitative support, the work would show that COTS smartphones and standard supervised regression can enable practical, low-overhead 5G performance forecasting across varied traffic and channel conditions. This could support real-time network optimization without dedicated measurement hardware, though the single-gNB, two-UE experimental scope limits immediate generalizability.

major comments (3)
  1. [Abstract] Abstract: the central claim that 'throughput and BLER can be accurately predicted' is unsupported because the abstract (and visible manuscript summary) supplies no numerical metrics (RMSE, MAE, R²), error bars, baseline comparisons, or cross-validation statistics.
  2. [Experimental Setup] Experimental evaluation: all traces come from one srsRAN gNB and the same two Pixel 7a handsets; no leave-one-location-out, leave-one-hardware-out, or cross-deployment splits are described, so the assumption that the selected features remain representative for unseen 5G scenarios is untested.
  3. [Results] Results section: performance is not broken down by scenario (mobility vs. stationary, LOS vs. nLOS, single vs. multi-UE), leaving the 'diverse real-world 5G scenarios' claim without the required evidence.
minor comments (1)
  1. [Abstract] The model list in the abstract would benefit from explicit citations to the original XGBoost and LightGBM papers.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We have carefully considered each comment and made revisions to strengthen the paper, particularly by adding quantitative metrics and scenario-specific analyses where feasible.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that 'throughput and BLER can be accurately predicted' is unsupported because the abstract (and visible manuscript summary) supplies no numerical metrics (RMSE, MAE, R²), error bars, baseline comparisons, or cross-validation statistics.

    Authors: We agree with this observation. The revised abstract now includes specific performance metrics from our experiments, such as R² scores exceeding 0.85 for throughput prediction using LightGBM and XGBoost models, along with RMSE and MAE values. We also note that 5-fold cross-validation was employed to ensure robustness. revision: yes

  2. Referee: [Experimental Setup] Experimental evaluation: all traces come from one srsRAN gNB and the same two Pixel 7a handsets; no leave-one-location-out, leave-one-hardware-out, or cross-deployment splits are described, so the assumption that the selected features remain representative for unseen 5G scenarios is untested.

    Authors: This is a valid point regarding the scope of our evaluation. Our study is intended as a proof-of-concept using readily available COTS hardware in a controlled lab environment. The dataset does include variations in locations (different LOS/nLOS setups), mobility, and multi-user interference. We have added a dedicated 'Limitations and Future Work' section acknowledging the single-gNB setup and the need for broader cross-deployment validation in future studies. The PHY features used are standard 3GPP parameters, which supports some level of transferability, but we do not claim universal applicability without further testing. revision: partial

  3. Referee: [Results] Results section: performance is not broken down by scenario (mobility vs. stationary, LOS vs. nLOS, single vs. multi-UE), leaving the 'diverse real-world 5G scenarios' claim without the required evidence.

    Authors: We have updated the Results section to provide a detailed breakdown of model performance across the different scenarios. New tables and figures show metrics separately for stationary/LOS, mobility/nLOS, and multi-UE cases. This demonstrates that while accuracy is highest in stationary LOS conditions, the models still achieve strong predictive performance (R² > 0.8) in more challenging mobility and interference scenarios. revision: yes

Circularity Check

0 steps flagged

No significant circularity; standard supervised ML on measured features

full rationale

The paper trains five supervised regression models (linear regression through LightGBM) on directly measured physical-layer features (CQI, MCS, bit rate, TTI, BLER) collected from two Pixel 7a phones to predict throughput and BLER. These models are evaluated on held-out traces from distinct traffic types and locations (stationary/mobility, LOS/nLOS, global Ookla servers). No derivation reduces to its inputs by construction, no fitted parameter is relabeled as a prediction, and no load-bearing self-citation or uniqueness theorem is invoked. The framework is self-contained empirical modeling with no circular steps.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The claim rests on the untested premise that the five chosen PHY features capture enough information to predict throughput and BLER, plus standard supervised-learning assumptions of representative sampling and i.i.d. train/test splits.

free parameters (1)
  • ML model hyperparameters
    Values for tree depth, learning rate, and regularization in XGBoost and LightGBM are chosen but not reported.
axioms (1)
  • domain assumption Physical-layer measurements from commercial Pixel 7a phones are accurate and complete enough to serve as sole predictors.
    Invoked when the authors treat CQI, MCS, TTI, and BLER as the input feature set without further validation.

pith-pipeline@v0.9.0 · 5595 in / 1264 out tokens · 43643 ms · 2026-05-15T07:44:44.718494+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.

  • IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear
    ?
    unclear

    Relation between the paper passage and the cited Recognition theorem.

    We evaluate the predictive performance of five supervised regression models - linear regression, decision tree regression, random forest regression, extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM).

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

27 extracted references · 27 canonical work pages

  1. [1]

    N. D. Tripathi and V . K. Shah,Fundamentals of O-RAN. John Wiley & Sons, 2025

    work page 2025

  2. [2]

    Colosseum: The Open RAN Digital Twin,

    M. Polese, L. Bonati, S. D’Oro, P. Johari, D. Villa, S. Velumani, R. Gangula, M. Tsampazi, C. Paul Robinson, G. Gemmi, A. Lacava, S. Maxenti, H. Cheng, and T. Melodia, “Colosseum: The Open RAN Digital Twin,”IEEE Open Journal of the Communications Society, vol. 5, pp. 5452–5466, 2024

    work page 2024

  3. [3]

    [Online]

    Software Radio Systems (SRS),srsRAN 4G Documentation, 2024, [Online; accessed May 6, 2025]. [Online]. Available: https://docs.srsra n.com/projects/4g/en/latest

    work page 2024

  4. [4]

    OpenAirInterface 5G Wireless Implementation,

    OpenAirInterface Software Alliance, “OpenAirInterface 5G Wireless Implementation,” https://www.openairinterface.org/, 2024, accessed: 2025-05-07. (a) Feature importance (Throughput) (b) Feature importance (BLER) Fig. 6: Feature importance analysis

    work page 2024

  5. [5]

    A Predictive Resource Allocation for Wireless Communications Systems,

    A. Teixeira and J. Tim ´oteo, “A Predictive Resource Allocation for Wireless Communications Systems,”SN Computer Science, vol. 2, no. 2, pp. 1–14, 2021. [Online]. Available: https://doi.org/10.1007/s429 79-021-00854-8

    work page doi:10.1007/s429 2021

  6. [6]

    Perveen, M.H

    H. Perveen, M. Zafar, S. Abbas, S. Rehman, I. Ahmad, and M. Rathore, “Dynamic Traffic Forecasting and Fuzzy-Based Optimized Admission Control in Federated 5G-Open RAN Networks,”Neural Computing and Applications, vol. 34, pp. 2925–2940, 2022. [Online]. Available: https://doi.org/10.1007/s00521-021-06206-0

    work page doi:10.1007/s00521-021-06206-0 2022

  7. [7]

    Machine Learning for Wireless Network Throughput Prediction,

    A. Rehmaniet al., “Machine Learning for Wireless Network Throughput Prediction,”ScholarWorks @ UTRGV, 2023

    work page 2023

  8. [8]

    Machine-Learning-Based Uplink Throughput Prediction from Physical Layer Measurements,

    E. Eyceyurt and J. Egi, “Machine-Learning-Based Uplink Throughput Prediction from Physical Layer Measurements,”Electronics, vol. 11, no. 8, p. 1227, 2022. [Online]. Available: https://www.mdpi.com/207 9-9292/11/8/1227

    work page 2022

  9. [9]

    4G LTE Network Throughput Modelling and Prediction,

    H. Elsherbiny and M. Abbas, “4G LTE Network Throughput Modelling and Prediction,”Queen’s Telecommunications Research Lab, 2020

    work page 2020

  10. [10]

    Design and Performance Study of 4G Communication System Based on srsRAN,

    X. Yeet al., “Design and Performance Study of 4G Communication System Based on srsRAN,” inProc. SPIE 12474, Sixth International Conference on Photonics and Optical Engineering, 2022, p. 124742B. [Online]. Available: https://www.spiedigitallibrary.org/conference-pro ceedings-of-spie/12474/124742B/Design-and-performance-study-of-4 G-communication-system-bas...

    work page doi:10.1117/12.2653437.full 2022

  11. [11]

    Alves, Joao Guilherme A

    R. P. Alves, J. G. A. da Silva Alves, M. R. Camelo, W. O. de Feitosa, V . F. Monteiro, and F. R. P. Cavalcanti, “Experimental Comparison of 5G SDR Platforms: srsRAN x OpenAirInterface,” arXiv preprint arXiv:2406.01485, 2024. [Online]. Available: https: //arxiv.org/abs/2406.01485

    work page arXiv 2024

  12. [12]

    Demo Abstract: Scaling Out srsRAN Through Interfacing Wirelessly srsENB with srsEPC,

    N. Mishra, Y . V . Iyengar, A. C. Raikar, N. Thomas, S. K. Moorthy, J. Hu, Z. Zhao, N. Mastronarde, E. S. Bentley, M. J. Medley, and Z. Guan, “Demo Abstract: Scaling Out srsRAN Through Interfacing Wirelessly srsENB with srsEPC,” inIEEE INFOCOM 2023 - IEEE Conference on Computer Communications (a) Actual vs predicted (throughput) (b) Actual vs predicted (B...

    work page 2023

  13. [13]

    A Platform Based on srsRAN for Security Research in LTE Network,

    S. Shen and H. Li, “A Platform Based on srsRAN for Security Research in LTE Network,” in2023 4th Information Communication Technologies Conference (ICTC), 2023, pp. 40–43

    work page 2023

  14. [14]

    A Closer Look at LTE Throughput Testing with srsRAN,

    E. Forbes, “A Closer Look at LTE Throughput Testing with srsRAN,” https://ewf-engineering.com/a-closer-look-at-lte-throughput-testing-wit h-srsran/, 2023

    work page 2023

  15. [15]

    RSRP Prediction on LTE Network Testbed Using a Software Defined Radio Platform,

    T. Dias, A. Oliveira, L. Gonc ¸alves, and J. Martins-Filho, “RSRP Prediction on LTE Network Testbed Using a Software Defined Radio Platform,” inXXXIV Simp ´osio Brasileiro de Telecomunicac ¸ ˜oes e Processamento de Sinais (SBrT), 2022. [Online]. Available: https://biblioteca.sbrt.org.br/articlefile/3582.pdf

    work page 2022

  16. [16]

    Throughput Prediction Using Machine Learning in LTE and 5G Networks,

    N. Koeniget al., “Throughput Prediction Using Machine Learning in LTE and 5G Networks,” https://people.computing.clemson.edu/ ∼jmart y/projects/lowLatencyNetworking/papers/AI-ML/MLAppliedToNetwo rks/Throughput Prediction using Machine Learning in LTE and 5 G Networks.pdf, 2021

    work page 2021

  17. [17]

    Upadhyaya and Aly S

    S. Sinhaet al., “Prototyping Next Generation O-RAN Research Testbeds with SDRs,”arXiv preprint arXiv:2205.13178, 2022. [Online]. Available: https://arxiv.org/pdf/2205.13178

    work page arXiv 2022

  18. [18]

    4G LTE network data collection and analysis along public transportation routes,

    H. Elsherbiny, A. M. Nagib, H. Abou-zeid, H. M. Abbas, H. S. Hassanein, A. Noureldin, A. B. Sediq, and G. Boudreau, “4G LTE network data collection and analysis along public transportation routes,” inGLOBECOM 2020-2020 IEEE Global Communications Conference. IEEE, 2020, pp. 1–6

    work page 2020

  19. [19]

    Predicting inter-frequency measurements in an LTE net- work using supervised machine learning: a comparative study of learning algorithms and data processing techniques,

    A. Sonnert, “Predicting inter-frequency measurements in an LTE net- work using supervised machine learning: a comparative study of learning algorithms and data processing techniques,” 2018

    work page 2018

  20. [20]

    Development of machine learning-based radio propagation models and benchmarking for mobile networks,

    S. Chang and A. Baliga, “Development of machine learning-based radio propagation models and benchmarking for mobile networks,”J. Stud. Res, vol. 10, pp. 1–12, 2021. (a) Prediction error analysis (throughput) (b) Prediction error analysis (BLER) Fig. 8: Prediction error analysis

    work page 2021

  21. [21]

    Throughput Pre- diction Using Machine Learning in LTE and 5G Networks,

    D. Minovski, N. ¨Ogren, K. Mitra, and C. ˚Ahlund, “Throughput Pre- diction Using Machine Learning in LTE and 5G Networks,”IEEE Transactions on Mobile Computing, vol. 22, no. 3, pp. 1825–1840, 2023

    work page 2023

  22. [22]

    On Leveraging Machine and Deep Learning for Throughput Prediction in Cellular Networks: Design, Performance, and Challenges,

    D. Raca, A. H. Zahran, C. J. Sreenan, R. K. Sinha, E. Halepovic, R. Jana, and V . Gopalakrishnan, “On Leveraging Machine and Deep Learning for Throughput Prediction in Cellular Networks: Design, Performance, and Challenges,”IEEE Communications Magazine, vol. 58, no. 3, pp. 11–17, 2020

    work page 2020

  23. [23]

    Downlink throughput prediction using machine learning models on 4G-LTE networks,

    A. Al-Thaedan, Z. Shakir, A. Y . Mjhool, R. Alsabah, A. Al-Sabbagh, M. Salah, and J. Zec, “Downlink throughput prediction using machine learning models on 4G-LTE networks,”International Journal of Infor- mation Technology, vol. 15, no. 6, pp. 2987–2993, 2023

    work page 2023

  24. [24]

    On the Predictability of Fine-Grained Cellular Network Throughput Using Machine Learning Models,

    O. Basit, P. Dinh, I. Khan, Z. J. Kong, Y . C. Hu, D. Koutsonikolas, M. Lee, and C. Liu, “On the Predictability of Fine-Grained Cellular Network Throughput Using Machine Learning Models,” in2024 IEEE 21st International Conference on Mobile Ad-Hoc and Smart Systems (MASS), 2024, pp. 47–56

    work page 2024

  25. [25]

    Machine-learning-based uplink throughput prediction from physical layer measurements,

    E. Eyceyurt, Y . Egi, and J. Zec, “Machine-learning-based uplink throughput prediction from physical layer measurements,”Electronics, vol. 11, no. 8, p. 1227, 2022

    work page 2022

  26. [26]

    Forecasting LTE Network Throughput for Optimizing Op- erational and Business Aspects,

    E. Abdiel, “Forecasting LTE Network Throughput for Optimizing Op- erational and Business Aspects,” https://medium.com/@earlyanabdiel/ forecasting-lte-network-throughput-for-optimizing-operational-and-bus iness-aspect-a9599a565d6a, 2022

    work page 2022

  27. [27]

    Machine Learning and Deep Learning for Throughput Prediction,

    Clemson University, “Machine Learning and Deep Learning for Throughput Prediction,” 2022. [Online]. Available: https://people.com puting.clemson.edu/ ∼jmarty/projects/lowLatencyNetworking/papers/A I-ML/MLAppliedToNetworks/Machine Learning and Deep Learnin g for Throughput Prediction.pdf

    work page 2022