arxiv: 2604.19935 · v2 · submitted 2026-04-21 · 📡 eess.SP

Recognition: unknown

A Hybrid Gauss Markov LSTM Mobility Model for Indoor OWC

Walter Zibusiso Ncube , Ahmad Adnan Qidan , Taisir El-Gorashi , Jaafar M. H. Elmirghani

Authors on Pith no claims yet

Pith reviewed 2026-05-10 01:16 UTC · model grok-4.3

classification 📡 eess.SP

keywords mobility modelingindoor optical wireless communicationGauss-Markov modelLSTMdevice orientationposition predictiondata rate stabilitychannel estimation

0 comments

The pith

The hybrid GM-LSTM model jointly forecasts user position and device orientation more accurately than Random Waypoint or pure Gauss-Markov models for indoor optical wireless links.

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

This paper introduces a hybrid mobility model that pairs the Gauss-Markov process, which tracks correlated movement over time, with an LSTM network that learns the remaining non-linear patterns in how users walk and tilt their devices. Optical wireless systems lose performance quickly when transmitter and receiver fall out of alignment, so realistic forecasts of both location and orientation matter for estimating channels and testing network designs. The authors measure success through lower prediction errors and steadier per-user data rates across simulated dynamic indoor scenes, showing the combined model beats the two standard baselines.

Core claim

The Gauss-Markov component supplies the temporal correlation of user motion while the LSTM learns residual non-linear behavior, allowing the model to output joint predictions of position and orientation that produce higher accuracy and more stable data-rate traces than Random Waypoint or standalone Gauss-Markov models.

What carries the argument

The hybrid GM-LSTM model, in which Gauss-Markov supplies linear temporal correlation and LSTM captures non-linear residuals for simultaneous position and orientation prediction.

If this is right

Channel estimates and link budgets for OWC networks become more reliable when mobility is modeled with joint position-orientation forecasts.
System-level simulations of indoor high-capacity networks can use the model to test performance under realistic user behavior rather than simplified random paths.
Design choices for transmitter placement and beam steering can be evaluated against more stable predicted data-rate curves.
The approach supplies a concrete way to incorporate orientation dynamics that pure geometric models omit.

Where Pith is reading between the lines

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

Training the LSTM on general pedestrian traces rather than OWC-specific recordings may still yield usable gains if the residuals are broadly similar across indoor settings.
The same hybrid structure could be tested for predicting blockage events or handover triggers in visible-light or infrared networks.
Real-time deployment would require checking whether the LSTM can run with low enough latency to support adaptive rate control or beam realignment.

Load-bearing premise

That the LSTM can extract genuine non-linear movement and orientation patterns from the training traces without simply memorizing noise or requiring data sets that fail to represent other indoor OWC environments.

What would settle it

Compare prediction error and data-rate variance on held-out real indoor movement and orientation traces against the same quantities produced by Random Waypoint and Gauss-Markov models; a clear, consistent advantage for the hybrid would support the claim.

Figures

Figures reproduced from arXiv: 2604.19935 by Ahmad Adnan Qidan, Jaafar M. H. Elmirghani, Taisir El-Gorashi, Walter Zibusiso Ncube.

**Figure 2.** Figure 2: Channel prediction RMSE versus prediction horizon a [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Figure 3: presents the achievable data rate of a user over time, moving at a constant speed of 1.0 m/s. The x-axis represents time in seconds, showing how the data rate evolves as the user moves within the indoor environment under different mobility models. The RWP model exhibits significant fluctuations, with frequent spikes and deep drops, due to its memoryless mobility pattern that leads to inaccurate channel pre… view at source ↗

read the original abstract

Optical wireless communication (OWC) has emerged as a promising candidate for future high-capacity indoor wireless networks, driven by its large unregulated spectrum, high spatial reuse, and ability to support multi-gigabit data rates. However, OWC systems are highly sensitive to user mobility, as link performance depends strongly on the spatial alignment between transmitter and receiver. Accurate modelling of user position and device orientation is therefore essential for reliable channel estimation and system evaluation. To that effect, this paper proposes a hybrid Gauss--Markov and long short-term memory (GM--LSTM) mobility model for indoor OWC environments. The Gauss--Markov component captures the temporal correlation of user motion, while the LSTM learns residual behaviour to model non-linear movement patterns and orientation dynamics. The proposed model jointly predicts user position and device orientation, enabling improved representation of mobility in OWC channels. Performance is evaluated using prediction accuracy and per-user data rate evolution. Results show that the proposed hybrid GM--LSTM model outperforms conventional Random Waypoint and Gauss--Markov models, providing more accurate mobility prediction and more stable communication performance in dynamic indoor environments.

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.

Desk Editor's Note private letter to a colleague

The hybrid GM-LSTM model gives a practical mobility simulator for indoor OWC by adding LSTM residuals to Gauss-Markov, but thin validation details make the gains hard to trust.

read the letter

The paper's main point is a hybrid mobility model that runs a Gauss-Markov process for correlated user motion and then uses an LSTM to capture the leftover non-linear patterns in both position and device orientation. The authors report that this produces more accurate predictions and steadier per-user data rates than random waypoint or plain Gauss-Markov in indoor OWC settings where alignment matters a lot. That is the one or two things worth knowing up front. The combination itself is the new piece here. Prior work already uses Gauss-Markov for temporal correlation and LSTMs for sequence prediction, but applying the LSTM specifically to residuals and doing the joint position-orientation task for OWC channel evaluation is a targeted application rather than a new derivation. The paper does well at linking the mobility output straight to a system-level metric like data-rate stability, which shows they understand why better traces matter for OWC performance studies. The soft spots sit in the validation. The abstract supplies no equations, no dataset description, no train-test split, and no ablation that removes the LSTM. Without those, it is difficult to know whether the LSTM is learning general non-linear dynamics or simply fitting the particular traces used for training. The stress-test concern about overfitting therefore lands until the full paper shows otherwise. This work is aimed at researchers who run simulations to evaluate indoor OWC networks under realistic movement. A reader who needs a mobility generator that includes orientation would get the most direct value. It deserves peer review because the core idea is reasonable and tied to a real technology need, even if the evidence needs more transparent checks on data and generalization.

Referee Report

3 major / 2 minor

Summary. The manuscript proposes a hybrid Gauss-Markov LSTM (GM-LSTM) mobility model for indoor optical wireless communication (OWC). The Gauss-Markov component captures temporal correlation in user motion while the LSTM learns residual non-linear patterns and orientation dynamics. The model jointly predicts position and orientation and is evaluated on mobility prediction accuracy and per-user data-rate stability, claiming outperformance over Random Waypoint and pure Gauss-Markov baselines in dynamic indoor environments.

Significance. A well-validated hybrid mobility model could improve the fidelity of OWC channel and system evaluations, where link performance is highly sensitive to alignment and orientation. If the LSTM component demonstrably extracts generalizable non-linear residuals rather than overfitting to training traces, the approach would offer a practical advance over purely stochastic models. However, the current lack of methodological transparency prevents assessment of whether the reported gains are robust.

major comments (3)

Abstract: the central claim of outperformance in prediction accuracy and data-rate stability is stated without any supporting equations, training details, error bars, dataset description, or quantitative results, rendering the claim unverifiable from the provided text.
Model and Evaluation sections: no information is given on the source or characteristics of the training traces, the train/test split, regularization techniques, or an ablation study that isolates the LSTM contribution; without these, it is impossible to rule out that reported gains arise from overfitting to the specific dataset rather than genuine capture of non-linear residuals.
Results: the manuscript provides no statistical validation (e.g., confidence intervals, multiple random seeds, or cross-validation) for the claimed superiority in data-rate stability, which is load-bearing for the practical utility asserted in the abstract.

minor comments (2)

Notation for the hybrid prediction step (position and orientation) should be defined explicitly with equations to allow reproduction.
Figure captions and axis labels for prediction-error and data-rate plots require additional detail on the simulation parameters and number of runs.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments, which have helped us identify areas where the manuscript can be strengthened for clarity and rigor. We address each major comment below and have prepared revisions to the manuscript accordingly.

read point-by-point responses

Referee: Abstract: the central claim of outperformance in prediction accuracy and data-rate stability is stated without any supporting equations, training details, error bars, dataset description, or quantitative results, rendering the claim unverifiable from the provided text.

Authors: We agree that the abstract would benefit from additional quantitative context to support the claims. In the revised manuscript we have updated the abstract to include specific quantitative results on prediction accuracy (RMSE) and data-rate stability (variance reduction), along with a brief reference to error bars and the evaluation dataset. Full equations, training procedures, and detailed methodology remain in the body of the paper due to abstract length constraints. revision: partial
Referee: Model and Evaluation sections: no information is given on the source or characteristics of the training traces, the train/test split, regularization techniques, or an ablation study that isolates the LSTM contribution; without these, it is impossible to rule out that reported gains arise from overfitting to the specific dataset rather than genuine capture of non-linear residuals.

Authors: We acknowledge the need for greater methodological transparency. The revised Model and Evaluation sections now include a description of the training traces (synthetic data generated from a validated indoor pedestrian mobility simulator with documented speed and orientation statistics), the train/test split procedure, the regularization methods applied (including dropout and L2 penalties), and results from an ablation study that isolates the LSTM component. This ablation demonstrates incremental gains attributable to the LSTM's modeling of non-linear residuals beyond the Gauss-Markov baseline. revision: yes
Referee: Results: the manuscript provides no statistical validation (e.g., confidence intervals, multiple random seeds, or cross-validation) for the claimed superiority in data-rate stability, which is load-bearing for the practical utility asserted in the abstract.

Authors: We agree that statistical validation strengthens the results. The revised Results section now reports confidence intervals for the data-rate stability metric, computed across multiple independent runs with varied random seeds, together with k-fold cross-validation outcomes. These additions confirm the robustness of the reported improvements over the baseline models. revision: yes

Circularity Check

0 steps flagged

No circularity; hybrid model combines standard components with empirical validation

full rationale

The abstract and description present a hybrid GM-LSTM mobility model in which the Gauss-Markov component models temporal correlation of motion while the LSTM is described as learning residual non-linear patterns and orientation dynamics. Performance is assessed via prediction accuracy and per-user data-rate stability, with reported outperformance over Random Waypoint and pure Gauss-Markov baselines. No equations appear in the provided text, no fitted parameters are renamed as independent predictions, and no self-citations or uniqueness theorems are invoked as load-bearing premises. The derivation chain therefore consists of a standard architectural proposal followed by empirical comparison rather than any reduction of outputs to inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No full text or equations available; therefore no free parameters, axioms, or invented entities can be extracted. The central claim rests on the unstated assumption that LSTM residuals add predictive power beyond the Gauss-Markov component.

pith-pipeline@v0.9.0 · 5511 in / 1149 out tokens · 27231 ms · 2026-05-10T01:16:08.623294+00:00 · methodology

discussion (0)

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

Mobility Aware Power Control for VCSEL Based Indoor OWC
eess.SP 2026-04 unverdicted novelty 4.0

A hybrid Gauss-Markov and learning-based mobility model guides power allocation in dynamic VCSEL indoor OWC networks, yielding more accurate allocation and higher energy efficiency than conventional schemes in simulations.

Reference graph

Works this paper leans on

19 extracted references · 3 canonical work pages · cited by 1 Pith paper

[1]

Terabit indoor laser-based wireles s communications: Liﬁ 2.0 for 6g,

M. D. Soltani, A. A. Qidan, S. Huang, B. Y osuf, S. Mohamed, R. Singh, Y . Liu, W. Ali, R. Chen, H. Kazemi, E. Sarbazi, B. Ber de, D. Chiaroni, B. B´ echadergue, F. Abdel-dayem, H. Soni, J. Tabu, M. Perruf el, N. Seraﬁmovski, T. E. El-Gorashi, J. Elmirghani, M. Cris p, R. Penty, I. H. White, H. Haas, and M. Safari, “Terabit indoor laser-based wireles s co...

2023
[2]

Optimised energy efﬁciency of various cell sizes in l aser-based optical wireless communications,

W. Z. Ncube, A. A. Qidan, T. El-Gorashi, and J. M. H. Elmirg hani, “Optimised energy efﬁciency of various cell sizes in l aser-based optical wireless communications,” in 2024 IEEE International Mediterranean Conference on Commu nications and Networking (MeditCom) , 2024, pp. 389–394

2024
[3]

Intelligent reﬂecting surfaces assiste d laser-based optical wireless communication networks,

A. N. Hamad, W. Z. Ncube, A. A. Qidan, T. E. El-Gorashi, and J. M. Elmirghani, “Intelligent reﬂecting surfaces assiste d laser-based optical wireless communication networks,” in 2024 24th International Conference on Transparent Optical Networks (ICTON) , 2024, pp. 1–5

2024
[4]

Achieving 70 gb/s over a vcsel-based op tical wireless link using a multi-mode ﬁber-coupled receiver,

H. Kazemi, I. N. O. Osahon, N. Ledentsov, I. Titkov, N. Led entsov, and H. Haas, “Achieving 70 gb/s over a vcsel-based op tical wireless link using a multi-mode ﬁber-coupled receiver,” Journal of Lightwave Technology , vol. 43, no. 24, pp. 10 986–10 994, 2025

2025
[5]

An orientation-based random waypoint mo del for user mobility in wireless networks,

M. Dehghani Soltani, A. A. Purwita, Z. Zeng, C. Chen, H. Ha as, and M. Safari, “An orientation-based random waypoint mo del for user mobility in wireless networks,” in 2020 IEEE International Conference on Communications W ork shops (ICC W orkshops), 2020, pp. 1–6

2020
[6]

A comparative study of rand om waypoint and gauss-markov mobility models in the perform ance evaluation of manet,

T. D. Nguyen and A. A. Kassem, “A comparative study of rand om waypoint and gauss-markov mobility models in the perform ance evaluation of manet,” International Journal of Computer Networks & Communicatio ns, vol. 7, no. 5, pp. 1–15, 2015

2015
[7]

Mobility-a ware optical random waypoint and transfer learning-based l oad balancing,

A. Ramakrishnan, T. Balaiah, and A. Kumar R, “Mobility-a ware optical random waypoint and transfer learning-based l oad balancing,” International Journal of Ad Hoc and Ubiquitous Computing , vol. 48, pp. 94–109, 01 2025

2025
[8]

Non-stationary mobile-to-mobile channel modeling using the gauss-markov mobility model,

R. He, B. Ai, G. L. St¨ uber, and Z. Zhong, “Non-stationary mobile-to-mobile channel modeling using the gauss-markov mobility model,” in IEEE International Conference on Communications (ICC) . IEEE, 2017, pp. 1–6

2017
[9]

An improved gaus s-markov mobility model for fanet using ns3 simulation in 3- dimension environment,

V . Kumar, R. K. Dwivedi, and S. Prakash, “An improved gaus s-markov mobility model for fanet using ns3 simulation in 3- dimension environment,” in 2023 14th International Conference on Computing Communica tion and Networking Technologies (ICCCNT) , 2023, pp. 1–6

2023
[10]

Adege, H.-P

A. Adege, H.-P . Lin, G. Tarekegn, and Y . Y ayeh, Mobility Prediction in Wireless Networks Using Deep Learni ng Algorithm . Springer, 03 2020, pp. 454–461

2020
[11]

Mobility-aware cluster federated learning in hierarchic al wireless networks,

C. Feng, H. H. Y ang, D. Hu, Z. Zhao, T. Q. S. Quek, and G. Min , “Mobility-aware cluster federated learning in hierarchi cal wireless networks,” 2021. [Online]. Available: https://arxiv.org/abs/2108.09103

work page arXiv 2021
[12]

Indoor mobility prediction for mmwave communication s using markov chain,

A. Turkmen, S. Ansari, P . V . Klaine, L. Zhang, and M. A. Im ran, “Indoor mobility prediction for mmwave communication s using markov chain,” in 2021 IEEE Wireless Communications and Networking Conferen ce (WCNC) , 2021, pp. 1–5

2021
[13]

Soci al-aware pedestrian trajectory prediction via states reﬁn ement lstm,

P . Zhang, J. Xue, P . Zhang, N. Zheng, and W. Ouyang, “Soci al-aware pedestrian trajectory prediction via states reﬁn ement lstm,” IEEE Transactions on Pattern Analysis and Machine Intelligence , vol. PP , pp. 1–1, 11 2020

2020
[14]

Generative models for simulating mobility trajectories ,

V . Kulkarni, N. Tagasovska, T. V atter, and B. Garbinato , “Generative models for simulating mobility trajectories ,” CoRR, vol. abs/1811.12801, 2018. [Online]. Available: http://arxiv.org/abs/1811.12801

work page arXiv 2018
[15]

Modelling individual rou tines and spatio-temporal trajectories in human mobility,

L. Pappalardo and F. Simini, “Modelling individual rou tines and spatio-temporal trajectories in human mobility, ” CoRR, vol. abs/1607.05952, 2016. [Online]. Available: http://arxiv.org/abs/1607.05952

work page arXiv 2016
[16]

User mobili ty synthesis based on generative adversarial networks: A su rvey,

S. Shin, H. Jeon, C. Cho, S. Y oon, and T. Kim, “User mobili ty synthesis based on generative adversarial networks: A su rvey,” in 2020 22nd International Conference on Advanced Communication Technology (ICACT) , 2020, pp. 94–103

2020
[17]

Social lstm: Human trajectory predict ion in crowded spaces,

A. Alahi, K. Goel, V . Ramanathan, A. Robicquet, L. Fei-F ei, and S. Savarese, “Social lstm: Human trajectory predict ion in crowded spaces,” in Proceedings of the IEEE Conference on Computer Vision and Pa ttern Recognition (CVPR) , June 2016

2016
[18]

Pedestrian walki ng speed analysis: A systematic review,

M. Giannoulaki and Z. Christoforou, “Pedestrian walki ng speed analysis: A systematic review,” Sustainability, vol. 16, no. 11, 2024. [Online]. Available: https://www.mdpi.com/2071-1050/16/11/4813

2024
[19]

Bidirection al optical spatial modulation for mobile users: Toward a practical design for liﬁ systems,

M. D. Soltani, M. A. Arfaoui, I. Tavakkolnia, A. Ghrayeb , M. Safari, C. M. Assi, M. O. Hasna, and H. Haas, “Bidirection al optical spatial modulation for mobile users: Toward a practical design for liﬁ systems, ” IEEE Journal on Selected Areas in Communications , vol. 37, no. 9, pp. 2069–2086, 2019

2069