XR Offloading Across Multiple Time Scales: The Roles of Power, Temperature, and Energy

Alessandro Catania; Antonella Molinaro; Carla Fabiana Chiasserini; Francesco Malandrino; Olga Chukhno

REVIEW 4 major objections 6 minor 16 references

XR Offloading Across Multiple Time Scales: The Roles of Power, Temperature, and Energy

T0 review · 4 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read This paper claims that a single stationary offloading probability can keep an XR wearable within its power, temperature, and battery limits at a tunable confidence, and that doing so cuts offloading cost by more than 35% relative to a…

desk verdict The thermal-aware offloading idea and COMSOL models are worth reading, but the central chance-constrained derivation is vacuous, the inequality directions are backwards, and the 35% cost claim is contradicted by the paper's own body. read the letter →

arxiv 2506.18584 v2 pith:LOL3AZZP submitted 2025-06-23 cs.NI

Francesco Malandrino , Olga Chukhno , Alessandro Catania , Antonella Molinaro , Carla Fabiana Chiasserini This is my paper · ORCID

classification cs.NI

keywords XRoffloadingwearablethermalconstraintsedgecomputingstochasticoptimizationPoissonarrivalstemperature-awarepowerbattery

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 argues that offloading decisions for XR wearables must be made jointly across three time scales: instantaneous power, short-term temperature, and long-term battery life. It proposes TAO, a stationary stochastic policy that sets the probability each incoming request is processed locally, and shows how to tune that probability so that power, temperature, and battery constraints are each met with a chosen confidence level. Under a linear time-invariant thermal model, the policy has a closed-form solution. In multiphysics 3D simulations of two commercial XR wearable designs, TAO keeps temperature below the 43°C burn-risk threshold while reducing offloading cost by over 35% compared with a state-of-the-art approach that ignores temperature.

What carries the argument

The load-bearing object is the per-device stationary local-execution probability $\alpha_d$, combined with closed-form Poisson quantile equations for the power, battery, and temperature constraints. The thermal model is the linear time-invariant convolution $\tau(d,t)=p(d,t)*h(d,t)$, reduced to the single-exponential step response $\tau(d,t)=B\bigl[(1-e^{-t/\theta})-(1-e^{-(t-T)/\theta})\bigr]$, which makes the safety threshold a quantile inequality on a Poisson random variable. This machinery collapses all three time scales into cumulative-distribution constraints on the same decision variable $\alpha_d$, so the stochastic optimization becomes solvable in closed form whenever closed-form impulse responses are available.

What would settle it

Run the paper's ten-request, one-hour workload on a physical XR wearable while TAO reports confidence 0.99 and measure skin or surface temperature directly; if temperature exceeds 43°C, or if the measured response to a step in local processing clearly deviates from the single-exponential form of Eq. (20), the central claim is falsified. A cheaper proxy is to compare the multiphysics-derived thermal constants against measured step-response curves for the same device.

Watch

Extended reading notes

Core claim

The central claim is that a time-invariant local-execution probability can simultaneously honor instantaneous power, short-term temperature, and long-term battery constraints when those constraints are expressed as quantiles of the request process. Because requests arrive as a Poisson process and each request consumes a fixed power increment, power, temperature, and battery level have Poisson-based distributions, so TAO computes the local-execution probability that maximizes local computation subject to each cumulative-distribution constraint. For the thermal branch, the paper adopts a first-order impulse response with thermal resistance $B$ and time constant $\theta$, which turns the 43°C limit into an explicit inequality on the Poisson cumulative distribution. In the reported simulations this policy never exceeds the temperature limit, whereas the baseline that drops the thermal constraint violates it for roughly 5% of the operating time.

Load-bearing premise

The no-overheating guarantee rests on assuming that each device heats and cools along a single-exponential linear curve with fixed thermal constants; if real wearables respond nonlinearly or with much slower dynamics, the temperature check is misspecified and the guarantee collapses.

Editorial extensions

If this is right

Lowering the confidence level from 0.99 to 0.80 cuts the normalized offloading cost from 0.84 to 0.33, so the same policy can be tuned to trade safety margin for cost in less critical scenarios.
Because the policy is stationary, it can be implemented as a simple per-device local-execution probability rather than an online per-request scheduler.
The method remains applicable when the thermal impulse response is not known in closed form, since TAO only needs the cumulative distributions, which can be obtained from measurements.
Thermal awareness comes nearly free on devices with ample thermal headroom but can require extra offloading on thermally constrained designs; the evaluation shows both behaviors across the two devices considered.
A purely battery- and power-aware strategy can violate the temperature limit for a non-negligible fraction of operating time, so thermal modeling is not a refinement but a necessary constraint for skin-contact XR devices.

Reading between the lines

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

Because the guarantee is probabilistic, an unlucky burst of arrivals can push temperature past 43°C with probability $1-\omega$; the reported safety is a tunable confidence, not a hard certificate for all possible sample paths.
The 35% cost saving is likely sensitive to the thermal time constant and thermal resistance, so measuring $B$ and $\theta$ for other wearable models would directly test how far the result generalizes.
The same cumulative-distribution-quantile machinery could be carried over to other skin-contact devices, such as VR headsets or smart glasses with different heat sinks, by replacing only the thermal impulse response.
A natural extension would be to replace the single-exponential thermal response with a measured multi-exponential response and re-derive the quantile constraint numerically, which the paper explicitly leaves open.

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

Quick take on arXiv:2506.18584. The framing is genuinely useful: offloading for XR wearables spans three time scales—instantaneous power, skin temperature, battery—and adding an explicit thermal constraint to a stochastic offloading policy is a legitimate extension. The COMSOL models of HoloLens and Google Glass are a real asset, and comparing against a temperature-blind baseline is the right experiment. The citation pattern looks fine.

The central math, though, does not hold up. Eq. (10) says F_p(T_p)≤ω where T_p is the ω-quantile of p. That is true by construction and says nothing about P_max. The actual chance constraint should be F_p(P_max)≥ω, and similarly for temperature and battery. Eq. (14) carries the wrong direction and uses T_p in place of P_max; Eq. (19) uses an undefined T_b; Eq. (21) repeats the same pattern; Eq. (15) is missing its bound. So TAO's advertised confidence guarantee is not derived. This is a load-bearing flaw in the theoretical section.

The evaluation has two additional soft spots. The thermal impulse response in Eq. (20) depends on B and θ, but the paper never reports their values or justifies the single-exponential form, so the COMSOL temperature traces cannot be tied to the optimization. The abstract's claim of a 35% reduction in offloading cost is contradicted by the body: for Google Glass TAO increases offloading cost relative to SoTA, and for HoloLens the cost is essentially unchanged. That claim appears to compare against something not in the paper.

None of this makes the central idea worthless. The three-time-scale decomposition is clear, and the COMSOL work is concrete and reproducible in principle. The errors are fixable: correct the inequalities, define the thresholds, report the impulse-response parameters and the α values, and re-state the cost comparison. As it stands, the paper does not establish its main theoretical or numerical claims.

My recommendation: this deserves peer review rather than desk rejection—the topic is timely and the modeling is a real contribution—but a serious referee should be directed to Sec. III-A first. I would ask for a revision with explicit derivations and a corrected baseline comparison.

Referee Report

4 major / 6 minor

Summary. This letter studies edge offloading for XR wearables under instantaneous power, short-term temperature, and long-term battery constraints. The authors formulate a stochastic optimization in which a stationary per-request offloading probability is chosen so that all constraints are met with tunable confidence omega, derive closed-form Poisson-based quantile constraints for the special case of homogeneous requests, and evaluate the resulting policy, TAO, in COMSOL simulations against a thermal-unaware baseline (SoTA) and a plugged-to-AC greedy policy. The paper claims that TAO reduces offloading cost by over 35% while avoiding violations of the 43 C temperature limit.

Significance. The multi-timescale problem framing is well motivated, and the use of detailed COMSOL models of commercial wearables (HoloLens and Google Glass) is a genuine strength that goes beyond abstract thermal abstractions. A tunable-confidence, stationary offloading policy that jointly respects power, thermal, and battery limits would be a useful contribution to the XR offloading literature. However, the central feasibility guarantee is not demonstrated: the quantile constraints are not correctly formulated, the closed-form expressions are internally inconsistent, and the evaluation does not report the quantities needed to connect the simulation results to the proposed optimization. As written, the paper's main claim is therefore unsupported.

major comments (4)

[Section III, Eq. (10)] The robust constraints in Eq. (10) are tautological and do not reference the actual operational limits. Since T_p is defined as the omega-quantile of p(d,t), the inequality F_{p(d,t)}(T_p) <= omega holds by definition up to atoms, and the same applies to T_tau and T_b. The required power chance constraint is Pr(p(d,t) <= P_max(d)) >= omega, which is never stated or used. Consequently Eq. (10) provides no feasibility guarantee for the problem described in Eqs. (6)-(8). This is a load-bearing defect in the central claim.
[Section III.A, Eqs. (12)-(21)] The closed-form constraints have the wrong inequality direction and an unexplained threshold substitution. From Eq. (13), enforcing a confidence omega requires the Poisson CDF evaluated at P_max(d)/pi_d to be >= omega. Eq. (14) instead replaces P_max(d)/pi_d by T_p/pi_d and writes '<= omega'; this both changes the threshold to the omega-quantile and reverses the inequality. No value of alpha satisfying Eq. (14) has been shown to imply Pr(p(d,t) <= P_max(d)) >= omega. The same defect propagates to Eq. (19), where T_b is never defined and the battery constraint is reduced to the deterministic mean inequality in Eq. (18), and to Eq. (21). The closed-form expressions therefore do not implement the advertised chance constraints.
[Section III.A, Eq. (20)] The thermal response used in the closed-form example is not justified and is inconsistent with the LTI convolution model in Eq. (5). Eq. (5) defines temperature as the convolution of power with an impulse response h(d,t), whereas Eq. (20) describes the response to a rectangular power pulse as a difference of two first-order step responses. Parameters B and theta are never fitted to the COMSOL models or otherwise justified, and the connection between the per-request power pi_d, the thermal resistance R_th, and the Poisson request count is not specified. Even after correcting the algebraic defect in Eq. (14), Eq. (21) would not express Pr(tau(d,t) <= tau_max) >= omega without a derivation of how the temperature limit maps to a count threshold.
[Section IV] The performance evaluation does not report the alpha values chosen by TAO, the empirical frequency of power/temperature/battery constraint violations, or which of the equations in Section III were actually solved in the simulations. The COMSOL temperature traces and the cost comparisons therefore cannot be attributed to the proposed stochastic optimization. In addition, the abstract's claim of a cost reduction exceeding 35% is not tied to a specific comparison number in the text or in Table II; the table reports only normalized TAO costs, with no corresponding SoTA value.

minor comments (6)

[Section II, Eqs. (7)-(8)] The constraints in Eqs. (7)-(8) are written as deterministic inequalities on random variables without a precise probabilistic quantifier; the intended meaning of 'with confidence omega' should be stated formally.
[Section III, Eq. (9)] The notation f_{p/b/tau}(d,t)(u) in Eq. (9) is difficult to parse; the subscript should be the random variable itself, e.g., f_{p(d,t)}(u), f_{b(d,t)}(u), f_{tau(d,t)}(u).
[Section III.A, Eq. (15)] The event N_d(t) >= N_d(T) in Eq. (15) is not the event used in the power constraint (12); please clarify what probability this expression is intended to compute.
[Fig. 3] The left panel and caption describe 'impulse responses,' but the plotted curves are the response to a finite-duration power pulse and are better described as step responses; the terminology should be corrected.
[Table II] The table heading contains the typo 'Normilized'; it should be 'Normalized.' Also, including the corresponding SoTA cost would make the claimed 35% improvement verifiable.
[Eq. (20)] The symbol T is used both for the time horizon and for the duration of the pulse stimulus in Eq. (20), which is confusing; a different symbol should be used for the pulse duration.

Circularity Check

2 steps flagged · score 6.0 of 10

Chance-constraint derivation is self-definitional: Eq. (10) evaluates each CDF at its own ω-quantile, and Eq. (14) substitutes that quantile for P_max, so the advertised confidence guarantee does not actually constrain TAO.

self definitional [Sec. III, Eq. (10)]
"Fp(d,t)(Tp)≤ ω, Fτ(d,t)(Tτ)≤ ω, and Fb(d,t)(Tb)≤ ω, (10) where Tp, Tτ, and Tb are the ω-quantiles of p(d,t), τ(d,t), and b(d,t), respectively."

By definition, the CDF of a continuous random variable evaluated at its own ω-quantile equals ω; for the discrete Poisson variable used later it is at least ω under the standard quantile convention. Thus Eq. (10) is either an identity or has the wrong inequality direction, and in neither case does it reference P_max, τ_max, or the battery limit. It is true for any offloading probability α, so it cannot be the robust constraint that determines TAO's behavior.
self definitional [Sec. III-A, Eq. (14)]
"By incorporating the ω-quantile, we obtain: Σ_{i=0}^{⌊Tp/πd⌋} (λdαdt)^i e^{-λdαdt}/i! ≤ ω. (14)"

Eq. (13) correctly uses the physical limit P_max(d)/π_d as the Poisson threshold. Eq. (14) replaces that threshold with T_p, the ω-quantile of p(d,t), and keeps the direction ≤ω. With T_p defined as the quantile, the left-hand side is identically ω (up to atoms), so the constraint is tautological; if T_p were meant to be P_max, the inequality should be ≥ω. The same substitution propagates to Eqs. (19) and (21), so the closed-form feasibility conditions do not enforce the advertised chance constraints.

full rationale

The numerical cost-reduction claim is not circular: it is obtained from COMSOL thermal simulations with fixed device models and compared against a state-of-the-art baseline, and no parameter of the optimization is fitted to the reported cost figures. The authors' prior work cited in [4] and [5] is background on user/content dynamics and is not load-bearing for TAO. However, the core chance-constraint derivation is self-definitional. Eq. (10) defines the constraints as CDFs evaluated at the quantiles of the same random variables, which is a tautology rather than a condition involving the physical limits. Eq. (14) then replaces the correct Poisson threshold P_max/π_d with the quantile T_p, again making the expression vacuous. Since the paper claims TAO guarantees power, thermal, and energy constraints with confidence ω, and this guarantee is the basis for the closed-form solution, the derivation reduces to quantile self-reference at its central step. The empirical COMSOL traces may still show that TAO avoids overtemperature in the specific scenarios, but the paper does not connect the reported α values or constraint slack to the derived inequalities, leaving the confidence guarantee unproven.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

No new physical entities or forces are introduced. The free parameters are thermal model constants (B, theta, heat transfer coefficient) and request statistics; they are inputs to the model, not fitted to the claimed cost reduction.

free parameters (2)

thermal impulse response parameters (B, theta) in Eq. (20) = not reported
The first-order thermal response tau(d,t) = B[(1-e^{-t/theta}) - (1-e^{-(t-T)/theta})] is assumed, with B = A*R_th and time constant theta; no numeric values are given, but the temperature curves in Fig. 3 require specific values.
heat transfer coefficient h = 8.3 W/(m^2 K) = 8.3 W/(m^2 K)
Set to simulate still air; this is a boundary condition chosen from literature, not fitted to TAO's result, but it strongly affects the temperature ceiling and therefore the policy.

assumptions (4)

domain assumption Requests arrive as a Poisson process with rate lambda_d, and local execution decisions are independent Bernoulli variables with probability alpha.
Invoked in Section III to obtain Poisson CDF quantile constraints. Real XR task arrivals are bursty and correlated, which the model does not capture.
domain assumption The linear power accumulation assumption p(d,t) = sum over locally-served requests of pi(d,r), with constant pi_d per device.
Justified via [11], [12]; nonlinear power interactions under multi-tasking are neglected.
domain assumption LTI thermal model: temperature is the convolution of power with impulse response, with constant heat capacity and thermal resistance.
Stated in Section II and supported by [11]-[13]; the closed-form solution further reduces to a two-step single-exponential response (Eq. 20).
domain assumption Requests are identical in size and computational requirements, and duration delta(r) is constant.
Stated in Section II as a simplification; heterogeneity is acknowledged but not modeled.

how reviews work

0 comments

Cite this review

Pith. "Pith review of XR Offloading Across Multiple Time Scales: The Roles of Power, Temperature, and Energy." pith.science (2026). https://pith.science/paper/LOL3AZZP

@misc{pith2026250618584,
  author       = {Pith},
  title        = {Pith review of: XR Offloading Across Multiple Time Scales: The Roles of Power, Temperature, and Energy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LOL3AZZP}},
  note         = {Machine review of arXiv:2506.18584}
}

read the original abstract

Extended reality (XR) devices, commonly known as wearables, must handle significant computational loads under tight latency constraints. To meet these demands, they rely on a combination of on-device processing and edge offloading. This letter focuses on offloading strategies for wearables by considering their impact across three time scales: instantaneous power consumption, short-term temperature fluctuations, and long-term battery duration. We introduce a comprehensive system model that captures these temporal dynamics, and propose a stochastic and stationary offloading strategy, called TAO (for temperature-aware offloading), designed to minimize the offloading cost while adhering to power, thermal, and energy constraints. Our performance evaluation, leveraging COMSOL models of real-world wearables, confirms that TAO reduces offloading cost by over 35% compared to state-of-the-art approaches, without violating the wearable operational limits.

Figures

Figures reproduced from arXiv: 2506.18584 by the authors.

**Figure 2.** Lateral view (left) and isometric view (right) of the 3D structure [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Impulse responses for the considered wearables (left); time evolution of the temperature for Microsoft HoloLens (center) and Google Glass (right). [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: Empirical distribution of the wearable temperatures (left), time evolution of the battery level (center), and cost (right) yielded by each strategy. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

16 extracted references · 16 canonical work pages

[1]

XR and 5G: Extended Reality at Scale with Time-Critical Communication

Fredrik Alriksson and others, “XR and 5G: Extended Reality at Scale with Time-Critical Communication.” [Online] (accessed August 25,

work page
[2]

Enabling Extended Reality Over 5G with Distributed Computing,

H. Sampath, P. Tinnakornsrisuphap, and P. Hande, “Enabling Extended Reality Over 5G with Distributed Computing,” IEEE Communications Magazine, 2024

work page 2024
[3]

Mobility-Aware Task Offloading Scheme for 6G Networks With Temporal Graph and Graph Matching,

J. Cai, X. Zhu, and A. E. Ackah, “Mobility-Aware Task Offloading Scheme for 6G Networks With Temporal Graph and Graph Matching,” IEEE Internet of Things Journal , 2024

work page 2024
[4]

Inter- play of User Behavior, Communication, and Computing in Immersive Reality 6G Applications,

O. Chukhno, O. Galinina, S. Andreev, A. Molinaro, and A. Iera, “Inter- play of User Behavior, Communication, and Computing in Immersive Reality 6G Applications,” IEEE Communications Magazine , 2022

work page 2022
[5]

User and Content Dynamics of Edge-Aided Immersive Reality Services,

O. Chukhno, O. Galinina, S. Andreev, A. Molinaro, and A. Iera, “User and Content Dynamics of Edge-Aided Immersive Reality Services,” IEEE Networking Letters , 2023

work page 2023
[6]

Future network requirements for ex- tended reality applications

Fredrik Alriksson and others, “Future network requirements for ex- tended reality applications.” [Online] (accessed August 25, 2025) https://www.ericsson.com/en/reports-and-papers/

work page 2025
[7]

Qualcomm and e& Collaborate on AI at the Edge to Drive Transformation in the UAE

Qualcomm Technologies, Inc, “Qualcomm and e& Collaborate on AI at the Edge to Drive Transformation in the UAE.” [Online] (accessed August 25, 2025) https://www.qualcomm.com/news/, 2025

work page 2025
[8]

Cost-minimized computation offloading of online multifunction ser- vices in collaborative edge-cloud networks,

C. Feng, P. Han, X. Zhang, Q. Zhang, Y . Zong, Y . Liu, and L. Guo, “Cost-minimized computation offloading of online multifunction ser- vices in collaborative edge-cloud networks,” IEEE Transactions on Network and Service Management , 2022

work page 2022

Show all 16 references

[9]

A deep reinforcement learning-based offloading scheme for multi-access edge computing-supported extended reality systems,

B. Trinh and G.-M. Muntean, “A deep reinforcement learning-based offloading scheme for multi-access edge computing-supported extended reality systems,” IEEE Transactions on Vehicular Technology, 2022

2022
[10]

Is It Getting Hot in Here? The Effects of VR Headset Microclimate Temperature on Perceived Thermal Discomfort, VR Sick- ness, and Skin Temperature,

M. A. Rupp, “Is It Getting Hot in Here? The Effects of VR Headset Microclimate Temperature on Perceived Thermal Discomfort, VR Sick- ness, and Skin Temperature,” Applied Ergonomics, 2024

2024
[11]

Parametric thermal modeling of heat transfer in handheld electronic devices,

J. Lee, D. W. Gerlach, and Y . K. Joshi, “Parametric thermal modeling of heat transfer in handheld electronic devices,” in 2008 11th Inter- society Conference on Thermal and Thermomechanical Phenomena in Electronic Systems, pp. 604–609, 2008

2008
[12]

Dynamic compact thermal models for skin temperature prediction of portable electronic devices based on convolution and fitting methods,

H. Liu, J. Yu, and R. Wang, “Dynamic compact thermal models for skin temperature prediction of portable electronic devices based on convolution and fitting methods,” International Journal of Heat and Mass Transfer, vol. 210, p. 124170, 2023

2023
[13]

Thermal model and countermeasures for future smart glasses,

K. Matsuhashi, T. Kanamoto, and A. Kurokawa, “Thermal model and countermeasures for future smart glasses,” Sensors, vol. 20, no. 5, 2020

2020
[14]

Using linear superposition to solve multiple heat source transient thermal problems,

D. T. Billings and R. P. Stout, “Using linear superposition to solve multiple heat source transient thermal problems,” in InterPACK, 2007

2007
[15]

Cengel and A

Y . Cengel and A. Chajar, Heat and Mass Transfer: Fundamentals and applications (Fifth Edition) . McGraw-Hill Education, 2015

2015
[2025]

https://www.ericsson.com/en/reports-and-papers, 2021

2021

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.

[1] [1]

XR and 5G: Extended Reality at Scale with Time-Critical Communication

Fredrik Alriksson and others, “XR and 5G: Extended Reality at Scale with Time-Critical Communication.” [Online] (accessed August 25,

work page

[2] [2]

Enabling Extended Reality Over 5G with Distributed Computing,

H. Sampath, P. Tinnakornsrisuphap, and P. Hande, “Enabling Extended Reality Over 5G with Distributed Computing,” IEEE Communications Magazine, 2024

work page 2024

[3] [3]

Mobility-Aware Task Offloading Scheme for 6G Networks With Temporal Graph and Graph Matching,

J. Cai, X. Zhu, and A. E. Ackah, “Mobility-Aware Task Offloading Scheme for 6G Networks With Temporal Graph and Graph Matching,” IEEE Internet of Things Journal , 2024

work page 2024

[4] [4]

Inter- play of User Behavior, Communication, and Computing in Immersive Reality 6G Applications,

O. Chukhno, O. Galinina, S. Andreev, A. Molinaro, and A. Iera, “Inter- play of User Behavior, Communication, and Computing in Immersive Reality 6G Applications,” IEEE Communications Magazine , 2022

work page 2022

[5] [5]

User and Content Dynamics of Edge-Aided Immersive Reality Services,

O. Chukhno, O. Galinina, S. Andreev, A. Molinaro, and A. Iera, “User and Content Dynamics of Edge-Aided Immersive Reality Services,” IEEE Networking Letters , 2023

work page 2023

[6] [6]

Future network requirements for ex- tended reality applications

Fredrik Alriksson and others, “Future network requirements for ex- tended reality applications.” [Online] (accessed August 25, 2025) https://www.ericsson.com/en/reports-and-papers/

work page 2025

[7] [7]

Qualcomm and e& Collaborate on AI at the Edge to Drive Transformation in the UAE

Qualcomm Technologies, Inc, “Qualcomm and e& Collaborate on AI at the Edge to Drive Transformation in the UAE.” [Online] (accessed August 25, 2025) https://www.qualcomm.com/news/, 2025

work page 2025

[8] [8]

Cost-minimized computation offloading of online multifunction ser- vices in collaborative edge-cloud networks,

C. Feng, P. Han, X. Zhang, Q. Zhang, Y . Zong, Y . Liu, and L. Guo, “Cost-minimized computation offloading of online multifunction ser- vices in collaborative edge-cloud networks,” IEEE Transactions on Network and Service Management , 2022

work page 2022