Pith. sign in

REVIEW 5 major objections 4 minor 23 references

ProCAVE: A Self-Adaptive, Full-Lifecycle Edge Caching Framework for Video Streaming via Predictive Bandwidth Estimation and Preference-Aware Deep Reinforcement Learning

T0 review · 5 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read ProCAVE claims that predictive, preference-aware DRL control across the full video lifecycle outperforms reactive edge caching, improving byte hit rate by about 56 percent, cutting backhaul traffic by 21 percent, and reducing delayed startu

desk verdict Plausible DRL integration for edge caching with a clean architecture, but an underspecified cache action mapping and thin statistics make the headline gains unverifiable. read the letter →

arxiv 2608.03313 v1 pith:QU6V3URB submitted 2026-08-04 cs.NI cs.MMeess.IV

Yeganeh Chatri , Behzad Akbari , Foad Ghaderi , Pejman Goudarzi This is my paper · ORCID
classification cs.NIcs.MMeess.IV
keywords edgecachingadaptivebitratestreamingdeepreinforcementlearningbandwidthpredictionTransformerqualityofexperiencebytehitratefull-lifecycle
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

ProCAVE is a proposed edge-caching framework that tries to show that video delivery improves when bandwidth prediction, bitrate selection, and cache admission/eviction are controlled together by learned agents instead of by reactive rules. The paper builds on FlyCache's four-region full-lifecycle cache and adds a Transformer that forecasts throughput four to twelve seconds ahead, a PPO policy for adaptive bitrate selection, and a DDPG controller that outputs continuous cache actions from a 220-dimensional global state. Evaluated on MovieLens preference traces and Ghent 4G bandwidth traces, the framework is reported to raise byte hit rate from about 0.45 to 0.70, lower backhaul traffic from 1.4e8 to 1.1e8 bytes, and cut delayed startup rate from 0.60 to 0.45 over 250 simulated steps. The paper concludes that predictive, cross-layer DRL orchestration is the key to efficient and user-centric edge video delivery.

What carries the argument

The load-bearing mechanism is the DDPG-based cache controller acting on a 220-dimensional global state $Z_t$ built from mean user preference $\bar{p}$, predicted bandwidth $\bar{\theta}$, cache utilization, average buffer, stall risk, and feature vectors of ten candidate videos. It outputs a continuous action vector $a_t \in [-1,1]^{10}$ in which each component is supposed to determine eviction, demotion, no-op, admission, or prefetching across the Static Buffer, Maintain Space, Dynamic Buffer, and Victim Space. The Transformer forecast at +4, +8, and +12 seconds feeds the ABR agent, and ABR outcomes feed the cache reward, closing the loop.

What would settle it

Reproduce the 250-step simulation with an explicit discretization of $a_t$ into the five named operations, using any reasonable fixed thresholds, and check whether the reported byte hit rate, backhaul, and delayed-startup gains survive; if the gains disappear under a fair discretization, the central claim fails.

Watch

Extended reading notes

Core claim

The central discovery is that coupling three learned controllers—a lightweight Transformer bandwidth predictor, a PPO-based ABR agent, and a DDPG-based cache controller—lets an edge cache anticipate short-term network changes and user preference drift instead of reacting to them. The paper reports that this integrated design improves byte hit rate by roughly 56%, cuts backhaul traffic by 21%, and lowers delayed startup rate by 25% relative to FlyCache, and that under cache scaling from 100 to 350 segments byte hit rate grows monotonically to about 0.34, an order of magnitude above FlyCache's 0.03, while FlyCache peaks early and degrades. These numbers are the paper's evidence that full-lifec

Load-bearing premise

The paper assumes that each real-valued cache action in $[-1,1]$ maps in a well-defined way to a concrete cache operation, but the thresholds and rules for that mapping are never specified.

Editorial extensions

If this is right

  • Edge video systems that treat bandwidth prediction, bitrate adaptation, and cache control as one control problem can expect higher byte hit rates and lower backhaul than reactive designs.
  • The reported cache-scaling monotonicity implies the framework can use added edge capacity effectively, whereas reactive baselines saturate or degrade.
  • The 25% drop in delayed startup rate suggests proactive prefix placement based on throughput forecasts can smooth playback startup under fluctuating wireless conditions.
  • Because cache actions are preference-aware and updated online, the framework can track drifting user interests without rerunning offline popularity analysis.

Reading between the lines

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

  • The unspecified action discretization means the quantitative results should be read as conditional on an implementation detail the paper does not expose; specifying it cleanly would make the comparisons reproducible.
  • ProCAVE's Transformer is trained on a single 531-second trace from one network; a natural extension is testing forecast-driven caching across longer, multi-day or multi-network traces to see whether proactive gains generalize.
  • The same three-agent structure could extend to live streaming or 360-degree video, where throughput prediction and viewport-aware prefetching interact strongly with cache admission.
  • A federated version, which the paper lists as future work, would need privacy-preserving preference aggregation before the global preference state $Z_t$ can be computed across edge nodes.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 4 minor

Summary. The paper proposes ProCAVE, a multi-agent DRL framework for edge video caching that couples a Transformer-based bandwidth predictor, a PPO-based ABR controller, and a DDPG-based cache controller. The claimed contribution is to extend the reactive FlyCache architecture into a predictive, coordinated full-lifecycle system. Evaluation on MovieLens preference traces and Ghent 4G throughput traces reports a ~56% byte hit rate gain, ~21% backhaul traffic reduction, and ~25% delayed startup rate reduction relative to FlyCache over 250 simulated steps, plus scalability gains under cache capacity variation.

Significance. If the reported gains are reproducible and the proposed control architecture is fully specified, the work would be a useful step toward jointly optimizing ABR and edge caching under realistic network dynamics. The paper uses real datasets, provides algorithms for the three learned components, and reports simulation parameters in Table II. However, the current manuscript does not specify a load-bearing implementation detail—the discretization of the DDPG continuous action into concrete cache operations—and the experimental validation lacks statistical grounding. As written, the central performance claims are not verifiable.

major comments (5)
  1. [Section III-B, Eq. (2); Algorithm 1, line 9] The mapping from the continuous action vector a_t in [-1,1]^10 to concrete cache operations is never defined. The text says each a_t,i "determines eviction, demotion, no-op, admission, or prefetching," but gives no thresholds, per-region rules, tie handling, or ordering. Since the cache is a discrete system, every continuous value must be discretized. Different reasonable mappings will produce different cache dynamics, and the reported BHR/backhaul/DSR numbers all depend on this mapping. Please specify the exact mapping and provide a sensitivity or ablation analysis over its parameters.
  2. [Section IV, Figures 2 and 3] All performance claims are based on single trajectories over 250 time steps with no error bars, confidence intervals, or multiple random seeds. The reported improvements (56% BHR, 21% backhaul, 25% DSR) could be within run-to-run variance, especially for DRL training, which is highly sensitive to initialization and exploration noise. Report mean and variance across at least 5-10 seeds, and state whether the plotted curves are representative runs or averages.
  3. [Section IV, datasets] The evaluation protocol is under-specified. The Transformer is trained on 510 overlapping samples from a single 531 s Ghent trace, but the paper does not state how the evaluation trace is split from training or whether the DRL agents are trained and evaluated on the same 250-step window. If the same trace is used for training and evaluation, the reported gains may include data leakage. Provide explicit train/validation/test splits and clarify the relationship between the predictor training window and the simulation horizon.
  4. [Section III-B and Table II] There is a direct inconsistency in the bitrate set. Section III-B defines B = {800, 1500, 3000, 6000} Kbps (four levels), while Table II lists "Bitrate levels: 6 levels (300–4300 kbps)." The abstract and experiments refer to 6 levels. This discrepancy affects the ABR action space and the comparability of the reported QoE/DSR results. Please correct and ensure the paper uses one consistent bitrate ladder throughout.
  5. [Section III-C, Eq. (4); Section IV, dataset description] Equation (4) says the Transformer maps a 20-second throughput history to a 3-step forecast, but the experimental description states the predictor input is a (10,1) window (10 seconds). The system is otherwise based on 4-second segments, so +4/+8/+12 s forecasts correspond to one, two, and three steps, which is consistent, but the input horizon is not. Clarify the exact input length and ensure the algorithm description matches the implementation.
minor comments (4)
  1. [Throughout] The paper contains formatting artifacts: "ProCA VE" is inconsistently spaced, the header reads "ProCA VE" and "A ware", and author names are split across lines. Please copyedit.
  2. [Section III-D, Eq. (3)] The objective weights beta_1, beta_2, beta_3 in Eq. (3) are never assigned values in Table II or the text. These weights control the trade-off among BHR, stall ratio, and backhaul, and the reported behavior is likely sensitive to them. Report the values or a sensitivity study.
  3. [Figure 3] The delayed startup rate for FlyCache is reported to exceed 1.0 (approximately 1.35 at 300 segments), which is impossible for a rate unless the metric is defined differently or includes multiple events per session. Please define DSR precisely and verify the plotted values.
  4. [Algorithm 4] The DDPG update is standard, but the use of the same R_t from Eq. (3) as the reward is not explicitly connected to the cache agent's reward. Clarify whether the cache reward is exactly the global objective or a decomposed per-decision reward.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: reported gains are empirical simulation outcomes against an external baseline, not derivations from fitted constants or self-citations.

full rationale

The paper's derivation chain is empirical, not definitional. ProCAVE is a multi-agent DRL framework trained on MovieLens/Ghent data and then compared against FlyCache, an external baseline. No equation in the paper defines the target result in terms of its own inputs. The reward in Eq. (3) includes BHR, stall, and backhaul, and the evaluation reports BHR, backhaul, and DSR; this is a standard RL setup where the optimization objective and evaluation metrics overlap, but the learned policy's realized performance is not logically forced by the reward definition—the DDPG/PPO training could in principle fail to improve over FlyCache. The self-citation [21] appears only in the related-work discussion and is not load-bearing. Two substantive concerns exist but are not circularity: (1) Section III-B and Algorithm 1 never specify how the continuous DDPG action a_t ∈ [-1,1]^10 is mapped to discrete cache operations (eviction, demotion, no-op, admission, prefetching) across the four cache regions—an omitted procedural definition that hurts reproducibility and verifiability, but does not make the claim equivalent to its inputs by construction. (2) Section IV-A says a 531 s Ghent trace is used to train the Transformer predictor and that the same Ghent 4G dataset family models bandwidth in the simulation, but no explicit held-out split is described—a data-leakage/overfitting risk, not circular reasoning. Because the central comparison is against an external baseline and the reported numbers are simulation outputs rather than fitted parameters renamed as predictions, no significant circularity is present.

Assumptions & free parameters 3 free parameters · 4 assumptions · 1 invented entities

The central claim (performance improvement over FlyCache) rests on the fidelity of the simulation, the representativeness of the traces, the precision of the preference model, and the correctness of the baseline. The unspecified objective weights and action mapping are free parameters that directly affect the results. No new entities are introduced.

free parameters (3)
  • Objective weights beta_1, beta_2, beta_3 in Eq. (3)
    These weights balance byte hit rate, stall ratio, and backhaul traffic in the reward. The paper does not give their values. If they are hand-tuned to favor the reported metrics, the comparison to FlyCache is tilted. They are free parameters that shape the learned behavior.
  • Transformer architecture hyperparameters (layers L, hidden size, heads)
    Algorithm 2 shows a generic encoder with L layers and MHSA, but no concrete size or layer count is given. These choices affect forecast accuracy and are not specified, so the predictor cannot be reproduced.
  • DDPG action-to-cache-operation thresholds
    The paper says a_t,i in [-1,1] determines eviction, demotion, no-op, admission, or prefetching, but does not define how each value is interpreted. These thresholds are essential free parameters for the cache controller.
assumptions (4)
  • domain assumption The simulation environment faithfully represents real-world edge video streaming dynamics.
    Section IV evaluates on a custom simulation with MovieLens and Ghent 4G traces, but the simulator internals (user behavior, network model, cache region sizes) are not described. If the simulation is too simplified, the reported improvements may not transfer to real deployments.
  • domain assumption User preferences are adequately captured by a 19-dimensional genre vector updated via EMA.
    Eq. (1) defines the preference update, but there is no validation that this model predicts real viewing behavior. The cache agent's usefulness depends on these preference signals being informative.
  • domain assumption The Ghent 4G traces are representative of the wireless conditions in which the system would be deployed.
    The Transformer predictor is trained on a 531 s segment of one trace and the system is evaluated on the same trace family. The assumption is that this small sample generalizes to other networks and time periods.
  • domain assumption FlyCache is implemented correctly and is a fair baseline.
    The comparison in Section IV relies on the FlyCache baseline being a reasonable, well-tuned representative of the prior art. If the baseline is under-performing or misconfigured, the reported gains are inflated.
invented entities (1)
  • None
    purpose: The paper introduces no new physical or computational entities.
    ProCAVE is a software framework composed of existing techniques (Transformer, PPO, DDPG, cache regions). No new particles, forces, or abstract objects are postulated.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ProCAVE: A Self-Adaptive, Full-Lifecycle Edge Caching Framework for Video Streaming via Predictive Bandwidth Estimation and Preference-Aware Deep Reinforcement Learning." pith.science (2026). https://pith.science/paper/QU6V3URB

@misc{pith2026260803313,
  author       = {Pith},
  title        = {Pith review of: ProCAVE: A Self-Adaptive, Full-Lifecycle Edge Caching Framework for Video Streaming via Predictive Bandwidth Estimation and Preference-Aware Deep Reinforcement Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QU6V3URB}},
  note         = {Machine review of arXiv:2608.03313}
}
read the original abstract

The growing demand for mobile video streaming requires edge delivery systems that adapt efficiently to rapid network fluctuations and diverse user preferences. Existing approaches such as FlyCache rely on reactive ABR heuristics and loosely coupled cache policies, limiting their responsiveness and coordination under real-world wireless dynamics. We propose ProCAVE (Proactive Caching with Adaptive Video Experience), a self-adaptive DRL-based framework that unifies predictive bandwidth modeling, proactive bitrate selection, and preference-aware cache control. ProCAVE employs: (i) a lightweight Transformer for short-term throughput forecasting; (ii) a PPO-driven ABR agent; and (iii) a DDPG-based continuous cache controller operating on a high-dimensional global state. Experiments using MovieLens preference traces and Ghent 4G bandwidth measurements show that ProCAVE improves byte hit rate, reduces backhaul load, and enhances QoE compared with FlyCache and other baselines. These results highlight the benefits of predictive, DRL-coordinated control for efficient and user-centric edge video delivery.

Figures

Figures reproduced from arXiv: 2608.03313 by the authors.

Figure 1
Figure 1. System Model TABLE I: Notation Table Symbol Meaning Users, Videos, and Preferences u, v User / video index Sv Segment set of video v fv Feature vector of video v pu t User preference vector at time t ρu,v t Playback progress for user u on v Network and ABR Variables θu(t) Measured throughput Hu t Throughput history window θ u t Predicted bandwidth (Transformer) au t ABR action selected by PPO ru t ABR reward Cache … view at source ↗
Figure 2
Figure 2. The Trend of Byte Hit Rate, Backhaul Traffi [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Impact of Cache Capacity on Byte Hit Rate, Backhaul Traffi [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

23 extracted references · 21 canonical work pages

  1. [1]

    Flycache: Recommendation-driven edge caching architecture for full life cycle of video streaming,

    S. Cao, Q. Zheng, Z. Zhan, Y . Y ang, H. Lv, D. Zheng, and W. Zhang, “Flycache: Recommendation-driven edge caching architecture for full life cycle of video streaming,”Digital Communications and Networks, vol. 11, no. 4, pp. 961–974, 2025. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S235286482500001X

    work page 2025

  2. [2]

    Segment prefetch- ing at the edge for adaptive video streaming,

    J. Aguilar-Armijo, C. Timmerer, and H. Hellwagner, “Segment prefetch- ing at the edge for adaptive video streaming,” in2022 18th International Conference on Wireless and Mobile Computing, Networking and Com- munications (WiMob), 2022, pp. 339–344

    work page 2022

  3. [3]

    Space: Segment prefetching and caching at the edge for adaptive video streaming,

    ——, “Space: Segment prefetching and caching at the edge for adaptive video streaming,”IEEE Access, vol. 11, pp. 21 783–21 798, 2023

    work page 2023

  4. [4]

    Cd-lwte: Cost- and delay-aware light-weight transcoding at the edge,

    A. Erfanian, H. Amirpour, F. Tashtarian, C. Timmerer, and H. Hell- wagner, “Cd-lwte: Cost- and delay-aware light-weight transcoding at the edge,”IEEE Transactions on Network and Service Management, vol. 20, no. 3, pp. 3104–3118, 2023

    work page 2023

  5. [5]

    Collaborative video caching in the edge network using deep reinforcement learning,

    A. Lekharu, P . Gupta, A. Sur, and M. Patra, “Collaborative video caching in the edge network using deep reinforcement learning,”ACM Trans. Internet Things, vol. 5, no. 3, Jun. 2024. [Online]. Available: https://doi.org/10.1145/3664613

    work page doi:10.1145/3664613 2024

  6. [6]

    Stochastic game based cooperative alternating q-learning caching in dynamic d2d networks,

    T. Zhang, X. Fang, Z. Wang, Y . Liu, and A. Nallanathan, “Stochastic game based cooperative alternating q-learning caching in dynamic d2d networks,”IEEE Transactions on V ehicular Technology, vol. 70, no. 12, pp. 13 255–13 269, 2021

    work page 2021

  7. [7]

    Joint cache placement and delivery design using reinforcement learning for cel- lular networks,

    M. Amidzadeh, H. Al-Tous, O. Tirkkonen, and J. Zhang, “Joint cache placement and delivery design using reinforcement learning for cel- lular networks,” in2021 IEEE 93rd V ehicular Technology Conference (VTC2021-Spring), 2021, pp. 1–6

    work page 2021

  8. [8]

    Cache placement optimization in mobile edge computing networks with unaware environ- ment—an extended multi-armed bandit approach,

    Y . Han, L. Ai, R. Wang, J. Wu, D. Liu, and H. Ren, “Cache placement optimization in mobile edge computing networks with unaware environ- ment—an extended multi-armed bandit approach,”IEEE Transactions on Wireless Communications, vol. 20, no. 12, pp. 8119–8133, 2021

    work page 2021

Show all 23 references
  1. [9]

    Optimal caching policy for d2d assisted cellular networks with different cache size devices,

    N. Abdolkhani, M. Eslami, J. Haghighat, and W. Hamouda, “Optimal caching policy for d2d assisted cellular networks with different cache size devices,”IEEE Access, vol. 10, pp. 99 353–99 360, 2022. Fig. 2: The Trend of Byte Hit Rate, Backhaul Traffic, and Delayed Startup Rate Ov...

    2022

  2. [10]

    Comparison of caching replacement policies in changing the number of interest packets on named data networks using mininet-ndn,

    F. Zharfan, L. D. Hasnaa, R. Muldina Negara, and N. R. Syambas, “Comparison of caching replacement policies in changing the number of interest packets on named data networks using mininet-ndn,” in2021 15th International Conference on Telecommunication Systems, Services, and Ap...

    2021

  3. [11]

    Mcucoder: Adaptive bitrate learned video compression for iot devices,

    A. Hojjat, J. Haberer, and O. Landsiedel, “Mcucoder: Adaptive bitrate learned video compression for iot devices,” 2024. [Online]. Available: https://arxiv.org/abs/2411.19442

    2024 arXiv

  4. [12]

    A two-stage deep reinforcement learning framework for mec-enabled adaptive 360-degree video streaming,

    S. Bi, H. Chen, X. Li, S. Wang, Y . Wu, and L. Qian, “A two-stage deep reinforcement learning framework for mec-enabled adaptive 360-degree video streaming,”IEEE Transactions on Mobile Computing, vol. 23, no. 12, pp. 14 313–14 329, 2024

    2024

  5. [13]

    Trim- stream: Adaptive realtime video streaming through intelligent frame retrospection in adverse network conditions,

    D. Zhang, L. Wei, K. Shen, H. Zhu, D. Wang, and F. Wang, “Trim- stream: Adaptive realtime video streaming through intelligent frame retrospection in adverse network conditions,”IEEE Transactions on Mobile Computing, vol. 23, no. 12, pp. 11 240–11 252, 2024

    2024

  6. [14]

    Qoe-driven cross-layer bitrate allocation approach for mec-supported adaptive video streaming,

    Y . F. Y eznabad, M. Helfert, and G.-M. Muntean, “Qoe-driven cross-layer bitrate allocation approach for mec-supported adaptive video streaming,” IEEE Transactions on Network and Service Management, vol. 21, no. 6, pp. 6857–6874, 2024

    2024

  7. [15]

    Task-oriented video compressive streaming for real-time semantic segmentation,

    X. Xiao, Y . Zuo, M. Y an, W. Wang, J. He, and Q. Zhang, “Task-oriented video compressive streaming for real-time semantic segmentation,”IEEE Transactions on Mobile Computing, vol. 23, no. 12, pp. 14 396–14 413, 2024

    2024

  8. [16]

    Cloud–edge learning for adaptive video streaming in b5g internet of things systems,

    H. Zhan, L. Fan, C. Li, X. Lei, and F. Li, “Cloud–edge learning for adaptive video streaming in b5g internet of things systems,”IEEE Internet of Things Journal, vol. 11, no. 24, pp. 40 140–40 148, 2024

    2024

  9. [17]

    Qoe- fairness-aware bandwidth allocation design for mec-assisted abr video transmission,

    A. Xiao, S. Wu, Y . Ou, N. Chen, C. Jiang, and W. Zhang, “Qoe- fairness-aware bandwidth allocation design for mec-assisted abr video transmission,”IEEE Transactions on Network and Service Management, vol. 22, no. 1, pp. 499–515, 2025

    2025

  10. [18]

    Acdv: Adaptive content delivery for vehicular digital twin networks,

    J. Zheng, T. H. Luan, G. Li, Z. Yin, Y . Wu, and M. Dong, “Acdv: Adaptive content delivery for vehicular digital twin networks,”IEEE Transactions on V ehicular Technology, vol. 74, no. 5, pp. 7084–7098, 2025

    2025

  11. [19]

    Enhancing qoe through adaptive bitrate allocation in collaborative mec-enabled wireless networks,

    Y . F. Y eznabad, M. Helfert, and G.-M. Muntean, “Enhancing qoe through adaptive bitrate allocation in collaborative mec-enabled wireless networks,”IEEE Transactions on V ehicular Technology, vol. 74, no. 6, pp. 9491–9505, 2025

    2025

  12. [20]

    How2compress: Scalable and efficient edge video analytics via adaptive granular video compression,

    Y . Wu, T.-T. Nguyen, L. Liebe, Q. Tau, P . E. Campos, J. Cheng, and D. Lee, “How2compress: Scalable and efficient edge video analytics via adaptive granular video compression,” 2025. [Online]. Available: https://arxiv.org/abs/2510.18409

    2025

  13. [21]

    Investigating the impact of chunk time on video streaming performance in tcp-based networks,

    Y . Chatri, B. Akbari, and P . Goudarzi, “Investigating the impact of chunk time on video streaming performance in tcp-based networks,” in2025 29th International Computer Conference, Computer Society of Iran (CSICC), 2025, pp. 1–5

    2025

  14. [22]

    The movielens datasets: History and context,

    F. M. Harper and J. A. Konstan, “The movielens datasets: History and context,”ACM Trans. Interact. Intell. Syst., vol. 5, no. 4, Dec. 2015. [Online]. Available: https://doi.org/10.1145/2827872

    2015 doi

  15. [23]

    HTTP/2-based adaptive streaming of HEVC video over 4G/LTE networks,

    J. van der Hooft, S. Petrangeli, T. Wauters, R. Huysegems, P . Rondao Al- face, T. Bostoen, and F. De Turck, “HTTP/2-based adaptive streaming of HEVC video over 4G/LTE networks,”IEEE Communications Letters, vol. 20, no. 11, pp. 2177–2180, 2016

    2016

Pith tools

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