Pith. sign in

REVIEW 3 major objections 4 minor 34 references

Intelligent Backhaul Link Selection for Traffic Offloading in B5G Networks

T0 review · 3 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read A small Double Deep Q-Network can select per-slice backhaul links—wired, IAB relay, or satellite—and train to 97 percent of the optimal reward in about 21 simulated episodes.

desk verdict A clean, narrowly-scoped DDQN simulation study for backhaul link selection; the 20-episode convergence claim rests on one best run and should be treated as provisional. read the letter →

arxiv 2501.09123 v1 pith:RQZ2WKRU submitted 2025-01-15 cs.NI

António J. Morgado , Firooz B. Saghezchi , Pablo Fondo-Ferreiro , Felipe Gil-Castiñeira , Jonathan Rodriguez This is my paper · ORCID
classification cs.NI
keywords integratedaccessandbackhaulnetworkslicingdeepreinforcementlearningsatellitetrafficoffloadingQoS-awarelinkselectionDoubleQ-Networktopologyadaptation
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

This paper tries to establish that a small deep-reinforcement-learning agent can decide, every fifteen minutes, which backhaul link—wired, wireless relay through a neighbour base station, or satellite—should carry each of several 5G network slices when a base station's wired backhaul is saturated. The agent must satisfy each slice's throughput and latency requirements. In the authors' simulations, a Double Deep Q-Network with one fully connected hidden layer of 80 neurons reaches 97 percent of the exhaustive-search training optimum in 21 episodes, and a testing reward of 70 out of the 75 achievable. If this holds outside the simulator, operators could adapt backhaul topology on demand with a very small, quickly trained model instead of solving an explicit optimization each time.

What carries the argument

The load-bearing machinery is the Double Deep Q-Network (DDQN) critic: a fully connected neural network whose input concatenates a 36-parameter observation with an 8-action vector, and whose output estimates $Q^*(s,a)$. DDQN's separation of an online critic (action selection) from a target critic (value evaluation) prevents the overoptimistic targets of plain DQN. The environment that this critic learns is an SDN-based simulator that computes path latency as the sum of link latencies plus waiting time at each interface modelled as an M/D/1 queue. The reward ($+1$ per successful slice allocation, $0$ otherwise) makes the episode reward equal to the number of QoS-compliant allocations, so the exhaustive-search optimum is directly comparable.

What would settle it

Run the same one-hidden-layer-80-neuron DDQN agent in a packet-level network simulator or a small IAB testbed where latency comes from actual queueing dynamics, and compare its per-slice allocation success against the exhaustive-search optimum over the same traffic traces; the central claim fails if the agent needs substantially more than 21 episodes to reach 97 percent of the optimum, or if its success rate on unseen intervals falls well below the reported 70 out of 75.

Watch

Extended reading notes

Core claim

The central claim is that per-slice backhaul link selection with integrated access and backhaul (IAB) plus satellite connectivity reduces to a small discrete-action reinforcement-learning problem that a Double Deep Q-Network solves quickly. The action is the choice of one of eight links per slice; the observation is a 36-parameter vector of slice QoS requirements, available link bandwidths, and path latencies; the reward is +1 for a QoS-compliant allocation and 0 otherwise. With four slices and one congested base station, the best trade-off critic—one hidden layer, 80 ReLU neurons, learning rate $10^{-3}$—collects an average training reward that reaches 97 percent of the optimum training reward of 262 after 21 episodes, and scores 70/75 in testing. The authors conclude that around 20–50 episodes are enough to train such an agent.

Load-bearing premise

The load-bearing premise is that path latency can be computed as link delays plus waiting time in an M/D/1 queue—a standard model with constant service times—and both the agent's reward and the exhaustive-search optimum depend on that computed number.

Editorial extensions

If this is right

  • With four slices on one congested base station, the trained agent matches the exhaustive-search optimum for most tested time steps, so dynamic backhaul reconfiguration can be automated rather than recomputed from scratch each interval.
  • The same agent learns to avoid the satellite link for slices needing under 100 ms latency and the wireless relays for the 1 ms uRLLC slice, so the learned policy respects slice QoS classes.
  • Training time of around 21 episodes implies the policy can be retrained quickly when traffic profiles or topology change, since the observation and action design stay fixed.
  • Scaling to multi-hop backhaul is not supported by the current design; the authors point to multi-agent DRL as the next step for larger networks.

Reading between the lines

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

  • If the 21-episode convergence transfers to real IAB nodes, a practical controller could re-train per site on a rolling basis with only a few hours of simulated or observed traffic, rather than relying on pre-trained models.
  • The exhaustive-search optimum already accounts for the same M/D/1 latency model, so the paper demonstrates RL matching a model-based oracle, not necessarily matching real networks; the gap to deployment depends on how realistic that queue model is.
  • A natural testable extension is to replace the synthetic slice profiles with measured 5G traffic traces and compare the agent's allocation success against a heuristic that always picks the least-loaded qualified link.
  • The action space's assumption that only BS1 is congested may be the easiest place to generalize; a multi-agent formulation with per-base-station congestion would test whether the small-critic result survives interaction effects.
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.

Referee Report

3 major / 4 minor

Summary. The paper proposes a Double Deep Q-Network (DDQN) agent that selects, every 15 minutes and for each of up to four network slices at a congested base station, a backhaul link from a pool of wired, IAB, and satellite options. The reward is +1 if the selected link meets the slice's throughput and latency requirements, as computed by an SDN simulator with M/D/1 queueing delays, and 0 otherwise. The authors compute an exhaustive-search maximum episode reward for the training, validation, and test sets, and sweep critic architectures (1, 3, and 5 hidden layers; 8--256 neurons). The selected critic, with one hidden layer and 80 neurons and learning rate 10^-3, reaches 97.0% of the training optimum in 21 episodes and a test reward of 70 out of 75 achievable (Section IV.C.2, Table 10, Fig. 12), which motivates the abstract's claim that around 20 episodes are sufficient to train the agent.

Significance. The problem is timely and practically relevant: dynamic per-slice backhaul selection that combines IAB and satellite links is a plausible component of B5G/6G network management. The paper has several genuine strengths: the reward is defined by simulator-computed QoS feasibility rather than by the agent's own outputs; an exhaustive-search upper bound is computed for the training/validation/test sets (262, 33, and 75); and model selection is informed by cross-validation. If the 21-episode convergence result were robust across random seeds, it would be a useful data point for lightweight DRL in backhaul reconfiguration. However, the headline quantitative claim currently rests on a single training run, and the test set is used during model selection, so the reported generalization evidence is weaker than the abstract suggests. The absence of code release also prevents independent verification of the simulation results.

major comments (3)
  1. [IV.C.2 (abstract, Table 10, Fig. 12)] The headline claim that 'around 20 episodes are sufficient' is supported by a single training run: the selected critic (one hidden layer, 80 neurons, learning rate 10^-3) reached the 97.0% threshold in 21 episodes, and no multiple-seed statistics or variance measures are reported anywhere. In deep RL, the number of episodes to reach a reward threshold is highly sensitive to random initialization, epsilon-greedy exploration order, and experience-replay sampling, and the 10-episode moving-average stopping rule can be crossed transiently. A single run cannot establish that 20 episodes are 'sufficient' even in this simulated environment. Please report the distribution of convergence episodes over at least 10--20 random seeds and state a formal sufficiency criterion (e.g., probability at least 0.9 of reaching the threshold by episode K).
  2. [IV.C.2 (Tables 9--10)] The model-selection protocol uses the testing results to choose among configurations that already passed cross-validation: the paper states that, from the configurations achieving the best cross-validation, the authors select 'the configurations with the best testing results,' and then choose one configuration. The reported test reward of 70/75 for the selected critic is therefore optimistic because the test set has been used during selection. To support a generalization claim, the test set should be used only once after all model selection is complete, or a nested cross-validation/selection procedure should be reported.
  3. [III.F and IV.C.1] The latency of every candidate path is computed with an M/D/1 queueing model, and a link at exactly full capacity is treated as having infinite latency. This latency determines both the agent's reward and the exhaustive-search upper bounds (262, 33, 75), so the headline convergence result is conditional on this specific queueing assumption. To assess transferability to real networks, please report sensitivity of the learning curves and the upper bounds to alternative queueing models (e.g., M/M/1 or finite-buffer approximations), or explicitly state that the claim is limited to the M/D/1 simulator.
minor comments (4)
  1. [Abstract and Section I] The phrase 'Fifth Generation (5G) mobile networks considers' should be 'consider' for subject-verb agreement; similar wording appears in the Introduction.
  2. [Section III.G] Please clarify whether each training episode is a full pass over the same 67 training timesteps and how epsilon is reset between episodes; Algorithm 1 only states that epsilon decays each timestep, and the current text does not specify the reset policy.
  3. [Fig. 13 and Table 10] The text says the agent 'learns rapidly in just five episodes' while Table 10 reports that the same critic requires 21 episodes to reach the 97% training-reward threshold; please reconcile these statements or clarify that Fig. 13 illustrates a different metric (allocated throughput and QoS failures).
  4. [Algorithm 1] Step 3.2.11 writes the epsilon update as ε = ε·(1-εdecay), which implies a geometric decay each timestep; if a linear decay schedule is intended, the notation should be changed, and the values of ε, εdecay, γ, N, M, and C used in the simulations should be explicitly listed.

Circularity Check

0 steps flagged · score 1.0 of 10

No circular reasoning: the DDQN convergence result is benchmarked against an independent exhaustive search on the simulator, and the only self-citation ([34]) is descriptive, not load-bearing.

full rationale

The paper does not derive any result from an equation that contains the result itself. The training objective and the benchmark are both computed by the simulator: the reward is +1 if the selected backhaul link meets the slice's throughput and latency constraints (Sec. III.D), and the claimed optimum (262 training, 33 cross-validation, 75 testing) is obtained by exhaustive search over the same feasibility rules (Sec. III.G and IV.C.2). The agent's ability to reach 97% of that optimum in 21 episodes is therefore an empirical performance claim with real content; the random-strategy baseline in Fig. 12 shows the target is not trivially reachable. The only self-citation, [34], appears in Section IV.D as a descriptive remark about the first, throughput-only scenario and is not used to justify the satellite-scenario critic or the '20-50 episodes' conclusion. The choice of the final critic from Tables 9-10 uses both cross-validation and testing rewards, which is a model-selection/data-leakage concern (and a statistical robustness concern because no multiple-seed averages are reported), but it is not a circular reduction: the reported training episode count is measured during training, and the test reward is not a fitted parameter renamed as a prediction. The M/D/1 latency model is an external modeling assumption [33], relevant to transferability but not to circularity.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central simulation result relies on the simulator's delay model and on hand-picked hyperparameters; the paper provides no independent evidence that the M/D/1 latency assumption or the synthetic traffic profiles hold in real networks. No new physical or conceptual entities are introduced.

free parameters (3)
  • learning_rate = 10^-3 (satellite scenario), 10^-4 (no-satellite scenario)
    Tuned manually across simulations (Fig. 7); directly affects the reported convergence episode count.
  • critic_architecture = 1 hidden layer, 80 neurons (satellite); 3 hidden layers, 40 neurons (no satellite)
    Selected as the best trade-off among configurations in Tables 9 and 10; the abstract highlights only the 1x80 case.
  • convergence_threshold = 97.5% of optimum (no satellite), 97.0% of optimum (satellite)
    Hand-chosen stopping rule; determines the reported number of episodes to convergence.
assumptions (4)
  • domain assumption The M/D/1 queueing model accurately estimates packet waiting time at each network interface and yields the latency used in QoS checks.
    Invoked in Section III.F to compute path latency and again in Section IV.C.1 where a link at full capacity is treated as having infinite latency. The exhaustive-search optimum and the agent reward both depend on this model.
  • domain assumption None of the wireless or satellite backhaul links interfere with incumbent networks operating in the same bands.
    Stated in Section III.A.1; removes a real-world constraint such as interference or spectrum sharing that would change link capacity and therefore the learned policy.
  • domain assumption For in-band IAB operation, no simultaneous transmit/receive limitation or directional transmission scheduling is modeled beyond per-link capacity.
    Section II.B notes the half-duplex constraint of in-band IAB, but the simulator only uses capacity and latency tuples and does not model this constraint in the action feasibility checks.
  • standard math Standard DRL convergence machinery, including DDQN, experience replay, and target networks, stabilizes training as claimed in the cited literature.
    Assumed from references [25]-[27]; not proved in this paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Intelligent Backhaul Link Selection for Traffic Offloading in B5G Networks." pith.science (2026). https://pith.science/paper/RQZ2WKRU

@misc{pith2026250109123,
  author       = {Pith},
  title        = {Pith review of: Intelligent Backhaul Link Selection for Traffic Offloading in B5G Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RQZ2WKRU}},
  note         = {Machine review of arXiv:2501.09123}
}
read the original abstract

Fifth Generation (5G) mobile networks considers an expansive set of heterogeneous services with stringent Quality of Service (QoS) requirements, and traffic demand with inherent spatial-temporal distribution, which places the backhaul network deployment under potential strain. In this paper, we propose to harness network slicing, Integrated Access and Backhaul (IAB) technology coupled with satellite connectivity to build a dynamic wireless backhaul network that can provide additional backhaul capacity to the base stations on demand when the wired backhaul link is temporarily out of capacity. To construct the network design, Deep Reinforcement Learning (DRL) models are used to select, for each network slice of the congested base station, an appropriate backhaul link from the pool of available IAB and satellite links that meets the QoS requirements (i.e., throughput and latency) of the slice. Simulation results show that around 20 episodes are sufficient to train a Double Deep Q-Network (DDQN) agent, with one fully-connected hidden layer and Rectified Linear Unit (ReLU) activation function, that adjusts the topology of the backhaul network.

Figures

Figures reproduced from arXiv: 2501.09123 by the authors.

Figure 1
Figure 1. FIGURE 1 [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. FIGURE 2 [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. FIGURE 3 [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: FIGURE 4 [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: FIGURE 5 [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: FIGURE 6 [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: FIGURE 7 [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: FIGURE 8 [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: FIGURE 9 [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 10
Figure 10. Figure 10: FIGURE 10 [PITH_FULL_IMAGE:figures/full_fig_p009_10.png]
Figure 11
Figure 11. Figure 11: illustrates the training episode reward obtained by the DDQN agent when the critic function is composed of 3 fully connected hidden layers with 40 neurons per hidden layer. The figure contrasts the result against 1) the optimum strategy (i.e., reward of 201) obtained …
Figure 13
Figure 13. Figure 13: , considering the traffic of all slices both in UL and DL. The figure shows the attained backhaul throughput and the failures of the agent due to the violation of QoS requirements of the slices (delay and throughput) over different training episodes. Analyzing [PITH_…
Figure 12
Figure 12. Figure 12: presents the training episode reward obtained by the DDQN agent when its critic function is composed of one fully connected hidden layer with 80 neurons. The figure contrasts the result against 1) the optimum strategy (i.e., reward of 262) obtained by the exhaustive s…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 27 canonical work pages

  1. [34]

    A. J. Morgado, F. B. Saghezchi, P . Fondo-Ferreiro, F. Gil-Castiñeira, M. Papaioannou, K. Ramantas, and J. Rodriguez, ‘‘Deep reinforcement learning for backhaul link selection for network slices in IAB networks,’’ in Proc. IEEE Global Commun. Conf., Kuala Lumpur, Malaysia, Dec. 2023, pp. 1–6. ANTÓNIO J. MORGADO received the degree in electronic and teleco...

    work page 2023

  2. [1]

    Ericsson Mobility Report, Ericsson, Stockholm, Sweden, Jun. 2023

    work page 2023

  3. [2]

    Polese, M

    M. Polese, M. Giordani, T. Zugno, A. Roy, S. Goyal, D. Castor, and M. Zorzi, ‘‘Integrated access and backhaul in 5G mmWave networks: Potential and challenges,’’IEEE Commun. Mag., vol. 58, no. 3, pp. 62–68, Mar. 2020

    work page 2020

  4. [3]

    M. M. Sande, M. C. Hlophe, and B. T. Maharaj, ‘‘Access and radio resource management for IAB networks using deep reinforcement learning,’’ IEEE Access, vol. 9, pp. 114218–114234, 2021

    work page 2021

  5. [4]

    W. Lei, Y . Y e, and M. Xiao, ‘‘Deep reinforcement learning-based spec- trum allocation in integrated access and backhaul networks,’’ IEEE Trans. Cogn. Commun. Netw., vol. 6, no. 3, pp. 970–979, Sep. 2020, doi: 10.1109/TCCN.2020.2992628

    arXiv 2020

  6. [5]

    Cheng, Z

    Q. Cheng, Z. Wei, and J. Y uan, ‘‘Deep reinforcement learning-based spectrum allocation and power management for IAB networks,’’ in Proc. IEEE Int. Conf. Commun. Workshops (ICC Workshops), Montreal, QC, Canada, Jun. 2021, pp. 1–6

    work page 2021

  7. [6]

    H. Yin, S. Roy, and L. Cao, ‘‘Routing and resource allocation for IAB multi-hop network in 5G advanced,’’ IEEE Trans. Commun., vol. 70, no. 10, pp. 6704–6717, Oct. 2022, doi: 10.1109/TCOMM.2022.3200673

    arXiv 2022

  8. [7]

    Zhang and I

    B. Zhang and I. Filippini, ‘‘Mobility-aware resource allocation for mmWave IAB networks: A multi-agent reinforcement learning approach,’’ IEEE/ACM Trans. Netw., early access, May 10, 2024, doi: 10.1109/TNET.2024.3396214

    arXiv 2024

Show all 34 references
  1. [8]

    Guidotti, A

    A. Guidotti, A. V anelli-Coralli, V . Schena, N. Chuberre, M. El Jaafari, J. Puttonen, and S. Cioni, ‘‘The path to 5G-advanced and 6G non-terrestrial network systems,’’ in Proc. 11th Adv. Satell. Multimedia Syst. Conf. 17th Signal Process. Space Commun. Workshop (ASMS/SPSC), G...

    2022

  2. [9]

    Zhang, M

    Y . Zhang, M. A. Kishk, and M.-S. Alouini, ‘‘A survey on integrated access and backhaul networks,’’ Frontiers Commun. Netw., vol. 2, Jun. 2021, Art. no. 647284

    2021

  3. [10]

    Perez, A

    A. Perez, A. Fouda, and A. S. Ibrahim, ‘‘Ray tracing analysis for UA V - assisted integrated access and backhaul millimeter wave networks,’’ in Proc. IEEE 20th Int. Symp. World Wireless, Mobile Multimedia Netw. (WoWMoM), Washington, DC, USA, Jun. 2019, pp. 1–5

    2019

  4. [11]

    Zhang and N

    L. Zhang and N. Ansari, ‘‘On the number and 3-D placement of in-band full-duplex enabled drone-mounted base-stations,’’ IEEE Wire- less Commun. Lett., vol. 8, no. 1, pp. 221–224, Feb. 2019, doi: 10.1109/LWC.2018.2867501

    2019

  5. [12]

    Kalantari, M

    E. Kalantari, M. Z. Shakir, H. Y anikomeroglu, and A. Y ongacoglu, ‘‘Backhaul-aware robust 3D drone placement in 5G+ wireless networks,’’ in Proc. IEEE Int. Conf. Commun. Workshops (ICC Workshops), Paris, France, May 2017, pp. 109–114

    2017

  6. [13]

    Almohamad, M

    A. Almohamad, M. O. Hasna, T. Khattab, and M. Haouari, ‘‘Maximiz- ing dense network flow through wireless multihop backhauling using UA Vs,’’ inProc. Int. Conf. Inf. Commun. Technol. Converg. (ICTC), Jeju, South Korea, Oct. 2018, pp. 526–531

    2018

  7. [14]

    B. Di, H. Zhang, L. Song, Y . Li, and G. Y . Li, ‘‘Data offloading in ultra- dense LEO-based integrated terrestrial-satellite networks,’’ in Proc. IEEE Global Commun. Conf. (GLOBECOM), Abu Dhabi, United Arab Emirates, Dec. 2018, pp. 1–6

    2018

  8. [15]

    Lagunas, S

    E. Lagunas, S. Maleki, L. Lei, C. Tsinos, S. Chatzinotas, and B. Ottersten, ‘‘Carrier allocation for hybrid satellite-terrestrial backhaul networks,’’ in Proc. IEEE Int. Conf. Commun. Workshops (ICC Workshops), Paris, France, May 2017, pp. 718–723

    2017

  9. [16]

    R. Deng, B. Di, S. Chen, S. Sun, and L. Song, ‘‘Ultra-dense LEO satellite offloading for terrestrial networks: How much to pay the satellite oper- ator?’’ IEEE Trans. Wireless Commun., vol. 19, no. 10, pp. 6240–6254, Oct. 2020

    2020

  10. [17]

    Abdullah, S

    Z. Abdullah, S. Kisseleff, E. Lagunas, V . N. Ha, F. Zeppenfeldt, and S. Chatzinotas, ‘‘Integrated access and backhaul via satellites,’’ 2023, arXiv:2304.01304

    2023 arXiv

  11. [18]

    Artiga, A

    X. Artiga, A. Perez-Neira, J. Baranda, E. Lagunas, S. Chatzinotas, R. Zetik, P . Gorski, K. Ntougias, D. Perez, and G. Ziaragkas, ‘‘Shared access satellite-terrestrial reconfigurable backhaul network enabled by smart antennas at mmWave band,’’ IEEE Netw., vol. 32, no. 5, pp. 4...

    2018

  12. [19]

    Y . Hu, M. Chen, and W. Saad, ‘‘Joint access and backhaul resource management in satellite-drone networks: A competitive market approach,’’ IEEE Trans. Wireless Commun., vol. 19, no. 6, pp. 3908–3923, Jun. 2020, doi: 10.1109/TWC.2020.2979127

    2020

  13. [20]

    Lee and H

    Y . Lee and H. Lee, ‘‘Resource allocation and placement for tethered flying platform-aided IAB network: Distributed DQN approach,’’ in Proc. IEEE 21st Consum. Commun. Netw. Conf. (CCNC), Las V egas, NV , USA, Jan. 2024, pp. 1096–1097, doi: 10.1109/CCNC51664.2024.10454748

    2024

  14. [21]

    R. S. Sutton and A. G. Barto, Reinforcement Learning—An Introduction, 1st ed., Cambridge, MA, USA: MIT Press, 1998

    1998

  15. [22]

    D. P . Bertsekas and J. N. Tsitsiklis, Neuro-dynamic Programming. Belmont, MA, USA: Athena Scientific, 1996

    1996

  16. [23]

    C. J. C. H. Watkins, ‘‘Learning from delayed rewards,’’ Ph.D. thesis, King’s College, London, U.K., 1989

    1989

  17. [24]

    C. J. C. H. Watkins and P . Dayan, ‘‘Q-learning,’’ Mach. Learn., vol. 8, nos. 3–4, pp. 279–292, May 1992, doi: 10.1007/BF00992698

    1992 doi

  18. [25]

    V . Mnih, K. Kavukcuoglu, D. Silver, A. Graves, I. Antonoglou, D. Wierstra, and M. Riedmiller, ‘‘Playing Atari with deep reinforcement learning,’’ in Proc. NIPS Deep Learning Workshop, Dec. 2013, pp. 1–9

    2013

  19. [26]

    V . Mnih, K. Kavukcuoglu, D. Silver, A. A. Rusu, J. V eness, M. G. Bellemare, A. Graves, M. Riedmiller, A. K. Fidjeland, G. Ostrovski, S. Petersen, C. Beattie, A. Sadik, I. Antonoglou, H. King, D. Kumaran, D. Wierstra, S. Legg, and D. Hassabis, ‘‘Human-level control through de...

    2015 doi

  20. [27]

    V an Hasselt, A

    H. V an Hasselt, A. Guez, and D. Silver, ‘‘Deep reinforcement learning with double Q-learning,’’ in Proc. 30th AAAI Conf. Artif. Intell. (AAAI), Phoenix, AZ, USA, Feb. 2016, pp. 2094–2100

    2016

  21. [28]

    Study on Integrated Access and Backhaul, document 3GPP TR 38.874, Dec. 2018

    2018

  22. [29]

    NR and NG-RAN Overall Description , document 3GPP TR 38.300, V ersion 16.1.0, Mar. 2020

    2020

  23. [30]

    Polese, M

    M. Polese, M. Giordani, A. Roy, D. Castor, and M. Zorzi, ‘‘Distributed path selection strategies for integrated access and backhaul at mmWaves,’’ in Proc. IEEE Global Commun. Conf. (GLOBECOM), Abu Dhabi, United Arab Emirates, Dec. 2018, pp. 1–7

    2018

  24. [31]

    Release 17 Description, document 3GPP TR 21.917, Jan. 2023

    2023

  25. [32]

    Mobile IAB for NR, document RP-222671, Qualcomm, Sep. 2022

    2022

  26. [33]

    Q. Y e, W. Zhuang, X. Li, and J. Rao, ‘‘End-to-end delay modeling for embedded VNF chains in 5G core networks,’’ IEEE Internet Things J., vol. 6, no. 1, pp. 692–704, Feb. 2019

    2019

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.