Pith. sign in

REVIEW 5 major objections 6 minor 42 references

Frequency Resource Management in 6G User-Centric CFmMIMO: A Hybrid Reinforcement Learning and Metaheuristic Approach

T0 review · 5 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read This paper claims that fusing the Aquila Optimizer with DDPG-based actor-critic reinforcement learning allocates frequency subbands in user-centric cell-free massive MIMO faster and with better balance than either method alone, reaching…

desk verdict A plausible hybrid method for subband allocation in UC-CFmMIMO, but the central convergence and fairness claims are not reproducible because the objective weights are unspecified and the Gini formula double-counts. read the letter →

arxiv 2505.22443 v1 pith:7CFGN4JM submitted 2025-05-28 cs.NI eess.SP

Selina Cheggour , Valeria Loscri This is my paper
classification cs.NIeess.SP
keywords 6Guser-centriccell-freemassiveMIMOfrequencyresourceallocationreinforcementlearningAquilaOptimizerDDPGspectralefficiencyvehicularnetworks
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 aims to show that a hybrid strategy, combining the metaheuristic Aquila Optimizer (AO) with Deep Deterministic Policy Gradient (DDPG) actor-critic reinforcement learning, solves the multi-objective frequency-subband allocation problem in user-centric cell-free massive MIMO (UC-CFmMIMO) networks. The target setting is dense 6G vehicular communications, where users share limited subbands under frequency-selective, bandwidth-dependent channels. The authors argue that the hybrid method, by using AO to generate exploratory actions that guide the RL agent, reaches the fastest and most stable convergence, achieves the highest spectral efficiency (over 200 bps/Hz), and keeps fairness at a Gini index around 0.02, outperforming both standalone AO and standalone DDPG. All results are evaluated with realistic 3GPP-3D channels generated by QuaDRiGa, making the claim about practical feasibility rather than idealized channel models.

What carries the argument

The load-bearing mechanism is the integration of the Aquila Optimizer into the DDPG actor-critic loop as an exploration generator: AO proposes exploratory actions that guide the actor network toward promising regions of the subband-assignment search space, while the critic evaluates those actions through Q-value estimates and the agent stores transitions in a replay buffer for stable off-policy training. The optimization objective is the weighted sum defined in equation (16), maximizing total spectral efficiency and the smallest eigenvalue of the channel matrix (to reduce interference) while penalizing the Gini index (to enforce fairness), subject to power, quality-of-service, and single-subband-per-user constraints. Supporting machinery includes zero-forcing precoding, equal power distribution, and QuaDRiGa-generated frequency-dependent channels at 5.9 GHz with 277 subbands, which supply the realistic propagation conditions the comparison rests on.

What would settle it

Rerun the comparison with deliberately corrupted or delayed channel estimates (for example, adding Gaussian error of increasing variance to the QuaDRiGa channels) and observe whether the hybrid's 200+ bps/Hz spectral efficiency and Gini index near 0.02 advantage over the baselines collapses; if it does, the claimed gains depend on perfect channel state information.

Watch

Extended reading notes

Core claim

The central discovery is that the hybrid AO-DDPG allocation scheme, referred to as HYM (or HRLM), consistently outperforms the standalone Aquila Optimizer and the standalone DDPG-based actor-critic model across all three objectives: total spectral efficiency, interference reduction, and fairness. In the reported simulations, HYM converges rapidly to an objective value close to 1.0, sustains spectral efficiency exceeding 200 bps/Hz, and stabilizes the Gini index at about 0.02, whereas AO plateaus near 0.6 objective value and about 100 bps/Hz, and the RL baseline reaches roughly 0.95 objective value and near 200 bps/Hz but with noticeable fluctuations. The paper also shows that spectral efficiency degrades gracefully as user density rises from 40 to 80 UEs (roughly 200 to 110 bps/Hz) and improves when more subbands are available, demonstrating that the hybrid framework remains stable under congestion.

Load-bearing premise

The entire performance comparison assumes the central processor has accurate channel state information for every user at every decision step, while realistic vehicle motion makes channel estimates stale, and the paper tests no sensitivity to estimation error.

Editorial extensions

If this is right

  • If the hybrid approach delivers the claimed convergence and spectral efficiency in realistic 3GPP-3D channels, it provides a practical frequency-allocation method for dense 6G vehicular deployments where subbands are shared among users.
  • The framework directly addresses the frequency-selectivity and bandwidth-dependence of propagation that simpler channel models ignore, so its performance estimates are more actionable for system designers.
  • The reported behavior under varying user density and subband counts gives a concrete scalability picture: spectral efficiency degrades with congestion but remains stable, and adding subbands mitigates the degradation.
  • Because the hybrid uses AO only to generate exploratory actions, the same architecture could be extended to other metaheuristics or other continuous resource-allocation variables, not just frequency subbands.
  • The fairness result, a Gini index near 0.02, suggests the scheme can prevent a few users from monopolizing good subbands while still maximizing total throughput, which is important for vehicular quality-of-service.

Reading between the lines

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

  • A natural extension, not made by the paper, is coupling frequency allocation with power allocation in the same hybrid loop, since the current equal-power scheme leaves the power budget unused as an additional lever.
  • The paper's assumed perfect channel state information is the weakest link in high-mobility scenarios; a testable extension is to feed stale or noisy channel estimates into the same algorithm and measure how quickly the 200+ bps/Hz gain erodes.
  • The claim that AO accelerates RL convergence suggests a broader principle: metaheuristic-guided exploration can substitute for extensive random exploration in other continuous-action wireless resource allocation problems, potentially reducing training time.
  • The scalability trends stopping at 80 UEs leave open how the hybrid behaves beyond that saturation point, so a direct extension would stress-test the method at 100+ UEs and larger subband counts to confirm graceful degradation.
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 / 6 minor

Summary. The manuscript proposes a hybrid frequency-resource allocation scheme for downlink user-centric cell-free massive MIMO in vehicular scenarios. Using QuaDRiGa-generated 3GPP-3D channels over 277 subbands, the authors formulate a multi-objective problem combining total spectral efficiency, the minimum eigenvalue of a channel matrix, and a Gini fairness index. They compare a standalone Aquila Optimizer, a DDPG actor-critic agent, and a hybrid AO-DDPG method. The central reported results are that the hybrid converges fastest to an objective value close to 1.0, achieves spectral efficiency above 200 bps/Hz, and reaches a Gini index near 0.02. Scalability experiments vary UE density and subband counts.

Significance. The topic is timely and relevant: the paper brings together a realistic 3GPP-3D channel model, a modern metaheuristic, and reinforcement learning for a 6G-relevant resource allocation problem, with explicit comparison against two baselines. The main spectral-efficiency and fairness metrics are computed from the channel and SINR, so the evaluation is not circular in its headline SE comparison. The paper also provides complexity expressions and a hyperparameter study. However, the central claims are currently under-supported: the objective weights and normalization are unspecified, the Gini formula is nonstandard, and the reported hyperparameters contradict the stated search ranges. With corrected definitions and reproducible details, the comparison could be a useful contribution, but the evaluation as written does not yet substantiate the claimed superior balance among objectives.

major comments (5)
  1. [Section 3.2, Eq. (16), and Section 7, Fig. 8] The objective weights w_eta, w_EVD, and w_Gini are never specified, and no normalization is described. Since eta_Total is a sum over 277 subbands and 40 UEs of log2(1+SINR), with order thousands of bps/Hz, while I_Gini is in [0,1] and lambda_min has its own scale, the weighted sum in Eq. (16) cannot, as written, take values near 1.0. The convergence curves in Fig. 8 therefore cannot be reproduced or interpreted, and the claim that the hybrid achieves the best balance is not supported. Please specify the weights, the normalization used for each term, and the exact objective formula used to produce Fig. 8.
  2. [Section 3.1.2, Eq. (13), and Section 7, Fig. 10] The Gini index is defined with an additional sum over subbands s, so it measures pairwise SE differences only among UEs assigned to the same subband, not among all UEs. Because constraint C3 (17c) assigns each UE to at most one subband, the double sum also introduces zero terms for non-assigned pairs. This is not the standard Gini coefficient, and the value 0.02 in Fig. 10 is not a valid fairness statistic for the system as a whole. Please replace Eq. (13) with the standard definition I_Gini = (1/(2 K^2 \bar{eta})) \sum_{i=1}^K \sum_{j=1}^K |\eta_i - \eta_j|, where \eta_i is the SE of UE i, and re-run the fairness comparison.
  3. [Section 3.2, constraint C2 (Eq. 17b)] As written, the constraint requires \eta(\mathcal{X}_{sk}) \ge \eta_{th} for every s and k, including subband-UE pairs for which no assignment exists; such entries are either zero or undefined under C3. This makes the constraint either infeasible for \eta_{th} > 0 or vacuous. Please rewrite C2 to range only over assigned pairs, for example \eta(\mathcal{X}_{sk}) \ge \eta_{th} \mathcal{X}_{sk}, or add the explicit condition \mathcal{X}_{sk} = 1.
  4. [Section 6.4 and Table 3] The selected Trial-5 configuration (actor and critic learning rate 0.00226, discount factor 0.882, batch size 272) contradicts the stated search ranges in Table 3 (learning rates 1e-5 to 1e-3, discount 0.9 to 0.99, batch size 32 to 256). This makes the reported hyperparameter optimization non-reproducible and suggests an error in either the table or the text. Please correct the inconsistency and confirm the exact values used.
  5. [Section 2 and Section 7] The system model assumes perfect CSI at the CPU for each decision step, but the target scenario is high-mobility vehicular communication. The paper provides no sensitivity analysis to CSI estimation error or Doppler-induced staleness. Without such an analysis, the reported performance advantage of the hybrid scheme over the baselines may not hold in the intended deployment. Please add a robustness experiment or, at minimum, a quantitative discussion of how CSI errors affect the compared algorithms.
minor comments (6)
  1. [Section 5 and Table 2] Section 5 states the simulation is conducted over a 1 km x 1 km area, while Table 2 reports R as 2 km x 2 km. This inconsistency should be corrected.
  2. [Section 4.4] In the complexity analysis, AO is described as 'Alternating Optimization', but elsewhere AO denotes the Aquila Optimizer. Please use consistent terminology.
  3. [Sections 4 and 7] The hybrid method is referred to as HRLM in Section 4.4 and HYM in Section 7. Please unify the acronym.
  4. [Figures 1 and 2] The figure numbering and captions are duplicated in the compiled text: 'Figure 1' appears both for the system architecture and for the channel frequency responses, and several figures have redundant 'Figure 1' captions in the supplement. Please renumber and clean the captions.
  5. [Section 6.1.1] The actor network output is described as a softmax layer, but DDPG normally uses a deterministic continuous action output. Clarify whether the action is a probability vector or an actual resource assignment, and how it is mapped to the assignment matrix.
  6. [Section 4.1] The Aquila Optimizer implementation is described only at a high level; for reproducibility, please state the population size, the number of AO iterations per step, and the encoding of the assignment matrix into AO solutions.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the claimed gains are evaluated on independent SE and Gini metrics, not on re-fitted parameters or a self-citation chain.

full rationale

The paper's main evaluation compares algorithms on standard, externally defined metrics: spectral efficiency (Eq. 5 and Figure 9) and Gini fairness (Figure 10). These metrics are computed from the channel and allocation matrices independently of the algorithm internals, so the central claim does not reduce to the optimization objective by construction. The convergence plot (Figure 8) plots the scalar objective of Eq. (16), which is of course the quantity each method is designed to optimize; this is a standard way to compare optimizers and is not a circular derivation of an independent result. The undisclosed weights/normalization in Eq. (16) and the nonstandard Gini formula in Eq. (13) are reproducibility and correctness concerns, not circularity: they do not cause a fitted parameter to be renamed as a prediction or a cited result to be load-bearing. The paper does not rely on any self-citation chain or imported uniqueness theorem; its cited background is external (QuaDRiGa, DDPG, Aquila Optimizer, etc.), and no claim is shown to be equivalent to its own input by construction.

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

The central performance claims rest on a simulation built from QuaDRiGa channel data plus a handful of unspecified weights and hyperparameters. No new physical entities are introduced. The assumptions of perfect CSI and localized interference are load-bearing for the precoding and allocation results. The simulation channel model is taken as a faithful representation of reality without external validation.

free parameters (3)
  • Objective weights w_eta, w_EVD, w_Gini = not reported
    Eq. (16) defines the multi-objective function, but the paper never provides the weight values used to produce Figures 8-12, so the objective values and relative importance of SE, interference, and fairness are unspecified.
  • DDPG hyperparameters (Trial 5) = alpha_pi = alpha_Q = 0.00226, gamma = 0.882, batch size 272, replay buffer 30,397
    Section 6.4 selects Trial 5 via random search based on training loss; these hyperparameters are fitted to the simulation setup rather than derived or validated on a separate set.
  • AO population size and iterations = not reported
    The complexity expressions in Section 4.4 include P and I, but no values are given, making the AO component and the hybrid overhead unquantified.
assumptions (4)
  • domain assumption Accurate channel state information for all UEs is available at the CPU
    Stated in Section 2; all allocation and precoding decisions use these channels, and no CSI error model is included.
  • domain assumption Interference originates predominantly from nearby UEs (set J_k)
    Section 2 defines J_k for ZF precoding; if interference is not localized in dense deployments, the precoding does not suppress all cross-subband interference.
  • domain assumption QuaDRiGa with 3GPP-38.901 UMi parameters accurately emulates real vehicular channels
    The simulation results depend on this channel model; no validation against measurements is provided.
  • standard math ZF precoding matrix inverse exists for the selected serving sets
    Equations (8)-(9) require H_{k,s} D_k H_{k,s}^H to be invertible, which is assumed without conditioning guarantees.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Frequency Resource Management in 6G User-Centric CFmMIMO: A Hybrid Reinforcement Learning and Metaheuristic Approach." pith.science (2026). https://pith.science/paper/7CFGN4JM

@misc{pith2026250522443,
  author       = {Pith},
  title        = {Pith review of: Frequency Resource Management in 6G User-Centric CFmMIMO: A Hybrid Reinforcement Learning and Metaheuristic Approach},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7CFGN4JM}},
  note         = {Machine review of arXiv:2505.22443}
}
read the original abstract

As sixth-generation (6G) networks continue to evolve, AI-driven solutions are playing a crucial role in enabling more efficient and adaptive resource management in wireless communication. One of the key innovations in 6G is user-centric cell-free massive Multiple-Input Multiple-Output (UC-CFmMIMO), a paradigm that eliminates traditional cell boundaries and enhances network performance by dynamically assigning access points (APs) to users. This approach is particularly well-suited for vehicular networks, offering seamless, homogeneous, ultra-reliable, and low-latency connectivity. However, in dense networks, a key challenge lies in efficiently allocating frequency resources within a limited shared subband spectrum while accounting for frequency selectivity and the dependency of signal propagation on bandwidth. These factors make resource allocation increasingly complex, especially in dynamic environments where maintaining Quality of Service (QoS) is critical. This paper tackles these challenges by proposing a hybrid multi-user allocation strategy that integrates reinforcement learning (RL) and metaheuristic optimization to enhance spectral efficiency (SE), ensure fairness, and mitigate interference within shared subbands. To assess its effectiveness, we compare this hybrid approach with two other methods: the bio-inspired Aquila Optimizer (AO) and Deep Deterministic Policy Gradient (DDPG)-based Actor-Critic Reinforcement Learning (AC-RL). Our evaluation is grounded in real-world patterns and channel characteristics, utilizing the 3GPP-3D channel modeling framework (QuaDRiGa) to capture realistic propagation conditions. The results demonstrate that the proposed hybrid strategy achieves a superior balance among competing objectives, underscoring the role of AI-driven resource allocation in advancing UC-CFmMIMO systems for next-generation wireless networks.

Figures

Figures reproduced from arXiv: 2505.22443 by the authors.

Figure 1
Figure 1. Illustration of UC mMIMO clustering in a CF network architecture. Selina Cheggour et al.: Preprint submitted to Elsevier Page 13 of 12 [PITH_FULL_IMAGE:figures/full_fig_p013_1.png] view at source ↗
Figure 1
Figure 1. Channel frequency responses of the 10 AP-UE pair links. Selina Cheggour et al.: Preprint submitted to Elsevier Page 1 of 1 [PITH_FULL_IMAGE:figures/full_fig_p014_1.png] view at source ↗
Figure 4
Figure 4. Architecture of Actor and Critic deep neural networks (DNNs). Selina Cheggour et al.: Preprint submitted to Elsevier Page 14 of 12 [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗
Figures from the paper (5 more)
Figure 5
Figure 5. Figure 5: Proposed HRLM framework. Hybrid AI & Metaheuristic for Frequency Allocation in 6G UC-CFmMIMO 10 20 30 40 50 60 70 80 90 100 −6 −4 −2 0 Steps Actor Loss Trial 1 Trial 2 Trial 3 Trial 4 Trial 5 Trial 6 [PITH_FULL_IMAGE:figures/full_fig_p015_5.png]
Figure 1
Figure 1. Figure 1: Loss variation across training steps for different trials of the actor network. Selina Cheggour et al.: Preprint submitted to Elsevier Page 1 of 1 [PITH_FULL_IMAGE:figures/full_fig_p015_1.png]
Figure 1
Figure 1. Figure 1: Objective value convergence across AO, RLM, and HYM. Selina Cheggour et al.: Preprint submitted to Elsevier Page 1 of 1 [PITH_FULL_IMAGE:figures/full_fig_p016_1.png]
Figure 11
Figure 11. Figure 11: SE comparison across 40, 60, and 80 UEs. Selina Cheggour et al.: Preprint submitted to Elsevier Page 16 of 12 [PITH_FULL_IMAGE:figures/full_fig_p016_11.png]
Figure 1
Figure 1. Figure 1: SE comparison across 40, 80, and 120 SBs distributions. Selina Cheggour et al.: Preprint submitted to Elsevier Page 1 of 1 [PITH_FULL_IMAGE:figures/full_fig_p017_1.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

42 extracted references · 37 canonical work pages

  1. [1]

    W. Wu, X. Huang, T. H. Luan, Ai-native network digital twin for intelligent network management in 6g, arXiv preprint 2410 (01584) (2024)

    work page 2024

  2. [2]

    H. Q. Ngo, G. Interdonato, E. G. Larsson, G. Caire, J. G. Andrews, Ultradense cell-free massive mimo for 6g: Technical overview and open questions, Proceedings of the IEEE (2024)

    work page 2024

  3. [3]

    Scalability Aspects of Cell-Free Massive MIMO

    G. Interdonato, P. Frenger, E. G. Larsson, Scalability aspects of cell- free massive mimo, arXiv preprint arXiv:1902.11275 (2019)

    work page Pith review arXiv 2019

  4. [4]

    O.T.Demir,E.Bjornson,L.Sanguinetti,Foundationsofuser-centric cell-freemassivemimo,FoundationsandTrendsinSignalProcessing 14 (3-4) (2021) 162–472

    work page 2021

  5. [5]

    Conceição, M

    F. Conceição, M. Gomes, V. Silva, R. Dinis, Survey on resource allocation for future 6g network architectures: Cell-free and radio stripe technologies, Electronics 13 (13) (2024) 2489

    work page 2024

  6. [6]

    Anand, C

    A. Anand, C. R. Murthy, R. Chopra, Impact of mobility on the downlink performance of cell-free massive mimo systems, Physical Communication 61 (2023) 102178

    work page 2023

  7. [7]

    N. M. H. Robbi, I. W. Mustika, W. Widyawan, A modified genetic algorithm for resource allocation in cognitive radio networks in the presence of primary users, in: 2019 International Seminar on Intelli- gent Technology and Its Applications (ISITIA), 2019, pp. 19–23

    work page 2019

  8. [8]

    Hussein, M

    A. Hussein, M. Anbar, M. Molhem, A. Y. Oudeh, M. Nassr, H. Far- doun, Channel allocation in 5g networks using ant colony optimiza- tionalgorithm,in:20246thInternationalYouthConferenceonRadio Electronics, Electrical and Power Engineering (REEPE), 2024, pp. 1–5.doi:10.1109/REEPE60449.2024.10479823

    arXiv 2024

Show all 42 references
  1. [9]

    C. Liu, J. Zhen, Diagonal loading beamforming based on aquila op- timizer, IEEE Access 11 (2023) 69091–69100.doi:10.1109/ACCESS. 2023.3293403

    2023

  2. [10]

    Anandan, S

    R. Anandan, S. Abdur Rahman, Seenuvasamurthi, G. Vishnu Vard- han Rao, Ccoa-adals: Hybrid beamforming using chaotic chebyshev aquilaoptimizationformmwavemassivemimo,InternationalJournal of Communication Systems (2024).doi:10.1002/dac.6069

    2024 doi

  3. [11]

    Alwarafy, M

    A. Alwarafy, M. Abdallah, B. S. Ciftler, A. Al-Fuqaha, M. Hamdi, Deep reinforcement learning for radio resource allocation and man- agement in next-generation heterogeneous wireless networks: A sur- vey, arXiv preprint arXiv:2106.00574 (2021)

    2021 arXiv

  4. [12]

    R.Wang,M.Shen,Y.He,X.Liu,Jointaccesspoints-userassociation and caching placement strategy for cell-free massive mimo systems based on soft actor-critic algorithm, IEEE Communications Letters (2023)

    2023

  5. [13]

    G. Wang, P. Cheng, Z. Chen, B. Vucetic, Y. Li, Green cell-free mas- sive mimo: An optimization-embedded deep reinforcement learning approach, IEEE Transactions on Signal Processing (2024)

    2024

  6. [14]

    Kölle, D

    M. Kölle, D. Seidl, M. Zorn, P. Altmann, J. Stein, T. Gabor, Opti- mizing variational quantum circuits using metaheuristic strategies in reinforcement learning, in: 2024 IEEE International Conference on Quantum Computing and Engineering (QCE), Vol. 2, IEEE, 2024, pp. 323–328

    2024

  7. [15]

    M.Wang,X.Liu,F.Wang,Y.Liu,T.Qiu,M.Jin,Spectrum-efficient user grouping and resource allocation based on deep reinforcement learning for mmwave massive mimo-noma systems, Scientific Re- ports 14 (1) (2024) 8884.doi:10.1038/s41598-024-34715-8

    2024 doi

  8. [16]

    Anand, C

    A. Anand, C. R. Murthy, R. Chopra, Impact of mobility on the downlink performance of cell-free massive MIMO systems, Phys. Commun. 61 (2023) 102178

    2023

  9. [17]

    W.Jiang,H.D.Schotten,Cell-freemassivemimo-ofdmtransmission overfrequency-selectivefadingchannels,IEEECommunicationsLet- ters 25 (8) (2021) 2718–2722

    2021

  10. [18]

    Farzanullah, Deep reinforcement learning for resource allocation in next-generation c-v2x and iiot networks (2023)

    M. Farzanullah, Deep reinforcement learning for resource allocation in next-generation c-v2x and iiot networks (2023)

    2023

  11. [19]

    K. Xu, S. Zhou, G. Y. Li, Federated reinforcement learning for resourceallocationinv2xnetworks,IEEEJournalofSelectedTopics in Signal Processing (2024)

    2024

  12. [20]

    Mohammadi, Z

    M. Mohammadi, Z. Mobini, H. Q. Ngo, M. Matthaiou, Next- generationmultipleaccesswithcell-freemassivemimo,Proceedings of the IEEE (2024)

    2024

  13. [21]

    J. F. D. Valgas, D. Martín-Sacristán, J. F. Monserrat, 5g new radio numerologies and their impact on v2x communications, in: Waves, Universitat Politecnica de Valencia, 2018, pp. 15–22

    2018

  14. [22]

    E.Bjornson,E.Jorswieck,Optimalresourceallocationincoordinated multi-cellsystems,FoundationsandTrends®inCommunicationsand Information Theory 9 (2–3) (2013) 113–381

    2013

  15. [23]

    Chakraborty, O

    S. Chakraborty, O. T. Demir, E. Bjornson, P. Giselsson, Efficient downlink power allocation algorithms for cell-free massive mimo systems,IEEEOpenJournaloftheCommunicationsSociety2(2020) 168–186

    2020

  16. [24]

    2, 2005, pp

    M.Dianati,X.Shen,S.Naik,Anewfairnessindexforradioresource allocation in wireless networks, in: IEEE Wireless Communications and Networking Conference, 2005, Vol. 2, 2005, pp. 712–717

    2005

  17. [25]

    Q. Sun, H. Wu, O. Petrosian, Optimal power allocation based on metaheuristic algorithms in wireless network, Mathematics 10 (18) (2022) 3336

    2022

  18. [26]

    Abualigah, D

    L. Abualigah, D. Yousri, M. Abd Elaziz, A. A. Ewees, M. A. Al- Qaness, A. H. Gandomi, Aquila optimizer: a novel meta-heuristic optimization algorithm, Computers & Industrial Engineering 157 (2021) 107250

    2021

  19. [27]

    M.S.Frikha,S.M.Gammar,A.Lahmadi,L.Andrey,Reinforcement and deep reinforcement learning for wireless internet of things: A survey, Computer Communications 178 (2021) 98–113

    2021

  20. [28]

    Ghasemi, A

    M. Ghasemi, A. H. Mousavi, D. Ebrahimi, Comprehensive survey of reinforcement learning: From algorithms to practical challenges, arXiv preprint arXiv:2411.18892 (2024)

    2024 arXiv

  21. [29]

    J. Kim, S. W. Choi, G. Noh, H. Chung, I. Kim, A study on frequency planning of mn system for 5g vehicular communications, in: 2019 International Conference on Information and Communication Tech- nology Convergence (ICTC), IEEE, 2019, pp. 1442–1445

    2019

  22. [30]

    S. D. Liyanaarachchi, C. B. Barneto, T. Riihonen, M. Valkama, Experimenting joint vehicular communications and sensing with op- timized 5g nr waveform, in: 2021 IEEE 93rd Vehicular Technology Conference (VTC2021-Spring), IEEE, 2021, pp. 1–5

    2021

  23. [31]

    P.Kumar,K.B.Ali,Intelligenttrafficsystemusingvehicletovehicle (v2v) & vehicle to infrastructure (v2i) communication based on wireless access in vehicular environments (wave) std, in: 2022 10th International Conference on Reliability, Infocom Technologies and Optimization (Tren...

    2022

  24. [32]

    Parkvall, E

    S. Parkvall, E. Dahlman, A. Furuskar, M. Frenne, NR: The new 5G radioaccesstechnology,IEEECommunicationsStandardsMagazine 1 (4) (2017) 24–30

    2017

  25. [33]

    Zhang, Channel Characterization and Modeling for Future Wire- less Communication Systems, 2023

    G. Zhang, Channel Characterization and Modeling for Future Wire- less Communication Systems, 2023

    2023

  26. [34]

    Zheng, C

    Y. Zheng, C. X. Wang, R. Yang, L. Yu, F. Lai, J. Huang, Z. Zhong, Ultra-massive mimo channel measurements at 5.3 ghz and a general 6gchannelmodel,IEEETransactionsonVehicularTechnology72(1) (2022) 20–34

    2022

  27. [35]

    Z. Wang, J. Zhang, E. Bjornson, B. Ai, Uplink performance of cell- free massive mimo over spatially correlated rician fading channels, IEEE Communications Letters 25 (4) (2020) 1348–1352

    2020

  28. [36]

    O.Ozdogan,E.Bjornson,J.Zhang,Performanceofcell-freemassive mimo with rician fading and phase shifts, IEEE Transactions on Wireless Communications 18 (11) (2019) 5299–5315

    2019

  29. [37]

    T. P. Lillicrap, J. J. Hunt, A. Pritzel, N. Heess, T. Erez, Y. Tassa, D. Silver, D. Wierstra, Continuous control with deep reinforcement learning, arXiv preprint arXiv:1509.02971 (2015)

    2015 arXiv

  30. [38]

    Eimer, M

    T. Eimer, M. Lindauer, R. Raileanu, Hyperparameters in reinforce- ment learning and how to tune them, in: International Conference on Machine Learning, PMLR, 2023, pp. 9104–9149. URLhttps://proceedings.mlr.press/v202/eimer23a.html

    2023

  31. [39]

    Zahavy, Z

    T. Zahavy, Z. Xu, V. Veeriah, M. Hessel, J. Oh, H. P. van Hasselt, D.Silver,S.Singh,Aself-tuningactor-criticalgorithm,in:Advances inNeuralInformationProcessingSystems,Vol.33,2020,pp.20913– 20924. URLhttps://proceedings.neurips.cc/paper_files/paper/2020/file/ 20a64c3e09657263cc...

    2020

  32. [40]

    Vasan, M

    G. Vasan, M. Elsayed, S. A. Azimi, J. He, F. Shahriar, C. Bellinger, A. Rajeswaran, R. Mahmood, Deep policy gradient methods without batch updates, target networks, or replay buffers, in: Advances in Neural Information Processing Systems, Vol. 37, 2025, pp. 845–891

    2025

  33. [41]

    N. M. Ashraf, R. R. Mostafa, R. H. Sakr, M. Z. Rashad, Optimizing hyperparametersofdeepreinforcementlearningforautonomousdriv- ing based on whale optimization algorithm, Plos One 16 (6) (2021) e0252754

    2021

  34. [42]

    C. Liu, Y. Zhang, Hyper-parameter optimization based on soft ac- tor critic and hierarchical mixture regularization, arXiv preprint arXiv:2112.04084 (2021). URLhttps://arxiv.org/abs/2112.04084 Selina Cheggourreceived an M.S. degree in Electronics and Telecommu- nicationsSystem...

    2021 arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.