arxiv: 2604.08197 · v1 · submitted 2026-04-09 · 📡 eess.SP

Recognition: 1 theorem link

· Lean Theorem

Discrete Diffusion for Codebook-Based Beam Candidate Generation

Amirhossein Azarbahram , Onel L. A. L\'opez

Authors on Pith no claims yet

Pith reviewed 2026-05-10 17:45 UTC · model grok-4.3

classification 📡 eess.SP

keywords mmWave beam managementlimited probingdiscrete diffusionbeam candidate generationcodebook-based systemshistory-conditioned generationbeam alignment

0 comments

The pith

A history-conditioned discrete diffusion model learns to generate effective beam candidates for limited-probing mmWave alignment from logged histories.

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

The paper establishes that training a discrete denoising diffusion model on past probing sequences allows it to sample promising beam indices from a learned conditional distribution, which then supplies the small set of beams that can be measured in each time slot. A reader would care because mmWave links require frequent realignment under blockage and mobility, yet only a handful of beams can be tested at once, so the quality of the candidate list directly determines whether the best direction is ever considered. When the model is right, the generated candidates produce higher received signal strength, fewer outright misses of the optimal beam, and lower regret relative to what an oracle would have chosen. The gains appear largest when the probing budget is smallest, the regime where conventional selection methods degrade fastest.

Core claim

The central claim is that a history-conditioned discrete denoising diffusion probabilistic model, trained on logged probing histories, learns the conditional distribution over promising beam indices and thereby constructs superior probing candidate sets for codebook-based mmWave systems; numerical results show this yields higher signal-to-noise ratio, lower beam-miss probability, and lower conditional probe regret than strong learning-based and discriminative baselines, with the advantage most visible under tight probing budgets.

What carries the argument

history-conditioned discrete denoising diffusion probabilistic model that learns a conditional distribution over beam indices from probing histories and samples candidate sets online

If this is right

Under the same probing budget the generated candidates raise average signal-to-noise ratio relative to baseline selection methods.
Beam-miss probability drops because the diffusion samples place higher mass on directions that would have been optimal.
Conditional probe regret, measured against an oracle that knows the best beam, decreases especially when only one or two beams can be measured per slot.
The advantage widens as the probing budget shrinks, confirming that accurate candidate generation matters most when measurement opportunities are scarcest.

Where Pith is reading between the lines

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

If the diffusion sampler can be run in a few forward passes, beam management latency could fall because fewer exhaustive searches are needed.
The same generative framing might apply to other sequential wireless decisions where only a subset of actions can be tested, such as channel sounding or handover candidate selection.
Adding explicit features like recent velocity estimates into the conditioning vector could tighten the learned distribution further.

Load-bearing premise

Logged probing histories contain enough information to learn a conditional distribution over promising beams that generalizes to new mobility and blockage patterns.

What would settle it

Evaluate the trained model on a test set whose mobility traces and blockage statistics are drawn from a distribution deliberately shifted from the training logs and check whether the reported gains in SNR and miss probability vanish.

Figures

Figures reproduced from arXiv: 2604.08197 by Amirhossein Azarbahram, Onel L. A. L\'opez.

**Figure 2.** Figure 2: The hierarchical Transformer encoder that aggre [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

**Figure 3.** Figure 3: A simple illustrative example of diffusion models’ [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

**Figure 4.** Figure 4: Illustration of the offline–online workflow. Offline, the BS collects probing–feedback interaction traces from [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗

**Figure 5.** Figure 5: The online operation loop: trained conditional [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗

**Figure 6.** Figure 6: The block diagram of the adapted baselines. [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗

**Figure 7.** Figure 7: The average training loss of the learning ap [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗

**Figure 9.** Figure 9: Top-m inclusion rate as a function of P for m ∈ {1, 2, 4} with L = 1. for predicting promising beams. Across all values of L, the D3PM-BM consistently achieves a smaller oracle gap than TRM and ODE-LSTM. Meanwhile, increasing L reduces the miss probability for all learning approaches, since additional past probing outcomes help the models better anticipate the future. Moreover, D3PM-BM consistently exhibit… view at source ↗

**Figure 10.** Figure 10: (a) SNR gap to the oracle (top), (b) oracle [PITH_FULL_IMAGE:figures/full_fig_p012_10.png] view at source ↗

**Figure 12.** Figure 12: Average (a) SNR (top) and (b) inference time [PITH_FULL_IMAGE:figures/full_fig_p013_12.png] view at source ↗

**Figure 13.** Figure 13: Average SNR at the user as a function of (a) [PITH_FULL_IMAGE:figures/full_fig_p013_13.png] view at source ↗

read the original abstract

Millimeter-wave (mmWave) communication enables high data rates through large bandwidths and highly directional beamforming, but its sensitivity to blockage and mobility makes reliable beam alignment a central challenge. Limited-probing beam management is a fundamental problem in codebook-based mmWave systems, where only a small subset of beams can be evaluated simultaneously, and the serving decision is restricted to the probed set. Under mobility and noisy feedback, this leads to a sequential and partially observable decision problem in which performance depends critically on the quality of the proposed beam candidates. In this paper, we consider limited-probing beam management and develop a history-conditioned discrete denoising diffusion probabilistic model for beam candidate generation. The proposed method learns from logged probing histories a conditional distribution over promising beam indices, which is then used to construct probing candidates online. Numerical analysis shows that the proposed approach consistently achieves better signal-to-noise ratio, beam-miss probability, and conditional probe regret under tight probing budgets compared with strong learning-based and discriminative baselines. The gains are especially pronounced in low-probing regimes, where accurate candidate generation is most critical.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper brings discrete diffusion to history-conditioned beam candidate generation in limited-probing mmWave systems and reports better SNR and lower miss rates than baselines in simulations, but the generalization claim rests on untested assumptions about mobility and blockage shifts.

read the letter

The paper takes discrete diffusion models and applies them to generating beam candidates in codebook-based mmWave systems under limited probing. It learns a conditional distribution over good beams from logged histories and uses that to pick what to probe next. The numerical results claim consistent improvements in signal-to-noise ratio, beam miss probability, and probe regret compared to other learning and discriminative methods, with bigger gains when the probing budget is tight. What stands out as new is bringing diffusion-based generative modeling to this specific beam management task. Most prior work uses reinforcement learning or simpler classifiers for beam selection, so this is a different technical route that might capture the distribution over promising indices more flexibly. The work does a reasonable job setting up the problem as a partially observable sequential decision and showing that the diffusion approach can outperform baselines in their simulations. That part is straightforward and the gains look plausible on the surface. The soft spot is around generalization. The stress-test concern is on point: the claimed robustness under tight budgets assumes the learned distribution transfers to new mobility and blockage conditions. If the experiments train and test on data from the same mobility models without holding out different velocities, blockage densities, or spatial correlations, then the results might just reflect fitting the training distribution rather than real robustness. The abstract does not mention any such out-of-distribution tests, so I would want to see those details in the full paper before being convinced the method is ready for varying real-world conditions. This paper is for people in wireless communications who work on learning-based beam alignment and mmWave systems. A reader already familiar with diffusion models or beam management literature would get the most out of it, as it is a targeted application rather than a broad methodological advance. It deserves a serious referee. The idea is fresh enough and the problem important enough that referees can help sharpen the evaluation and clarify the scope of the claims. I would recommend sending it for peer review, with specific feedback on adding generalization experiments and more experimental details.

Referee Report

2 major / 1 minor

Summary. The manuscript develops a history-conditioned discrete denoising diffusion probabilistic model for generating beam candidates in limited-probing mmWave beam management. By learning a conditional distribution over promising beam indices from logged probing histories, the method constructs probing candidates online. Numerical results indicate superior performance in terms of signal-to-noise ratio, beam-miss probability, and conditional probe regret relative to strong baselines, with gains most notable under tight probing budgets.

Significance. If the reported gains prove robust, the approach could meaningfully improve beam alignment efficiency in dynamic mmWave systems by leveraging generative modeling on logged data. The application of discrete diffusion to codebook-based candidate generation is a novel framing that may better handle the partially observable sequential decision problem than purely discriminative baselines.

major comments (2)

The central claim of consistent gains under tight probing budgets rests on the model's ability to produce a useful conditional distribution p(beam indices | history) that generalizes beyond training. The numerical analysis description does not indicate whether held-out test scenarios alter user velocity distributions, blockage densities, or spatial correlation lengths relative to training data; without such explicit out-of-distribution validation, the improvements may reflect in-distribution interpolation rather than the claimed robustness.
The abstract asserts performance gains but supplies no details on model architecture, training procedure, dataset characteristics, or statistical significance testing. These elements are load-bearing for verifying that the data support the stated superiority over learning-based and discriminative baselines.

minor comments (1)

The abstract could be strengthened by briefly noting key implementation choices (e.g., diffusion steps, conditioning mechanism, or loss) to aid immediate assessment of reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thoughtful review and valuable feedback on our manuscript. We have carefully considered the major comments and provide point-by-point responses below. We plan to incorporate revisions to strengthen the paper as outlined.

read point-by-point responses

Referee: The central claim of consistent gains under tight probing budgets rests on the model's ability to produce a useful conditional distribution p(beam indices | history) that generalizes beyond training. The numerical analysis description does not indicate whether held-out test scenarios alter user velocity distributions, blockage densities, or spatial correlation lengths relative to training data; without such explicit out-of-distribution validation, the improvements may reflect in-distribution interpolation rather than the claimed robustness.

Authors: We agree that explicit out-of-distribution testing would strengthen the claims regarding robustness. The current numerical results use held-out test scenarios drawn from the same underlying distributions as the training data, which demonstrates performance on unseen histories but within the same environment statistics. To address this concern, we will add new experiments in the revised manuscript where we vary user velocity distributions, blockage densities, and spatial correlation lengths in the test set. These will be compared against the baselines to show generalization. revision: yes
Referee: The abstract asserts performance gains but supplies no details on model architecture, training procedure, dataset characteristics, or statistical significance testing. These elements are load-bearing for verifying that the data support the stated superiority over learning-based and discriminative baselines.

Authors: We acknowledge that the abstract is concise and omits these details, as is typical to meet length constraints. The full manuscript provides descriptions of the model architecture in Section III, the training procedure in Section IV, and dataset characteristics in Section V. To further support the claims, we will include statistical significance testing, such as results from multiple independent runs with error bars, in the revised numerical analysis section. We believe this addresses the verification concern without altering the abstract substantially. revision: partial

Circularity Check

0 steps flagged

No circularity: derivation relies on external logged data and independent numerical evaluation

full rationale

The paper frames beam candidate generation as learning a conditional distribution p(beam indices | history) via a discrete denoising diffusion model trained on logged probing histories. Performance claims rest on numerical comparisons of SNR, beam-miss probability, and conditional probe regret against external baselines, with no equations or steps that reduce the target quantities to fitted parameters by construction, no load-bearing self-citations, and no uniqueness theorems imported from prior author work. The derivation chain is therefore self-contained against external benchmarks and does not exhibit any of the enumerated circular patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Based solely on the abstract, no explicit free parameters, axioms, or invented entities are described beyond the standard components of a discrete diffusion model.

pith-pipeline@v0.9.0 · 5490 in / 1080 out tokens · 27177 ms · 2026-05-10T17:45:02.133849+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean, IndisputableMonolith/Foundation/RealityFromDistinction.lean, IndisputableMonolith/Foundation/AlexanderDuality.lean reality_from_one_distinction, washburn_uniqueness_aczel, alexander_duality_circle_linking unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

history-conditioned discrete denoising diffusion probabilistic model for beam candidate generation... D3PM-BM... hierarchical Transformer encoder... soft oracle labels... sampling-to-ranking

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

55 extracted references · 10 canonical work pages · 1 internal anchor

[1]

Millimeter Wave Communication: A Compre- hensive Survey,

X. Wang et al., “Millimeter Wave Communication: A Compre- hensive Survey,” IEEE Commun. Surveys Tuts., vol. 20, no. 3, pp. 1616–1653, 2018

2018
[2]

A Tutorial on Beam Management for 3GPP NR at mmWave Frequencies,

M. Giordani et al., “A Tutorial on Beam Management for 3GPP NR at mmWave Frequencies,” IEEE Commun. Surveys Tuts., vol. 21, no. 1, pp. 173–196, 2019

2019
[3]

NR; physical layer procedures for data,

3GPP, “NR; physical layer procedures for data,” 3GPP, Tech. Rep. TS 38.214, release 18
[4]

Millimeter Wave Mobile Communica- tions for 5G Cellular: It Will Work!

T. S. Rappaport et al., “Millimeter Wave Mobile Communica- tions for 5G Cellular: It Will Work!” IEEE Access, vol. 1, pp. 335–349, 2013

2013
[5]

An Overview of Signal Processing Tech- niques for Millimeter Wave MIMO Systems,

R. W. Heath et al., “An Overview of Signal Processing Tech- niques for Millimeter Wave MIMO Systems,” IEEE J. Sel. Topics Signal Process., vol. 10, no. 3, pp. 436–453, 2016

2016
[6]

Beam Codebook Based Beamforming Protocol for Multi-Gbps Millimeter-Wave WPAN Systems,

J. Wang et al., “Beam Codebook Based Beamforming Protocol for Multi-Gbps Millimeter-Wave WPAN Systems,” IEEE J. Sel. Areas Commun., vol. 27, no. 8, pp. 1390–1399, 2009

2009
[7]

Beam Design for Beam Switching Based Millimeter Wave Vehicle- to-Infrastructure Communications,

V. Va, T. Shimizu, G. Bansal, and R. W. Heath, “Beam Design for Beam Switching Based Millimeter Wave Vehicle- to-Infrastructure Communications,” in IEEE ICC, 2016, pp. 1–6

2016
[8]

The Complexity of Markov Decision Processes,

C. H. Papadimitriou and J. N. Tsitsiklis, “The Complexity of Markov Decision Processes,” Math. Oper. Res., vol. 12, no. 3, pp. 441–450, 1987

1987
[9]

Deep Learning for mmWave Beam-Management: State-of-the-Art, Opportunities and Challenges,

K. Ma et al., “Deep Learning for mmWave Beam-Management: State-of-the-Art, Opportunities and Challenges,” IEEE Wire- less Commun., vol. 30, no. 4, pp. 108–114, 2023

2023
[10]

Generative AI for Physical Layer Communications: A Survey,

N. Van Huynh et al., “Generative AI for Physical Layer Communications: A Survey,” IEEE Trans. Cogn. Commun. Netw., vol. 10, no. 3, pp. 706–728, 2024

2024
[11]

A comprehensive survey of ai-generated content (aigc): A history of generative ai from gan to chatgpt,

Y. Cao et al., “A Comprehensive Survey of AI-Generated Content (AIGC): A History of Generative AI from GAN to ChatGPT,” 2023. [Online]. A vailable: https://arxiv.org/abs/ 2303.04226

work page arXiv 2023
[12]

Denoising Diffusion Proba- bilistic Models,

J. Ho, A. Jain, and P. Abbeel, “Denoising Diffusion Proba- bilistic Models,” in NeurIPS, vol. 33. Curran Associates, Inc., 2020, pp. 6840–6851

2020
[13]

MmWave Beam Prediction with Situational Awareness: A Machine Learning Approach,

Y. Wang, M. Narasimha, and R. W. Heath, “MmWave Beam Prediction with Situational Awareness: A Machine Learning Approach,” in IEEE SPA WC, 2018, pp. 1–5

2018
[14]

Deep Learning for mmWave Beam and Blockage Prediction Using Sub-6 GHz Channels,

M. Alrabeiah and A. Alkhateeb, “Deep Learning for mmWave Beam and Blockage Prediction Using Sub-6 GHz Channels,” IEEE Trans. Commun., vol. 68, no. 9, pp. 5504–5518, 2020

2020
[15]

Deep Learning Assisted mmWave Beam Pre- diction for Heterogeneous Networks: A Dual-Band Fusion Approach,

K. Ma et al., “Deep Learning Assisted mmWave Beam Pre- diction for Heterogeneous Networks: A Dual-Band Fusion Approach,” IEEE Trans. Commun., vol. 71, no. 1, pp. 115– 130, 2023

2023
[16]

A Low-Complexity Machine Learning De- sign for mmWave Beam Prediction,

M. Q. Khan et al., “A Low-Complexity Machine Learning De- sign for mmWave Beam Prediction,” IEEE Wireless Commun. Lett., vol. 13, no. 6, pp. 1551–1555, 2024. 14

2024
[17]

LSTM-Based Predictive mmWave Beam Track- ing via Sub-6 GHz Channels for V2I Communications,

Y. Zhao et al., “LSTM-Based Predictive mmWave Beam Track- ing via Sub-6 GHz Channels for V2I Communications,” IEEE Trans. Commun., vol. 72, no. 10, pp. 6254–6270, 2024

2024
[18]

Enhancing mmWave Beam Prediction through Deep Learning with Sub-6 GHz Channel Estimate,

W. Deng, M. Li, Y. Liu, M.-M. Zhao, and M. Lei, “Enhancing mmWave Beam Prediction through Deep Learning with Sub-6 GHz Channel Estimate,” in IEEE WCNC, 2024, pp. 1–6

2024
[19]

FusionNet: Enhanced Beam Prediction for mmWave Communications Using Sub-6 GHz Channel and a Few Pilots,

F. Gao et al., “FusionNet: Enhanced Beam Prediction for mmWave Communications Using Sub-6 GHz Channel and a Few Pilots,” IEEE Trans. Commun., vol. 69, no. 12, pp. 8488– 8500, 2021

2021
[20]

Vision-Position Multi-Modal Beam Predic- tion Using Real Millimeter Wave Datasets,

G. Charan et al., “Vision-Position Multi-Modal Beam Predic- tion Using Real Millimeter Wave Datasets,” in IEEE WCNC, 2022, pp. 2727–2731

2022
[21]

LiDAR Aided Future Beam Prediction in Real-World Millimeter Wave V2I Commu- nications,

S. Jiang, G. Charan, and A. Alkhateeb, “LiDAR Aided Future Beam Prediction in Real-World Millimeter Wave V2I Commu- nications,” IEEE Wireless Commun. Lett., vol. 12, no. 2, pp. 212–216, 2023

2023
[22]

Multimodal transformers for wireless communications: A case study in b eam pre- diction,

Y. Tian et al., “Multimodal Transformers for Wireless Communications: A Case Study in Beam Prediction,” 2023. [Online]. A vailable: https://arxiv.org/abs/2309.11811

work page arXiv 2023
[23]

Fast Initial Access with Deep Learning for Beam Prediction in 5G mmWave Networks,

T. S. Cousik, V. K. Shah, J. H. Reed et al., “Fast Initial Access with Deep Learning for Beam Prediction in 5G mmWave Networks,” in MILCOM, 2021, pp. 664–669

2021
[24]

Integrated Probing-Beam Pattern Learning and Beam Prediction for mmWave Massive MIMO,

Q. Xue et al., “Integrated Probing-Beam Pattern Learning and Beam Prediction for mmWave Massive MIMO,” IEEE Trans. Commun., vol. 73, no. 8, pp. 6499–6513, 2025

2025
[25]

Learning Site-Specific Probing Beams for Fast mmWave Beam Alignment,

Y. Heng, J. Mo, and J. G. Andrews, “Learning Site-Specific Probing Beams for Fast mmWave Beam Alignment,” IEEE Trans. Wireless Commun., vol. 21, no. 8, pp. 5785–5800, 2022

2022
[26]

Deep Learning Assisted mmWave Beam Prediction with Prior Low-frequency Informa- tion,

K. Ma, D. He, H. Sun, and Z. Wang, “Deep Learning Assisted mmWave Beam Prediction with Prior Low-frequency Informa- tion,” in IEEE ICC, 2021, pp. 1–6

2021
[27]

Deep Learning- Based Beam Tracking for Millimeter-Wave Communications Under Mobility,

S. H. Lim, S. Kim, B. Shim, and J. W. Choi, “Deep Learning- Based Beam Tracking for Millimeter-Wave Communications Under Mobility,” IEEE Trans. Commun., vol. 69, no. 11, pp. 7458–7469, 2021

2021
[28]

Multi-Cell Multi-Beam Predic- tion Using Auto-Encoder LSTM for mmWave Systems,

S. H. A. Shah and S. Rangan, “Multi-Cell Multi-Beam Predic- tion Using Auto-Encoder LSTM for mmWave Systems,” IEEE Trans. Wireless Commun., vol. 21, no. 12, pp. 10 366–10 380, 2022

2022
[29]

Machine Learning Based Time Domain Millimeter-Wave Beam Prediction for 5G-Advanced and Be- yond: Design, Analysis, and Over-The-Air Experiments,

Q. Li et al., “Machine Learning Based Time Domain Millimeter-Wave Beam Prediction for 5G-Advanced and Be- yond: Design, Analysis, and Over-The-Air Experiments,” IEEE J. Sel. Areas Commun., vol. 41, no. 6, pp. 1787–1809, 2023

2023
[30]

Continuous- Time mmWave Beam Prediction With ODE-LSTM Learning Architecture,

K. Ma, F. Zhang, W. Tian, and Z. Wang, “Continuous- Time mmWave Beam Prediction With ODE-LSTM Learning Architecture,” IEEE Wireless Commun. Lett., vol. 12, no. 1, pp. 187–191, 2023

2023
[31]

Generative AI for the Opti- mization of Next-Generation Wireless Networks: Basics, State- of-the-Art, and Open Challenges,

F. Khoramnejad and E. Hossain, “Generative AI for the Opti- mization of Next-Generation Wireless Networks: Basics, State- of-the-Art, and Open Challenges,” IEEE Commun. Surveys Tuts., pp. 1–1, 2025

2025
[32]

Beam Tracking for High-Speed UA V via Generative Diffusion Model-Enabled Joint Optimization Ap- proach,

J. Zhang et al., “Beam Tracking for High-Speed UA V via Generative Diffusion Model-Enabled Joint Optimization Ap- proach,” IEEE Trans. Veh. Technol., vol. 74, no. 9, pp. 14 054– 14 068, 2025

2025
[33]

Echo-Conditioned Denoising Diffusion Probabilistic Models for Multi-Target Tracking in RF Sensing,

A. Azarbahram and O. L. A. López, “Echo-Conditioned Denoising Diffusion Probabilistic Models for Multi-Target Tracking in RF Sensing,” ICC 2026. [Online]. A vailable: https://arxiv.org/abs/2510.25464

work page arXiv 2026
[34]

Enhanced Secure Beamforming for IRS- Assisted IoT Communication Using a Generative-Diffusion- Model-Enabled Optimization Approach,

J. Zhang et al., “Enhanced Secure Beamforming for IRS- Assisted IoT Communication Using a Generative-Diffusion- Model-Enabled Optimization Approach,” IEEE Internet Things J., vol. 12, no. 10, pp. 13 398–13 414, 2025

2025
[35]

Coordinated Downlink Beamforming in Multi- Cell MIMO Networks: A Diffusion Model-Enhanced Multi- Agent Reinforcement Learning Perspective,

H. Liu et al., “Coordinated Downlink Beamforming in Multi- Cell MIMO Networks: A Diffusion Model-Enhanced Multi- Agent Reinforcement Learning Perspective,” IEEE Trans. Wireless Commun., vol. 25, pp. 7617–7634, 2026

2026
[36]

Beam-brainstorm: A generative site- specific beamforming approach,

Z. Zhou, Z. Wang, and Y. Liu, “Beam-Brainstorm: A Generative Site-Specific Beamforming Approach,” 2026. [Online]. A vailable: https://arxiv.org/abs/2601.02219

work page arXiv 2026
[37]

Leveraging Generative Diffusion Models for Enhanced Beam Alignment in Cell-Free MIMO Systems,

J. Zhang et al., “Leveraging Generative Diffusion Models for Enhanced Beam Alignment in Cell-Free MIMO Systems,” in ICCCN, 2025, pp. 1–6

2025
[38]

Beam Prediction Based on Large Language Models,

Y. Sheng et al., “Beam Prediction Based on Large Language Models,” IEEE Wireless Commun. Lett., vol. 14, no. 5, pp. 1406–1410, 2025

2025
[39]

Generative Diffusion Model-Based Variational Inference for MIMO Channel Estima- tion,

Z. Chen, H. Shin, and A. Nallanathan, “Generative Diffusion Model-Based Variational Inference for MIMO Channel Estima- tion,” IEEE Trans. Commun., vol. 73, no. 10, pp. 9254–9269, 2025

2025
[40]

Generative Diffusion Models for High Dimen- sional Channel Estimation,

X. Zhou et al., “Generative Diffusion Models for High Dimen- sional Channel Estimation,” IEEE Trans. Wireless Commun., vol. 24, no. 7, pp. 5840–5854, 2025

2025
[41]

Diffusion-Based Generative Prior for Low- Complexity MIMO Channel Estimation,

B. Fesl et al., “Diffusion-Based Generative Prior for Low- Complexity MIMO Channel Estimation,” IEEE Wireless Com- mun. Lett., vol. 13, no. 12, pp. 3493–3497, 2024

2024
[42]

Structured Denoising Diffusion Models in Dis- crete State-Spaces,

J. Austin et al., “Structured Denoising Diffusion Models in Dis- crete State-Spaces,” in NeurIPS, vol. 34. Curran Associates, Inc., 2021, pp. 17 981–17 993

2021
[43]

Beam squint and channel estimation for wideband mmWave massive MIMO-OFDM systems,

B. Wang, et al., “Beam squint and channel estimation for wideband mmWave massive MIMO-OFDM systems,” IEEE Trans. Signal Process., vol. 67, no. 23, pp. 5893–5908, 2019

2019
[44]

Tse and P

D. Tse and P. Viswanath, Fundamentals of Wireless Commu- nication. Cambridge Univ. Press, 2005

2005
[45]

Solving POMDPs by Searching the Space of Finite Policies,

N. Meuleau, K.-E. Kim, L. P. Kaelbling, and A. R. Cassandra, “Solving POMDPs by Searching the Space of Finite Policies,”
[46]

A vailable: https://arxiv.org/abs/1301.6720

[Online]. A vailable: https://arxiv.org/abs/1301.6720

work page arXiv
[47]

Kukučka et al

M. Ilse, J. M. Tomczak, and M. Welling, “Attention-based Deep Multiple Instance Learning,” arXiv preprint arXiv:1802.04712, 2018

work page arXiv 2018
[48]

Diffusion Models Beat GANs on Image Synthesis,

P. Dhariwal and A. Nichol, “Diffusion Models Beat GANs on Image Synthesis,” in NeurIPS, vol. 34. Curran Associates, Inc., 2021, pp. 8780–8794

2021
[49]

Classifier-Free Diffusion Guidance,

J. Ho and T. Salimans, “Classifier-Free Diffusion Guidance,”
[50]

Classifier-Free Diffusion Guidance

[Online]. A vailable: https://arxiv.org/abs/2207.12598

work page internal anchor Pith review Pith/arXiv arXiv
[51]

Eﬀicient diffusion models: A survey,

H. Shen et al., “Eﬀicient diffusion models: A survey,” 2025. [Online]. A vailable: https://arxiv.org/abs/2502.06805

work page arXiv 2025
[52]

R. S. Sutton, A. G. Barto et al., Reinforcement Learning: An Introduction. MIT Press, 1998, vol. 1, no. 1

1998
[53]

Finite-time analysis of the multiarmed bandit problem.Mach

P. Auer, N. Cesa-Bianchi, and P. Fischer, “Finite-time Analysis of the Multiarmed Bandit Problem,” Mach. Learn., vol. 47, no. 2, pp. 235–256, 2002. [Online]. A vailable: https://doi.org/10.1023/A:1013689704352

work page doi:10.1023/a:1013689704352 2002
[54]

DeepMIMO: A Generic Deep Learning Dataset for Millimeter Wave and Massive MIMO Applications

A. Alkhateeb, “DeepMIMO: A Generic Deep Learning Dataset for Millimeter Wave and Massive MIMO Applications,” 2019. [Online]. A vailable: https://arxiv.org/abs/1902.06435

work page Pith review arXiv 2019
[55]

Survey of Maneuvering Target Tracking. Part I. Dynamic Models,

X. R. Li and V. P. Jilkov, “Survey of Maneuvering Target Tracking. Part I. Dynamic Models,” IEEE Trans. Aerosp. Electron. Syst., vol. 39, no. 4, pp. 1333–1364, 2003

2003