pith. sign in

arxiv: 2606.23730 · v1 · pith:NKVZEQQYnew · submitted 2026-06-20 · 📡 eess.SP

Physics-Informed Path-Parametric Learning for Efficient and Lightweight CSI Feedback

Chunyu Ling , Jiajia Guo , Yiming Cui , Chao-Kai Wen , Shi Jin , Shuangfeng Han , Xiaoyun Wang This is my paper

Pith reviewed 2026-06-26 11:48 UTC · model grok-4.3

classification 📡 eess.SP
keywords CSI feedbackPhysics-informed neural networkMultipath parameter estimationHierarchical sensingWireless communicationsChannel state informationDeep learningXL-MIMO
0
0 comments X

The pith

HS-PINNnet reformulates CSI reconstruction as multipath parameter estimation to reduce feedback overhead and computation.

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

The paper introduces HS-PINNnet to solve high overhead in CSI feedback for wireless systems by embedding a multipath channel model directly into the neural network. This turns the problem from high-dimensional image-like reconstruction into low-dimensional estimation of parameters such as amplitude and angle. A hierarchical sensing encoder creates a compact representation, while a heterogeneous decoder with dedicated branches handles parameter-specific reconstruction. An adaptive PCD module estimates the number of dominant paths per sample, and subchannel-wise encoding with parallel decoding lowers training difficulty for large-scale systems. Simulations show the approach beats prior methods while cutting FLOPs by 92.8 percent and FPGA latency by two orders of magnitude.

Core claim

HS-PINNnet integrates a multipath channel model into the network, reformulating high-dimensional CSI reconstruction as low-dimensional multipath parameter estimation. It features a hierarchical sensing encoder to produce a compact multipath representation, and a heterogeneous decoder for parameter-specific CSI reconstruction with dedicated branches to estimate different parameters. A PCD module adaptively estimates the number of dominant paths in each CSI sample, and a subchannel-wise shared encoding and parallel decoding strategy decomposes high-dimensional CSI processing into low-dimensional subchannel tasks.

What carries the argument

HS-PINNnet, which embeds a multipath channel model into a hierarchical sensing physics-informed neural network to convert CSI reconstruction into multipath parameter estimation.

If this is right

  • HS-PINNnet achieves a 92.8 percent reduction in FLOPs compared with state-of-the-art CSI feedback methods.
  • The design exhibits two orders of magnitude lower FPGA simulation latency.
  • The PCD module improves generalization across different channel environments.
  • Subchannel-wise processing makes the method scalable to XL-MIMO systems.

Where Pith is reading between the lines

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

  • The parameter-based approach could allow direct fusion with ray-tracing or geometry-based channel models in future network simulators.
  • Lower computational cost may enable real-time CSI feedback on resource-constrained devices such as IoT sensors.
  • The interpretability gain might simplify regulatory or standardization discussions around learned CSI codecs.

Load-bearing premise

Reformulating CSI reconstruction as multipath parameter estimation will produce outputs consistent with multipath propagation principles in diverse real-world environments.

What would settle it

Measured CSI from a real propagation environment where HS-PINNnet's reconstructed values deviate from the actual multipath components by more than the error margin of conventional methods.

Figures

Figures reproduced from arXiv: 2606.23730 by Chao-Kai Wen, Chunyu Ling, Jiajia Guo, Shi Jin, Shuangfeng Han, Xiaoyun Wang, Yiming Cui.

Figure 1
Figure 1. Figure 1: Illustration of conventional CSI feedback and the [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The architecture of HS-PINNnet. The encoder is designed with a hierarchical sensing convolution mechanism, with [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The difference between conventional convolution and [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Parameter distribution of the simulations’ scenario. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Accumulated error of µ mismatch (µ = π/2, Nt = 32) Subcarrier index R 0 10 20 30 40 50 60 70 e l a t i v e e r r o r m a g n i t u d e j" H = Hj 0 0.2 0.4 0.6 0.8 1 Exact, 10% 1st-order, 10% Exact, 5% 1st-order, 5% Exact, 1% 1st-order, 1% Exact, 0% 1st-order, 0% [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Accumulated error of τ mismatch (τ = 1.5 × 10−7 , Nc = 64) introduced in Section IV. Finally, the branch outputs are concatenated into a structured parameter matrix Pˆ ∈ R L×4 , representing the recovered multipath parameters. These param￾eters are then used to calculate the downlink CSI through the channel model in (6). 3) Theoretical Analysis of Multipath Parameter Estimation Errors: To quantitatively ch… view at source ↗
Figure 7
Figure 7. Figure 7: The process of guidance and refinement training strategy. STEP1 guides HS-PINNnet to learn the map between CSI [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: The PCD module estimates the number of dominant [PITH_FULL_IMAGE:figures/full_fig_p007_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: The illustration of subchannel-wise shared encoding and parallel decoding strategy for large BS arrays. [PITH_FULL_IMAGE:figures/full_fig_p008_9.png] view at source ↗
Figure 11
Figure 11. Figure 11: Comparison of CSI reconstruction accuracy of differ [PITH_FULL_IMAGE:figures/full_fig_p010_11.png] view at source ↗
Figure 10
Figure 10. Figure 10: Channel reconstruction visualization of different [PITH_FULL_IMAGE:figures/full_fig_p010_10.png] view at source ↗
Figure 12
Figure 12. Figure 12: Comparison of CSI reconstruction accuracy when [PITH_FULL_IMAGE:figures/full_fig_p011_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: CSI reconstruction performance under different quan [PITH_FULL_IMAGE:figures/full_fig_p012_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: (a) Total FPGA inference latency of different methods. The size of the scatter point is proportional to the inference [PITH_FULL_IMAGE:figures/full_fig_p013_14.png] view at source ↗
read the original abstract

Channel State Information (CSI) feedback is vital for high spectral efficiency in wireless systems, yet high-dimensional CSI introduce significant feedback overhead. Recent deep learning (DL) approaches alleviate this issue by treating CSI as a visual image, but such "black-box" designs often lack interpretability, producing CSI that is not consistent with multipath propagation principles. To address these limitations, this paper proposes HS-PINNnet, a Hierarchical Sensing mechanism assisted Physics-Informed Neural Network for CSI Feedback. Unlike vision-inspired methods, HS-PINNnet integrates a multipath channel model into the network, reformulating high-dimensional CSI reconstruction as low-dimensional multipath parameter estimation (e.g., amplitude, angle). HS-PINNnet features a hierarchical sensing encoder to produce a compact multipath representation, and a heterogeneous decoder for parameter-specific CSI reconstruction, with dedicated branches to estimate different parameters. Moreover, a PCD module adaptively estimates the number of dominant paths in each CSI sample to enhance generalization across diverse environments. A subchannel-wise shared encoding and parallel decoding strategy is further designed to decompose high-dimensional CSI processing into low-dimensional subchannel tasks, reducing training difficulty and improving scalability of HS-PINNnet for future extremely large-scale multiple-input multiple-output (XL-MIMO) systems. Simulation results show that HS-PINNnet outperforms the state-of-the-art under different configurations, achieving a 92.8% reduction in FLOPs and exhibiting two orders of magnitude lower FPGA simulation latency.

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.

Referee Report

0 major / 3 minor

Summary. The paper proposes HS-PINNnet, a hierarchical sensing physics-informed neural network for CSI feedback. It integrates a multipath channel model to reformulate high-dimensional CSI reconstruction as low-dimensional multipath parameter estimation (amplitude, angle, etc.), using a hierarchical encoder, heterogeneous decoder with parameter-specific branches, a PCD module for adaptive estimation of dominant paths per sample, and subchannel-wise shared encoding/parallel decoding for scalability in XL-MIMO. Simulations claim outperformance over SOTA with 92.8% FLOPs reduction and two orders of magnitude lower FPGA latency.

Significance. If the simulation results hold under the reported conditions, the work provides a more interpretable alternative to black-box vision-inspired DL methods for CSI feedback by grounding the architecture in multipath propagation principles, with potential efficiency benefits for high-dimensional systems.

minor comments (3)
  1. The abstract states that HS-PINNnet 'outperforms the state-of-the-art under different configurations' but does not name the specific baselines or configurations; this should be clarified with explicit references to prior methods and parameter settings.
  2. The claim of consistency with multipath propagation principles is central but rests on simulation results; the manuscript should include a dedicated section or figure quantifying how closely the reconstructed CSI matches independent multipath model predictions (e.g., via path parameter error metrics) beyond aggregate NMSE.
  3. The PCD module's adaptive path-number estimation is described as enhancing generalization, but the training procedure for this module and any associated loss terms are not detailed in the provided abstract; add explicit description of the loss formulation and training schedule.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive review, recognition of the interpretability benefits of grounding the architecture in multipath propagation, and the recommendation for minor revision. No specific major comments were listed in the report.

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper's core contribution is an architectural design (hierarchical encoder, heterogeneous decoder, PCD module, subchannel-wise processing) that explicitly incorporates a multipath channel model as an external domain-knowledge constraint to reformulate CSI feedback as parameter estimation. Performance claims rest on simulation benchmarks comparing against baselines, with no equations or derivations shown that reduce a claimed prediction or uniqueness result back to a fitted parameter or self-citation by construction. The multipath integration is a modeling choice whose validity is tested externally via those simulations rather than assumed tautologically. No load-bearing self-citations, ansatz smuggling, or renaming of known results appear in the abstract or described structure.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The approach depends on the standard domain assumption of multipath modeling in wireless channels and the effectiveness of the neural network in estimating those parameters from compressed representations.

free parameters (1)
  • Estimated multipath parameters (amplitude, angle, etc.)
    These are learned by the network branches rather than fixed a priori.
axioms (1)
  • domain assumption Wireless channels follow a multipath propagation model that can be parameterized by amplitude, angle, and similar quantities.
    The paper reformulates the CSI reconstruction task based on this model to achieve consistency with physical principles.

pith-pipeline@v0.9.1-grok · 5810 in / 1255 out tokens · 35138 ms · 2026-06-26T11:48:01.888818+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

44 extracted references · 1 canonical work pages

  1. [1]

    Physics-Informed Neural Networks for Wireless CSI Feedback,

    C. Linget al., “Physics-Informed Neural Networks for Wireless CSI Feedback,”Proc. IEEE ICC, pp. 1–6, 2026

    2026

  2. [2]

    5G-Advanced Toward 6G: Past, Present, and Future,

    W. Chenet al., “5G-Advanced Toward 6G: Past, Present, and Future,” IEEE J. Sel. Areas Commun., vol. 41, no. 6, pp. 1592–1619, 2023

    2023

  3. [3]

    A Speculative Study on 6G,

    F. Tariq, M. R. A. Khandaker, K.-K. Wong, M. A. Imran, M. Bennis, and M. Debbah, “A Speculative Study on 6G,”IEEE Wireless Commun., vol. 27, no. 4, pp. 118–125, 2020

    2020

  4. [4]

    Overview of Deep Learning- Based CSI Feedback in Massive MIMO Systems,

    J. Guo, C.-K. Wen, S. Jin, and G. Y . Li, “Overview of Deep Learning- Based CSI Feedback in Massive MIMO Systems,”IEEE Trans. Com- mun., vol. 70, no. 12, pp. 8017–8045, 2022

    2022

  5. [5]

    Physical Layer Procedures for Data (Release 17),

    3GPP, “Physical Layer Procedures for Data (Release 17),” 3rd Generation Partnership Project (3GPP), Technical Specification (TS) 38.214, 2022, Version 17.3.0. [Online]. Available: https://www.3gpp.org

    2022

  6. [6]

    Sparse Representation for Wireless Communications: A Compressive Sensing Approach,

    Z. Qin, J. Fan, Y . Liu, Y . Gao, and G. Y . Li, “Sparse Representation for Wireless Communications: A Compressive Sensing Approach,”IEEE Signal Process. Mag., vol. 35, no. 3, pp. 40–58, 2018

    2018

  7. [7]

    Remote State Estimation of Nonlinear Systems Over Fading Channels via Recurrent Neural Networks,

    S. Cai and V . K. N. Lau, “Remote State Estimation of Nonlinear Systems Over Fading Channels via Recurrent Neural Networks,”IEEE Trans. Neural Netw. Learn. Syst., vol. 33, no. 8, pp. 3908–3922, 2022

    2022

  8. [8]

    Graph Neural Networks for Wireless Communications: From Theory to Practice,

    Y . Shen, J. Zhang, S. H. Song, and K. B. Letaief, “Graph Neural Networks for Wireless Communications: From Theory to Practice,” IEEE Trans. Wireless Commun., vol. 22, no. 5, pp. 3554–3569, 2023

    2023

  9. [9]

    Deep Learning- Based Design of Uplink Integrated Sensing and Communication,

    Q. Qi, X. Chen, C. Zhong, C. Yuen, and Z. Zhang, “Deep Learning- Based Design of Uplink Integrated Sensing and Communication,”IEEE Trans. Wireless Commun., vol. 23, no. 9, pp. 10 639–10 652, 2024

    2024

  10. [10]

    Graph Neural Network-Based Node Deployment for Throughput Enhancement,

    Y . Yang, D. Zou, and X. He, “Graph Neural Network-Based Node Deployment for Throughput Enhancement,”IEEE Trans. Neural Netw. Learn. Syst., vol. 35, no. 10, pp. 14 810–14 824, 2024

    2024

  11. [11]

    Prompt-Enabled Large AI Models for CSI Feedback,

    J. Guo, Y . Cui, C.-K. Wen, and S. Jin, “Prompt-Enabled Large AI Models for CSI Feedback,”IEEE J. Sel. Areas Commun., vol. 44, pp. 2654–2668, 2026

    2026

  12. [12]

    Towards CSI Acquisition in 6G: A 3GPP Perspective,

    Y . Cui, C. Jiang, C. Ling, J. Liu, J. Guo, C.-K. Wen, and S. Jin, “Towards CSI Acquisition in 6G: A 3GPP Perspective,” 2026, dOI: 10.13140/RG.2.2.12766.19527. [On- line]. Available: https://www.researchgate.net/publication/405232078 Towards CSI Acquisition in 6G A 3GPP Perspective

    work page doi:10.13140/rg.2.2.12766.19527 2026

  13. [13]

    Deep Learning for Massive MIMO CSI Feedback,

    C.-K. Wen, W.-T. Shih, and S. Jin, “Deep Learning for Massive MIMO CSI Feedback,”IEEE Wireless Commun. Lett., vol. 7, no. 5, pp. 748– 751, 2018

    2018

  14. [14]

    Convolutional Neural Network- Based Multiple-Rate Compressive Sensing for Massive MIMO CSI Feedback: Design, Simulation, and Analysis,

    J. Guo, C.-K. Wen, S. Jin, and G. Y . Li, “Convolutional Neural Network- Based Multiple-Rate Compressive Sensing for Massive MIMO CSI Feedback: Design, Simulation, and Analysis,”IEEE Trans. Wireless Commun., vol. 19, no. 4, pp. 2827–2840, 2020

    2020

  15. [15]

    Multi-resolution CSI Feedback with Deep Learning in Massive MIMO System,

    Z. Lu, J. Wang, and J. Song, “Multi-resolution CSI Feedback with Deep Learning in Massive MIMO System,” inProc. IEEE ICC, 2020, pp. 1–6

    2020

  16. [16]

    Going Deeper With Convolutions,

    C. Szegedyet al., “Going Deeper With Convolutions,” inProc. IEEE/CVF CVPR, 2015, pp. 1–9

    2015

  17. [17]

    DualConv: Dual Convolutional Kernels for Lightweight Deep Neural Networks,

    J. Zhong, J. Chen, and A. Mian, “DualConv: Dual Convolutional Kernels for Lightweight Deep Neural Networks,”IEEE Trans. Neural Netw. Learn. Syst., vol. 34, no. 11, pp. 9528–9535, 2023

    2023

  18. [18]

    Rethinking the Inception Architecture for Computer Vision,

    C. Szegedy, V . Vanhoucke, S. Ioffe, J. Shlens, and Z. Wojna, “Rethinking the Inception Architecture for Computer Vision,” inProc. IEEE/CVF CVPR, 2016, pp. 2818–2826

    2016

  19. [19]

    Transformer Empowered CSI Feed- back for Massive MIMO Systems,

    Y . Xu, M. Yuan, and M.-O. Pun, “Transformer Empowered CSI Feed- back for Massive MIMO Systems,” inProc. IEEE WOCC, 2021, pp. 157–161

    2021

  20. [20]

    A Novel Approach Using Convolu- tional Transformer for Massive MIMO CSI Feedback,

    X. Bi, S. Li, C. Yu, and Y . Zhang, “A Novel Approach Using Convolu- tional Transformer for Massive MIMO CSI Feedback,”IEEE Wireless Commun. Lett., vol. 11, no. 5, pp. 1017–1021, 2022

    2022

  21. [21]

    TransNet: Full Attention Network for CSI Feedback in FDD Massive MIMO System,

    Y . Cui, A. Guo, and C. Song, “TransNet: Full Attention Network for CSI Feedback in FDD Massive MIMO System,”IEEE Wireless Commun. Lett., vol. 11, no. 5, pp. 903–907, 2022

    2022

  22. [22]

    A Spatially Separable Attention Mechanism for Massive MIMO CSI Feedback,

    S. Mourya, S. Amuru, and K. K. Kuchi, “A Spatially Separable Attention Mechanism for Massive MIMO CSI Feedback,”IEEE Wireless Commun. Lett., vol. 12, no. 1, pp. 40–44, 2023

    2023

  23. [23]

    Attention is All you Need,

    A. Vaswaniet al., “Attention is All you Need,” inProc. NeurIPS, vol. 30, 2017

    2017

  24. [24]

    A Survey of Visual Transformers,

    Y . Liuet al., “A Survey of Visual Transformers,”IEEE Trans. Neural Netw. Learn. Syst., vol. 35, no. 6, pp. 7478–7498, 2024

    2024

  25. [25]

    Deep Learning-Based CSI Feedback Approach for Time-Varying Massive MIMO Channels,

    T. Wang, C.-K. Wen, S. Jin, and G. Y . Li, “Deep Learning-Based CSI Feedback Approach for Time-Varying Massive MIMO Channels,”IEEE Wireless Commun. Lett., vol. 8, no. 2, pp. 416–419, 2019

    2019

  26. [26]

    Compressive Sampled CSI Feedback Method Based on Deep Learning for FDD Massive MIMO Systems,

    J. Wang, G. Gui, T. Ohtsuki, B. Adebisi, H. Gacanin, and H. Sari, “Compressive Sampled CSI Feedback Method Based on Deep Learning for FDD Massive MIMO Systems,”IEEE Trans. Commun., vol. 69, no. 9, pp. 5873–5885, 2021

    2021

  27. [27]

    Learning How to Transfer From Uplink to Downlink via Hyper-Recurrent Neural Network for FDD Massive MIMO,

    Y . Liu and O. Simeone, “Learning How to Transfer From Uplink to Downlink via Hyper-Recurrent Neural Network for FDD Massive MIMO,”IEEE Trans. Wireless Commun., vol. 21, no. 10, pp. 7975– 7989, 2022

    2022

  28. [28]

    Physics-Inspired Deep Learning Anti-Aliasing Framework in Efficient Channel State Feedback,

    Y . Lin, Y . Xin, T. Lee, C. Zhang, and Z. Ding, “Physics-Inspired Deep Learning Anti-Aliasing Framework in Efficient Channel State Feedback,”IEEE Trans. Wireless Commun., vol. 24, no. 2, pp. 1117– 1131, 2024

    2024

  29. [29]

    Physics-Informed Data Augmentation for CSI Data-Limited Deep Learning-Driven Communication Tasks,

    M. Yuan, L. Lian, and S. Ji, “Physics-Informed Data Augmentation for CSI Data-Limited Deep Learning-Driven Communication Tasks,” in Proc. IEEE ICC Workshops, 2025, pp. 1634–1639

    2025

  30. [30]

    Deep Learning- Based CSI Feedback for RIS-Assisted Multi-User Systems,

    J. Guo, X. Yang, C.-K. Wen, S. Jin, and G. Ye Li, “Deep Learning- Based CSI Feedback for RIS-Assisted Multi-User Systems,”IEEE Trans. Commun., vol. 73, no. 7, pp. 4974–4989, 2025

    2025

  31. [31]

    MuSAC: Mutualistic Sensing and Communication for Mobile Crowdsensing,

    S. Ji, L. Lian, Y . Zheng, and C. Wu, “MuSAC: Mutualistic Sensing and Communication for Mobile Crowdsensing,” inProc. IEEE ICDCS, 2024, pp. 243–254

    2024

  32. [32]

    CSI-StripeFormer: Exploiting Stripe Features for CSI Compression in Massive MIMO System,

    Q. Hu, H. Kang, H. Chen, Q. Huang, Q. Zhang, and M. Cheng, “CSI-StripeFormer: Exploiting Stripe Features for CSI Compression in Massive MIMO System,” inProc. IEEE INFOCOM, 2023, pp. 1–10

    2023

  33. [33]

    Newtonized Or- thogonal Matching Pursuit: Frequency Estimation Over the Continuum,

    B. Mamandipoor, D. Ramasamy, and U. Madhow, “Newtonized Or- thogonal Matching Pursuit: Frequency Estimation Over the Continuum,” IEEE Trans. Signal Process., vol. 64, no. 19, p. 5066–5081, 2016

    2016

  34. [34]

    Transformer-Assisted Parametric CSI Feedback for mmWave Massive MIMO Systems,

    H. Ju, S. Jeong, S. Kim, B. Lee, and B. Shim, “Transformer-Assisted Parametric CSI Feedback for mmWave Massive MIMO Systems,”IEEE Trans. Wireless Commun., vol. 23, no. 12, pp. 18 774–18 787, 2024

    2024

  35. [35]

    Foundation Models for Wireless Communications: From PHY Intelligence to Network Autonomy,

    L. Lianget al., “Foundation Models for Wireless Communications: From PHY Intelligence to Network Autonomy,”arXiv:2606.06239, 2026

    Pith/arXiv arXiv 2026

  36. [36]

    Deformable Convolutional Networks,

    J. Dai, H. Qi, Y . Xiong, Y . Li, G. Zhang, H. Hu, and Y . Wei, “Deformable Convolutional Networks,” inProc. IEEE ICCV, 2017, pp. 764–773

    2017

  37. [37]

    Large Kernel Matters — Improve Semantic Segmen- tation by Global Convolutional Network,

    C. Penget al., “Large Kernel Matters — Improve Semantic Segmen- tation by Global Convolutional Network,” inProc. IEEE/CVF CVPR, 2017, pp. 1743–1751

    2017

  38. [38]

    Receptive Fields as Experts in Convolu- tional Neural Architectures,

    D. Lian, W. Yu, and X. Wang, “Receptive Fields as Experts in Convolu- tional Neural Architectures,” inProc. ICML, 2024, pp. 29 531–29 544

    2024

  39. [39]

    LSNet: See Large, Focus Small,

    A. Wang, H. Chen, Z. Lin, J. Han, and G. Ding, “LSNet: See Large, Focus Small,” inProc. IEEE/CVF CVPR, 2025, pp. 9718–9729

    2025

  40. [40]

    A Survey of Convolutional Neural Networks: Analysis, Applications, and Prospects,

    Z. Li, F. Liu, W. Yang, S. Peng, and J. Zhou, “A Survey of Convolutional Neural Networks: Analysis, Applications, and Prospects,”IEEE Trans. Neural Netw. Learn. Syst., vol. 33, no. 12, pp. 6999–7019, 2022

    2022

  41. [41]

    Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift,

    S. Ioffe and C. Szegedy, “Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift,” inProc. ICML, vol. 37, 2015, pp. 448–456

    2015

  42. [42]

    Deep Residual Learning for Image Recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep Residual Learning for Image Recognition,” inProc. IEEE/CVF CVPR, 2016, pp. 770–778

    2016

  43. [43]

    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,” inProc. IEEE ITA, 2019, pp. 1–8

    2019

  44. [44]

    Custom Hardware Architectures for Deep Learning on Portable Devices: A Review,

    K. S. Zamanet al., “Custom Hardware Architectures for Deep Learning on Portable Devices: A Review,”IEEE Trans. Neural Netw. Learn. Syst., vol. 33, no. 11, pp. 6068–6088, 2022

    2022