Pith. sign in

REVIEW 3 major objections 4 minor 51 references

Autonomous Discovery of Wireless Communications Algorithms

T0 review · 3 major / 4 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read An LLM-driven evolutionary search, called AITE, autonomously designs wireless receiver algorithms that beat known OTFS equalizers in speed and match neural-network receivers without pilots.

desk verdict The pilotless receiver is the real result; the framework-level claim needs repeated runs before it is established. read the letter →

arxiv 2607.17762 v1 pith:CO7WEVL5 submitted 2026-07-20 cs.IT cs.AIcs.MAmath.IT

Fayçal Aït Aoudia , Jakob Hoydis , Sebastian Cammerer , Gian Marti , Merlin Nimier-David , Nicolas Roussel , Alexander Keller This is my paper
classification cs.ITcs.AIcs.MAmath.IT
keywords algorithmdiscoveryLLM-drivenevolutionarysearchOTFSequalizationpilotlessOFDMexpectationpropagationblindchannelestimationperformance-complexitytradeoffwirelessphysicallayer
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 introduces AITE, a framework in which a large language model orchestrates an evolutionary search over wireless receiver algorithms, guided only by a task description and an evaluation tool that reports each candidate's error rate and latency. The paper reports two results: for an OTFS equalizer, AITE found algorithms that outperform the strong expectation-propagation baseline while running about 3.6 times faster; for a pilotless OFDM receiver, it found the first explicit, explainable algorithms that achieve performance on par with a state-of-the-art neural receiver, with no pilots and no trainable network. The authors' aim is to show that LLM-driven evolutionary search can autonomously produce competitive physical-layer algorithms, expanding the search to the performance–complexity frontier rather than a single solution. A sympathetic reader would care because it points toward a mode of algorithm design where a machine proposes, implements, and optimizes the code itself, and the human picks an operating point from a set of trade-off options.

What carries the argument

AITE is the central mechanism: an orchestrator LLM repeatedly proposes N algorithmic ideas spanning the accuracy–latency plane; a pool of tool-using worker agents independently implement each idea in isolated workspaces, iterating with the help of an evaluation tool that returns a scalar metric and a scalar complexity; the orchestrator then aggregates implementations, records a leaderboard, and generates the next round of ideas conditioned on Pareto-front and sampled off-front entries. The evaluation tool is deliberately immutable, preventing the search from gaming the score. For the two tasks, the crucial enabling operations are, respectively, the Doppler block-diagonalization that turns th

What would settle it

Re-running AITE on the same two tasks from different random seeds and with a different frontier LLM, and checking whether it again produces an OTFS equalizer with NVE below 1 and latency well below UAMP, and a pilotless receiver with goodput within a small margin of the neural receiver. If most runs fail to reach those thresholds, the framework-level claim falls even though the specific discovered algorithms stand.

Watch

Extended reading notes

Core claim

The central claim is that AITE, a two-tier LLM-driven evolutionary framework, can autonomously discover wireless physical-layer algorithms that are competitive with or better than hand-designed and learned baselines. On OTFS equalization, the generated algorithm builds on an expectation-consistency approximation, diagonalizes the channel along the Doppler axis, avoids matrix inversion via Cholesky and a single Richardson half-step, fuses three log-metrics before max-log demapping, and is tuned by Bayesian hyperparameter optimization; the paper reports a lower NVE than the reference and 3.6x lower latency than the strongest baseline UAMP. On the pilotless OFDM task, AITE discovered a receiver

Load-bearing premise

The reported headline results each come from a single AITE run with a single underlying LLM, so the paper's claim that AITE can reliably discover such algorithms presumes that these runs are typical, not a lucky draw.

Editorial extensions

If this is right

  • If AITE's results are correct, OTFS equalizers can be both more accurate and substantially faster than current message-passing and linear baselines.
  • Pilotless OFDM can be made explicit and non-neural, using a single learned constellation over all resource elements, which was previously only possible with a neural receiver.
  • The metric–complexity Pareto front gives system designers a menu of algorithms to pick from without further search.
  • LLM-driven search can be pointed at new physical-layer tasks by simply providing a task description and an evaluation tool, reducing manual engineering effort.
  • Generated algorithms come with hyperparameter tuning applied, making them deployable as code with measured latency and error-rate operating points.

Reading between the lines

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

  • AITE's discovered pilotless receiver suggests the neural receiver is not the only way to exploit a non-symmetric learned constellation; the moment-matching trick could be reused for other blind-estimation problems.
  • The framework's dependence on a single frontier LLM and a fixed evaluation metric implies that the search outcome may vary with model version or channel model; a validation tool that detects overfitting to the evaluation channel models would strengthen the method.
  • The latency of the generated pilotless receiver is higher than the neural receiver's on GPU, so a dedicated non-PyTorch implementation could close that gap.
  • Because the equalizer result came from one run, an obvious test is to repeat the search with different random seeds and compare the distribution of best NVE values.
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

3 major / 4 minor

Summary. The paper introduces AITE, an LLM-driven evolutionary search framework for autonomous discovery of wireless communications algorithms, and evaluates it on two physical-layer tasks: an OTFS equalizer design and a pilotless OFDM receiver. AITE generates candidate algorithms via agentic LLM workers, evaluates them with an immutable task-specific evaluation tool, and builds a Pareto front over a performance metric (NVE) and measured latency. For the OTFS task, the authors report that a selected generated equalizer achieves NVE 0.549 with a per-frame latency of 2.17 ms on an RTX PRO 6000, outperforming conventional baselines and running about 3.6x faster than the UAMP baseline. For the pilotless OFDM task, AITE discovers an explicit, non-neural receiver that achieves goodput parity with a neural receiver on the TDL-A and TDL-D channel models, using a learned constellation from prior work. The appendices give detailed descriptions of the two selected algorithms.

Significance. If the reported results are robust, this is a significant demonstration of LLM-driven evolutionary search producing competitive physical-layer algorithms, including a novel explicit pilotless receiver where only neural solutions previously existed. The paper is commendable for releasing the AITE code, providing substantial algorithmic detail in the appendices, and testing the pilotless receiver on two unseen channel models. The OTFS equalizer and the pilotless receiver are concrete, implementable algorithms. However, the central claim about AITE as a framework—that it reliably and autonomously discovers such algorithms—rests on a single run per task with one LLM and on selected Pareto-front points, so the framework-level conclusion is not yet established to the standard the headline claims imply.

major comments (3)
  1. [§V-A, Figs. 5 and 10, Tables 2 and 3] The framework-level claim that AITE autonomously discovers competitive algorithms is supported by exactly one run per task using GPT-5.5. The reported headline numbers are selected from the Pareto front of that single run, after Optuna tuning on the same NVE/latency objectives used for reporting. Section V-A explicitly states that 'Quantifying the run-to-run variability of the search outcome remains open.' Since selection over ~1,890 OTFS and ~1,240 pilotless candidates can inflate the best observed point under Monte Carlo noise, this is load-bearing. Please add multiple independent runs (e.g., different seeds or LLM temperatures) and report the distribution of NVE and latency, or clearly re-scope the claims to the specific discovered algorithms rather than to AITE's general capability.
  2. [§III-C/D and §V-B] The OTFS equalizer is evaluated only on the same channel model and SNR set S={13,16} dB used during the search. Unlike the pilotless receiver, which is tested on unseen TDL-A and TDL-D models, the OTFS evaluation has no holdout channel condition. The overfitting risk is acknowledged in §V-B, but no mitigation or quantification is provided for the OTFS task. Please add evaluation on unseen channel profiles or SNR points, or at least report the gap between search-time and holdout NVE to support the generalization implied by the title and abstract.
  3. [§III-C, Eq. (1), Figs. 6 and 11] NVE and coded BLER are reported without confidence intervals or the number of Monte Carlo frames/blocks simulated. Given the small SNR grid (two points) and the selection over thousands of candidates, the reported values—such as NVE=0.549 and the 3.6x speedup—may not be statistically stable. Please report BLER confidence intervals and the Monte Carlo sample size, at least for the final selected algorithms and baselines.
minor comments (4)
  1. [Abstract] Typo: 'Y et' should be 'Yet'.
  2. [Appendix B and C] Several implementation details (windowing, damping, reliability gating, exact hyperparameter values) are omitted and deferred to the repository. Since the appendices are meant to make the algorithms explainable, a short pointer to the exact file/function names in the repository would help readers verify the descriptions.
  3. [§II-C1] The hyperparameter tuning stage is central to the results, but the paper does not report the Optuna budget (number of trials per candidate) or the acceptance criteria for Pareto-optimal configurations. This information would strengthen reproducibility.
  4. [Fig. 6 caption] The caption mentions 'values in parentheses show the speedup relative to LMMSE,' but the figure as printed does not clearly show these values; please ensure the figure and caption match.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: reported algorithms are empirical search outputs evaluated by a fixed, immutable tool; the reused prior-work constellation is an input, not a target.

full rationale

The paper does not claim to derive its headline results from the task inputs; it reports the output of an evolutionary search. AITE is handed a task description and a fixed evaluation tool (Sections II.A and II.C.3), and the discovered algorithms are assessed by Monte Carlo simulations of the defined system models. The OTFS equalizer result is a selected Pareto-front candidate whose NVE and latency were measured by the evaluation tool, with all baselines receiving the same Optuna tuning; no fitted parameter is renamed as a prediction. The pilotless-receiver experiment explicitly takes the learned constellation from [21] as an input ('the constellation optimized through end-to-end learning, shown in Fig. 7, is reused unchanged') and targets the receiver only; the neural receiver from [21] serves as a baseline, not as an ingredient in AITE's search. The receiver's stages are derived from the stated system model and constellation-moment relations, and hyperparameters are openly tuned and reported as such. The self-citations to [1], [18], [21], and [24] provide provenance for the framework, prior constellation work, and simulation library; none is invoked as a load-bearing external theorem that forces the conclusions. Section V.A's admission that 'Quantifying the run-to-run variability of the search outcome remains open' is a reproducibility/correctness caveat about single-run selection, not evidence that the result was baked into the input. The central claims therefore have independent empirical content.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The free parameters are mostly algorithm hyperparameters tuned by Optuna on the evaluation metric; fitted values are not reported. The main axioms are domain assumptions about the channel and constellation structure that the discovered algorithms exploit, plus the unstated representativeness of a single stochastic LLM run. No invented physical entities are introduced; AITE is a software framework, not a new natural-kind entity.

free parameters (4)
  • OTFS equalizer hyperparameters (top-k sparsity, damping/relaxation, fusion weights, noise/LLR scalings)
    Selected by Optuna postrun tuning (Section II-C1); fitted values not reported. The headline NVE and latency depend on these values.
  • Pilotless receiver hyperparameters (moment window scales, Ntop, CRF weights, EM weights, BP weights, affine weights, dem
    Selected by Optuna postrun tuning; fitted values not reported. The parity claim depends on these hyperparameters.
  • AITE search configuration (LLM choice, 32 workers/generation, 60/42 generations, temperature T, prompt-refinement schedu
    Chosen by the authors rather than optimized; different choices could change search outcomes and the reported results.
  • Evaluation protocol (NVE SNR set S={13,16} dB, reference equalizer EP top-k=256, latency hardware)
    Defines the metric being optimized and the baseline for speedup/significance; the 3.6x latency claim is relative to this protocol and hardware.
assumptions (5)
  • standard math Hammersley-Clifford theorem and mean-field KL fixed-point update for the conditional random field over channel candidates
    Used in Appendix C-B to justify the pairwise MRF factorization and the iterative update (31)-(32).
  • domain assumption The OTFS effective channel is block-circulant along the Doppler index for the rectangular-CP waveform with continuous delay and Doppler, so an N-point DFT decouples the system into N MxM blocks
    Invoked in Section III-B and Appendix B-C1; underpins the latency reduction of the selected equalizer.
  • domain assumption The channel is approximately constant within the local time-frequency windows used for moment matching, CRF smoothing, EM, and affine correction
    Required in Appendix C-A/B/F; degrades under faster fading than the tested 0-3 m/s or larger delay spreads.
  • domain assumption The learned constellation C has nonzero higher moments for m in {2,3,4,6} and no rotational symmetry, so the m-fold phase ambiguity can be resolved by likelihood selection
    Appendix C-A, Eq. (28)-(29): the moment channel estimate divides by the constellation moment and relies on asymmetry of C.
  • ad hoc to paper A single GPT-5.5 run per task is representative of AITE's capability; stochastic LLM search outcomes are not repeated or quantified
    All headline results come from one run per task; Section V-A states run-to-run variability remains open.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Autonomous Discovery of Wireless Communications Algorithms." pith.science (2026). https://pith.science/paper/CO7WEVL5

@misc{pith2026260717762,
  author       = {Pith},
  title        = {Pith review of: Autonomous Discovery of Wireless Communications Algorithms},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CO7WEVL5}},
  note         = {Machine review of arXiv:2607.17762}
}
read the original abstract

Large language model (LLM)-driven evolutionary search is an emerging algorithm-discovery paradigm that has already produced novel results in several scientific fields. Yet its application to wireless communications remains largely unexplored. To bridge this gap, we introduce The AI Telco Engineer (AITE), a framework to autonomously design algorithms for complex communication problems, while navigating performance-complexity tradeoffs. We showcase AITE on two challenging physical-layer problems: designing an equalizer for an orthogonal time-frequency space (OTFS) system, and constructing a receiver algorithm for an orthogonal frequency-division multiplexing (OFDM) system using a custom constellation and operating without pilots. For the first task, AITE develops algorithms that outperform the best-known solutions while reducing computational latency by a factor of 3.6 compared to the strongest baseline. For the second task, it discovers the first explicit, explainable algorithms that achieve performance parity with state-of-the-art neural receivers. These results demonstrate the strong potential of LLM-driven evolutionary search for the autonomous discovery of next-generation wireless communications algorithms.

Figures

Figures reproduced from arXiv: 2607.17762 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_p004_3.png] view at source ↗
Figures from the paper (7 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_p007_5.png]
Figure 7
Figure 7. Figure 7: FIGURE 7 [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: contrasts the resource allocation in a conventional pilot-assisted grid and in the pilotless grid. In the former case, a subset P of REs carries known demodulation refer￾ence signal (DMRS) symbols. In the latter case, P = ∅ and all REs are occupied by data symbols. On …
Figure 9
Figure 9. Figure 9: (b). The neural receiver substitutes for channel esti￾mation, equalization, and demapping. It maps the received resource grid to LLRs. In this work, the neural receiver is a residual convolutional neural network as in [21]. On the transmitter side, only the constellati…
Figure 10
Figure 10. Figure 10: FIGURE 10 [PITH_FULL_IMAGE:figures/full_fig_p010_10.png]
Figure 11
Figure 11. Figure 11: FIGURE 11 [PITH_FULL_IMAGE:figures/full_fig_p011_11.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

51 extracted references · 6 linked inside Pith

  1. [1]

    The AI Telco Engineer: Toward autonomous discov ery of wireless communications algorithms,

    F. Aït Aoudia, J. Hoydis, S. Cammerer, L. Maggi, G. Marti, and A. Keller, “The AI Telco Engineer: Toward autonomous discov ery of wireless communications algorithms,” in European Wireless, Rimini, Italy, Jun. 2026. [Online]. Available: https://arxiv.org /abs/2604.19803

    arXiv 2026

  2. [2]

    J. R. Koza, Genetic Programming: On the Programming of Computers by Means of Natural Selection . Cambridge, MA: MIT Press, 1992

    1992

  3. [3]

    Evolution through large models,

    J. Lehman, J. Gordon, S. Jain, K. Ndousse, C. Y eh, and K. O. Stanley, “Evolution through large models,” in Handbook of Evolutionary Machine Learning . Springer Nature Singapore, 2024, pp. 331–366. [Online]. Available: https://arxiv.org/abs/2206.08896

    arXiv 2024

  4. [4]

    Language model crossover: V ariation through few-shot prompting,

    E. Meyerson et al. , “Language model crossover: V ariation through few-shot prompting,” ACM Trans. Evol. Learn. Optim. , vol. 4, no. 4, pp. 1–40, Nov. 2024. [Online]. Available: https://doi.org/10.1145/3694791

    doi:10.1145/3694791 2024

  5. [5]

    EvoPrompting: Languag e models for code-level neural architecture search,

    A. Chen, D. M. Dohan, and D. R. So, “EvoPrompting: Languag e models for code-level neural architecture search,” in Proc. 37th Conf. Neural Inf. Process. Syst. (NeurIPS) , New Orleans, LA, USA, Dec. 2023, pp. 7787–7817

    2023

  6. [6]

    Mathematical discoveries from program search with large language models,

    B. Romera-Paredes et al. , “Mathematical discoveries from program search with large language models,” Nature, vol. 625, no. 7995, pp. 468–475, Jan. 2024

    2024

  7. [7]

    AlphaEvolve: A coding agent for scientific and algorithmic discovery,

    A. Novikov et al. , “AlphaEvolve: A coding agent for scientific and algorithmic discovery,” arXiv preprint arXiv:2506.13131 , Jun. 2025. [Online]. Available: https://arxiv.org/abs/2506.13131

    arXiv 2025

  8. [8]

    OpenEvolve: An open-source evolutionary co ding agent,

    A. Sharma, “OpenEvolve: An open-source evolutionary co ding agent,” https://github.com/algorithmicsuperintelligence/openevolve , 2025

    2025

Show all 51 references
  1. [9]

    ShinkaEvolve: Towards open-ended and sample-efficient program evolution ,

    R. T. Lange, Y . Imajuku, and E. Cetin, “ShinkaEvolve: Towards open-ended and sample-efficient program evolution ,” arXiv preprint arXiv:2509.19349 , 2025. [Online]. Available: https://arxiv.org/abs/2509.19349

    2025 arXiv

  2. [10]

    GEPA: Reflective prompt evolution can outperform reinforcement learning,

    L. A. Agrawal et al. , “GEPA: Reflective prompt evolution can outperform reinforcement learning,” in Int. Conf. Learn. Representations (ICLR) , Rio de Janeiro, Brazil, Apr. 2026. [Online]. Available: https://openreview.net/forum?id=RQm2KQTM5r

    2026

  3. [11]

    SkyDiscover: A flexible, adaptive framework for AI- driven scientific and algorithmic discovery,

    S. Liu et al. , “SkyDiscover: A flexible, adaptive framework for AI- driven scientific and algorithmic discovery,” in Proc. ACM Conf. AI Agentic Syst. (CAIS’26) , San Jose, CA, USA, May 2026, pp. 1223– 1227

    2026

  4. [12]

    AdaEvolve: Adaptive LLM driven zeroth-order optimization,

    M. Cemri et al. , “AdaEvolve: Adaptive LLM driven zeroth-order optimization,” arXiv preprint arXiv:2602.20133 , Feb. 2026. [Online]. Available: https://arxiv.org/abs/2602.20133

    2026

  5. [13]

    EvoX: Meta-evolution for automated discovery,

    S. Liu et al. , “EvoX: Meta-evolution for automated discovery,” arXiv preprint arXiv:2602.23413 , Feb. 2026. [Online]. Available: https://arxiv.org/abs/2602.23413

    2026

  6. [14]

    A VO: Agentic variation operators for autonomous evolutionary search,

    T. Chen et al. , “A VO: Agentic variation operators for autonomous evolutionary search,” arXiv preprint arXiv:2603.24517 , Mar. 2026. [Online]. Available: https://arxiv.org/abs/2603.24517

    2026

  7. [15]

    LLM-guided search for deleti on-correcting codes,

    F. Weindel and R. Heckel, “LLM-guided search for deleti on-correcting codes,” Trans. Mach. Learn. Res. , Jun. 2026. [Online]. Available: https://openreview.net/forum?id=qZ69Ozpo6v

    2026

  8. [16]

    LLM-enabled automated a lgorithm design for multiuser fluid antenna communications,

    G. Zheng, F. Liu, and Q. Zhang, “LLM-enabled automated a lgorithm design for multiuser fluid antenna communications,” IEEE Trans. Wireless Commun. , May 2026, to appear. [Online]. Available: https://arxiv.org/abs/2605.14661

    2026 arXiv

  9. [17]

    GENESIS: Harnessing AI agents for autonomous 6G RAN synthesis, research, and testing,

    T. Aghayev et al. , “GENESIS: Harnessing AI agents for autonomous 6G RAN synthesis, research, and testing,” arXiv preprint arXiv:2605.27360 , May 2026. [Online]. Available: https://arxiv.org/abs/2605.27360

    2026 arXiv

  10. [18]

    The AI Telco Engineer,

    F. Aït Aoudia, J. Hoydis, S. Cammerer, L. Maggi, G. Marti, and A. Keller, “The AI Telco Engineer,” https://github.com/nvlabs/the-ai-telco-engineer, 202 6

  11. [19]

    LangChain,

    LangChain, Inc., “LangChain,” https://github.com/l angchain-ai/langchain, 2022, version 1.3.11

    2022

  12. [20]

    OTFS—a mathematical foundation for communication and rad ar sensing in the delay-Doppler domain,

    S. K. Mohammed, R. Hadani, A. Chockalingam, and R. Calde rbank, “OTFS—a mathematical foundation for communication and rad ar sensing in the delay-Doppler domain,” IEEE BITS Inf. Theory Mag. , vol. 2, no. 2, pp. 36–55, Nov. 2022

    2022

  13. [21]

    End-to-end learning for OF DM: From neural receivers to pilotless communication,

    F. Aït Aoudia and J. Hoydis, “End-to-end learning for OF DM: From neural receivers to pilotless communication,” IEEE Trans. Wireless Commun., vol. 21, no. 2, pp. 1049–1063, Feb. 2022

    2022

  14. [22]

    ReAct: Synergizing reasoning and acting in language models,

    S. Y ao et al. , “ReAct: Synergizing reasoning and acting in language models,” in Int. Conf. Learn. Representations (ICLR) , Kigali, Rwanda, May 2023. [Online]. Available: https://arxiv.org/abs/22 10.03629

    2023

  15. [23]

    Opt una: A next-generation hyperparameter optimization framework ,

    T. Akiba, S. Sano, T. Y anase, T. Ohta, and M. Koyama, “Opt una: A next-generation hyperparameter optimization framework ,” in Proc. 25th ACM SIGKDD Int. Conf. Knowl. Discovery Data Mining (KDD ), Anchorage, AK, USA, Aug. 2019, pp. 2623–2631

    2019

  16. [24]

    Hoydis et al

    J. Hoydis et al. , “Sionna,” https://nvlabs.github.io/sionna/, 2022, ver - sion 2.0.1

    2022

  17. [25]

    PyTorch 2: Faster machine learning through dynamic Python bytecode transformation and graph compilation,

    J. Ansel et al. , “PyTorch 2: Faster machine learning through dynamic Python bytecode transformation and graph compilation,” in ACM Int. Conf. Architectural Support Program. Lang. Operating Syst. (ASPLOS), La Jolla, CA, USA, May 2024, pp. 929–947. [Online]. Available: https://...

    2024

  18. [26]

    Billion-scale simi larity search with GPUs,

    J. Johnson, M. Douze, and H. Jégou, “Billion-scale simi larity search with GPUs,” IEEE Trans. Big Data , vol. 7, no. 3, pp. 535–547, Jul. 2021

    2021

  19. [27]

    On d eep learning-based channel decoding,

    T. Gruber, S. Cammerer, J. Hoydis, and S. ten Brink, “On d eep learning-based channel decoding,” in Proc. 51st Annu. Conf. Inf. Sci. Syst. (CISS) , Baltimore, MD, USA, May 2017, pp. 1–6

    2017

  20. [28]

    Interf erence cancellation and iterative detection for orthogonal time f requency space modulation,

    P . Raviteja, K. T. Phan, Y . Hong, and E. Viterbo, “Interf erence cancellation and iterative detection for orthogonal time f requency space modulation,” IEEE Trans. Wireless Commun. , vol. 17, no. 10, pp. 6501–6515, Oct. 2018

    2018

  21. [29]

    Ite rative detection for orthogonal time frequency space modulation w ith uni- tary approximate message passing,

    Z. Y uan, F. Liu, W. Y uan, Q. Guo, Z. Wang, and J. Y uan, “Ite rative detection for orthogonal time frequency space modulation w ith uni- tary approximate message passing,” IEEE Trans. Wireless Commun. , vol. 21, no. 2, pp. 714–725, Feb. 2022

    2022

  22. [30]

    Low com- plexity receiver via expectation propagation for OTFS modu lation,

    H. Li, Y . Dong, C. Gong, Z. Zhang, X. Wang, and X. Dai, “Low com- plexity receiver via expectation propagation for OTFS modu lation,” IEEE Commun. Lett. , vol. 25, no. 10, pp. 3180–3184, Oct. 2021

    2021

  23. [31]

    Low complexity LMMSE receiver for OTFS,

    S. Tiwari, S. S. Das, and V . Rangamgari, “Low complexity LMMSE receiver for OTFS,” IEEE Commun. Lett. , vol. 23, no. 12, pp. 2205– 2209, Dec. 2019

    2019

  24. [32]

    Low-complexity lin ear equal- ization for OTFS modulation,

    G. D. Surabhi and A. Chockalingam, “Low-complexity lin ear equal- ization for OTFS modulation,” IEEE Commun. Lett. , vol. 24, no. 2, pp. 330–334, Feb. 2020

    2020

  25. [33]

    V ector appr oximate message passing,

    S. Rangan, P . Schniter, and A. K. Fletcher, “V ector appr oximate message passing,” IEEE Trans. Inf. Theory , vol. 65, no. 10, pp. 6664– 6684, Oct. 2019

    2019

  26. [34]

    Orthogonal AMP,

    J. Ma and L. Ping, “Orthogonal AMP,” IEEE Access, vol. 5, pp. 2020– 2033, Jan. 2017

    2020

  27. [35]

    Expectation consistent appro ximate inference,

    M. Opper and O. Winther, “Expectation consistent appro ximate inference,” J. Mach. Learn. Res. , vol. 6, no. 73, pp. 2177–2204, Dec

  28. [36]

    Proximal splitting methods in signal processing,

    P . L. Combettes and J.-C. Pesquet, “Proximal splitting methods in signal processing,” in Fixed-Point Algorithms for Inverse Problems in Science and Engineering . New Y ork, NY , USA: Springer, 2011, pp. 185–212

    2011

  29. [37]

    S. Boyd, N. Parikh, E. Chu, B. Peleato, and J. Eckstein, “ Distributed optimization and statistical learning via the alternating direction 17 F . Aït Aoudia et al.: Autonomous Discovery of Wireless Communications Algorith ms method of multipliers,” F ound. Trends Mach. Learn. ...

    2011

  30. [38]

    On the sphere-decoding algor ithm I. Expected complexity,

    B. Hassibi and H. Vikalo, “On the sphere-decoding algor ithm I. Expected complexity,” IEEE Trans. Signal Process. , vol. 53, no. 8, pp. 2806–2818, Aug. 2005

    2005

  31. [39]

    A VLSI architecture of a K-best lattice decoding algorithm fo r MIMO channels,

    K. W. Wong, C. Y . Tsui, R. S. K. Cheng, and W. H. Mow, “A VLSI architecture of a K-best lattice decoding algorithm fo r MIMO channels,” in Proc. IEEE Int. Symp. Circuits Syst. (ISCAS) , Scottsdale, AZ, USA, May 2002, pp. III-273–III-276

    2002

  32. [40]

    Generalized approximate message passing f or estimation with random linear mixing,

    S. Rangan, “Generalized approximate message passing f or estimation with random linear mixing,” in Proc. IEEE Int. Symp. Inf. Theory (ISIT), St. Petersburg, Russia, Aug. 2011, pp. 2168–2172

    2011

  33. [41]

    Message-pas sing algorithms for compressed sensing,

    D. L. Donoho, A. Maleki, and A. Montanari, “Message-pas sing algorithms for compressed sensing,” Proc. Nat. Acad. Sci. , vol. 106, no. 45, pp. 18 914–18 919, Nov. 2009

    2009

  34. [42]

    Saad, Iterative Methods for Sparse Linear Systems , 2nd ed

    Y . Saad, Iterative Methods for Sparse Linear Systems , 2nd ed. Philadelphia, PA, USA: SIAM, 2003

    2003

  35. [43]

    Adaptive mixtures of local experts,

    R. A. Jacobs, M. I. Jordan, S. J. Nowlan, and G. E. Hinton, “Adaptive mixtures of local experts,” Neural Comput. , vol. 3, no. 1, pp. 79–87, Mar. 1991

    1991

  36. [44]

    G. H. Golub and C. F. V an Loan, Matrix Computations , 4th ed. Baltimore, MD, USA: Johns Hopkins Univ. Press, 2013

    2013

  37. [45]

    Efficient blind detection f or OFDM systems in selective fading channels,

    M.-X. Chang and J.-H. Chen, “Efficient blind detection f or OFDM systems in selective fading channels,” IEEE Trans. V eh. Technol. , vol. 64, no. 5, pp. 1779–1792, May 2015

    2015

  38. [46]

    Low-complexity blind equalization for OFDM systems with g eneral constellations,

    T. Y . Al-Naffouri, A. A. Dahman, M. S. Sohail, W. Xu, and B . Hassibi, “Low-complexity blind equalization for OFDM systems with g eneral constellations,” IEEE Trans. Signal Process., vol. 60, no. 12, pp. 6395– 6407, Dec. 2012

    2012

  39. [47]

    OFDM with iterative blin d channel estimation,

    S. A. Banani and R. G. V aughan, “OFDM with iterative blin d channel estimation,” IEEE Trans. V eh. Technol., vol. 59, no. 9, pp. 4298–4308, Nov. 2010

    2010

  40. [48]

    Blind channel estimation technique for OFDM systems over time varying channels,

    L. Bariah, A. Al-Dweik, and S. Muhaidat, “Blind channel estimation technique for OFDM systems over time varying channels,” in Proc. IEEE 87th V eh. Technol. Conf. (VTC Spring) , Porto, Portugal, Jul. 2018, pp. 1–7

    2018

  41. [49]

    One-shot b lind channel estimation for OFDM systems over frequency-select ive fading channels,

    A. Saci, A. Al-Dweik, A. Shami, and Y . Iraqi, “One-shot b lind channel estimation for OFDM systems over frequency-select ive fading channels,” IEEE Trans. Commun., vol. 65, no. 12, pp. 5445–5458, Dec. 2017

    2017

  42. [50]

    Automating wireless research and development: What remains human?

    E. Björnson, M. Dohler, J. Hoydis, and R. W. Heath, Jr., “Automating wireless research and development: What remains human?” Preprints, Apr. 2026. [Online]. Available: https://doi.org/10.20944/preprints202604.1172.v1 18

    2026

  43. [2005]

    Available: https://jmlr.org/papers/v6/ opper05a.html

    [Online]. Available: https://jmlr.org/papers/v6/ opper05a.html

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.